It's a common scenario: you want to solve a problem by reading online help or contacting customer support and a window pops up "Hi it's Brian, can I help you?" Of course you know it's not Brian, it answers by spiting out some pieces of badly-prepared script, can't answer any specific question, and sooner or later you realize it's worse than useless as it only took your time you could use to actually solve the problem.
Users never wanted chatbots, really. But platforms wanted them because it meant yet another way to ensure content is locked away inside their platform. So, platforms shout to the high heavens, and everyone rushes to adopt they realise there is no audience.
> An oversized assumption has been that apps are ‘over’, and would be replaced by bots
That was never going to be the case for sufficiently complicated tasks with enough decisions to make. Same reason Google Home/Echo are awkward interfaces for almost anything other than getting music to start playing.
GUIs on phones/desktop are tremendously productive because they're information rich and trivial to use. I can search, choose, customize, and purchase an item on Amazon in like 15-30 seconds flat. A bot isn't going to beat that for the general case.
Yeah, right now the home/echo/siri/etc seem completely worthless to me for all but the simplest tasks, but the idea is for the use cases to keep increasing.
Tools do get better and better. Automation and AI will continue to improve. It's more a question of why chatbots haven't gotten there yet and what needs to be done differently.
Sure, my main thought is that it's an awkward interface for doing things that require choices specifically - it's slow for chatbots / voicebots to present available choices.
> GUIs on phones/desktop are tremendously productive
They used to say that GUIs were toy interfaces and only CLIs could offer real productivity. I don't know where chatbots are going but I wouldn't expect the current state of the art to be anything like where they end up.
The reason the productivity of CLIs was often extolled is because throughout most of interactive computing's lifetime, CLIs and scripting have been intertwined.
It was possible and commonplace to invoke parameter-oriented CLIs from another process, and use the CLI as a perfectly adequate API, while other interface paradigms like TUIs and GUIs never quite achieved such ease of programmatic manipulation.
The key here, I believe, is that CLIs are still vastly more productive for purposes on which you're familiar with the pertinent commands; GUIs are generally better for discovery of new functionality and one-off tasks.
Chatbots and NL interfaces in general are thus far a failure because they neither improve on these purposes nor provide a third, unique benefit. I could see them being useful for consolidation of services (instead of comparing eight puzza places' sites, go to one third-party pizza chatbot), but there's a very, very long way to go.
I seem to remember these were the objections that were made at the time.
The basic tech dates back to SHRDLU and ELIZA. In some ways this is similar to the partially-self-driving-car problem: if it's a mechanical interface where the human is in control it's fine, and if it's a hypothetical human-equivalent AI it's fine, but in the middle people forget to adapt to the limitations of the system. And the limitations are very severe as soon as you go outside the lines.
You see, it takes a huge amount of humility to admit that self-driving cars are possible, but we need several long years of supervised training to say we're actually there. Nobody would like that statement, people - including investors - want to have advanced tech here and now, they don't want vague and distant promises. That's why we deal with alpha and beta versions of everything now, including the partially-self-driving cars.
AsI recall, Terry Winograd later claimed SHRDLU was the beginning of the end of the strong AI trajectory, at least at that time. I believe he, or someone close, said the central weakness of the system was that when it failed, it did not have any understanding that it had failed, much less why. Seems this is the current SOTA in chatbots now, given the somewhat snarky examples in the article.
Perhaps even more worrying then that nothing came out of it: now all that money has flowed into the hands of short-term opportunistic thinkers who might amplify or incentivize even more hype driven companies in the next iteration of this mechanism. Sure, by no means are all of them bad apples (some just jumped on the bandwagon but will long term perhaps become the next Elon Musk), but I think a strong argument could be made that there are enough people of that ilk in the cohort to be worried.
If the chatbot revolution didn’t happen then did those people make a return? Or are you saying it’s the employees getting paid (who also piled onto the dad) who are now the ones with capital?
The chatbot revolution didn't happen but there were plenty of companies sold, with founders who've now gone on to become investors (of the, I claim, possibly hype-jumping kind).
"Why don't you just tell me the name of the movie you've just selected?"
But seriously, it's the same reason so many sites won't accept a credit card number with spaces in it. Someone writing from scratch the same thing that's been written a thousand times, but only thinking about the way they would use it (or the way they were told others would use it).
I do think that eventually these technologies will supersede trivial human transactions like customer service calls. The problem is that no one seems to understand the sheer difficulty of the problem. You have to create a machine that can realistically speak to a person, with all the nuance that entails. We're far further from that than AI marketers would lead us to believe.
The chatbot graph is secret because its purpose is to discourage certain actions, by making you sit through a humiliating hold period with brain-melting repeated announcements, until you reach a person who has strict orders to not be reasonable.
They already have those online, in the form of FAQs and knowledge bases that don't tell you anything new.
I didn't mean trivial in the sense you mean. I mean trivial in that, in comparison to other human interactions, the variables are more limited and therefore lower hanging fruit for automation.
I agree but really only to a certain extent. Coaxing out of humans what thier actual issue is, is more of an art than a science. IMHO and from antecdotal experience this problem seems easier to automate than I think it really is which is why I don’t personally think anyone is crushing it in this space.
Exactly. I did tech support in college, and there's often a huge gulf between what people ask for and what they want. Especially with the rise of the web and apps, we've shifted plenty of power into the hands of users and customers already.
Voice-based systems currently combine all of the limitations of human conversation (e.g., low bandwidth, strongly linear, possibilities hidden) with all the limitations of computers (e.g., not very bright, highly literal, inflexible).
I think the main reason they're so popular for support is that people are bad at accounting. If I install an IVR system, I see that calls to human agents go down, saving me easily measured cash money. But I probably haven't measured the time and cognitive load burden shifted to customers, or the value lost by suboptimal use of whatever they're supporting.
Particularly that last point stands out to me. Lots of automations I’ve seen assume the customer knows the problem. Which they may not they are contacting with a symptom to which they may not know the problem. They aren’t necessarily going to say billing issue/expired credit card when contacting about why they can’t access thier premium features on your app anymore.
But this aligns precisely with what I said. I didn't say it was easy. In fact, I said that the industry vastly underestimates the technical challenge of the problem. My point in mentioning customer service calls was that when we finally are able to generate viable AI (which seems like a long ways out), short and limited interactions will likely be the first inroads. I stand by what I originally said, which was that in comparison to all other human interactions, these are likely the lowest hanging fruit. Not that they are, in absolute terms, low hanging fruit.
Given all the recordings I hear when I call a customer service line, I suspect that most customer service calls are from the kind of people who can't help themselves.
I think customer service departments find them boring and want to automate them away.
I'd take issue that it needs to realistically speak like a person.
In a sense, the Google search bar is a type of chatbot, but we don't converse with it in grammatical English. It doesn't present as a human, so there's no uncanny valley effect. What gets typed into Google is a sort of lingua franca that we've all collectively learned through 20+ years of increasingly capable search engines. What we need is that level of lingua franca, but for a full, state-change-driving conversation, instead of just a one-step search.
It's not the machines that need to learn, it's us.
Except now Google are trying to make their search engine respond to more natural language and so the established search methods like "domain keyword specific-keyword" are often less successful than trying to think of a more human, but not complex, question that might be commonly asked.
Particularly, of course, they no longer care whether results include your required words.
One has to learn and re-learn how to get the best from these systems.
Aside, I'm impressed with Alexa on FireTV, but until it enters data in to the apps, and searches within media libraries, it remains a novelty.
Ah, good point. I remember working with some chatbot tech a few years ago, and came to the conclusion that realistic speaking is not just unnecessary, it's directly undesirable.
The uncanny valley effect, in my opinion, is due to the fact that 99% of our speech and writing is not dictated by content or necessity, but instead driven by social tendencies.
If I ask my friend to lunch, I'll text him "lunch?". But if I ask my boss out to lunch, it's going to be something like a paragraph, explaining what I'd like to cover during the lunch.
Most of what we say is a kind of dance to ensure other people that we aren't stupid. We don't need this formality when dealing with non-human entities, so speaking in grammatical language when chatting with a bot feels incredibly stupid.
Probably, a good AI could get me to the proper representative or solve my problem a lot faster than a tree menu... But I don't think we're really close to doing that.
The issue is that an AI is limited to what the user interface programmed by the company behind it provides. The issue is not the AI algorithms and understanding but rather extremely substandard automation behind it.
Of course one can say that this is not an exclusively AI issue. Call centers, especially offshore ones, have the same problem.
We can translate spoken words to text strings pretty reliably. AFAIK, we can do this locally without much processing power, but I may be mistaken. From the point that spoken words become strings, chatbots and voicebots are effectively the same.
Those strings are piped through a Switch statement of static responses, or fall out into the Default response. This was the situation in 2013 when I created a speech bot in Powershell, and it sounds like the state of the art here hasn't progressed far.
What's needed is logic to dynamically build the Switch statements, or otherwise better parse human entry and build responses. There has been much work on a few different fronts, but I'm not aware of any which were wildly successful.
>We can translate spoken words to text strings pretty reliably. AFAIK, we can do this locally without much processing power, but I may be mistaken.
Cloud services backed by big data sets tend to be better although I admit I haven't tried a local copy of Dragon Naturally Speaking for a long time.
In any case, at least assuming fairly mainstream American/English accents, the voice recognition isn't really the problem any longer. Sophisticated NLP and responses are. We're a long way from virtual assistants that can do anything sophisticated.
> ...no better features than a menu system, but you had to guess what the menus were.
This is also true for CLI interfaces. It illustrates the popularity of GUIs - much to the dismay of CLI enthusiasts everywhere.
Edit: Yes, a -help or "help" command can be used to list the menu. But then this command has to be known beforehand. What if a clever designer decides to use -assist or "assistance" instead?
Not to be too curmudgeonly but you don't have to guess, almost all those CLI interfaces have a manual and a --help function. You can't ask a chatbot what its options are.
"I need to find out how much space I have left on this machine."
> freespace
bash: freespace: command not found
> diskspace
bash: diskspace: command not found
> disk info
bash: disk: command not found
> diskinfo
bash: diskinfo: command not found
> man disk space
No manual entry for disk
No manual entry for space
> help disk space
bash: help: no help topics match `space'. Try `help help' or `man -k space' or `info space'.
(looks it up on the internet)
> df /dev/sda1
Filesystem 1K-blocks Used Available Use% Mounted on
/dev/sda1 19478160 1370768 17094912 8% /
... AND that's why GUIs are preferred by most non-technical computer users.
EDIT: Maybe I should have said 'CLI Newbie' - I am in no-way singling out Linux here.
"Aha, there's my disk and little bar showing the free space"
It's two clicks if you have your wits about you, maybe a few more if its not that obvious.
I don't know why you singled out Windows. It's just as easy in Gnome or KDE etc - the point here is that CLIs can never be as intuitive as GUIs. As the original article said - Humans are very visual people. Seeing things in a visual space is much easier for us.
Further, Windows will pop up a notification that your disk space is running low, and this opens another window with suggestions for different types of files (temporary update files, temporary internet files, recycle bin, installed programs) that can be removed to release space.
Press windows key, type "space". First item is "Storage". Press enter or click it, you see a prominent progress bar like thing and text: "300GB used, 164GB free".
Thank you, you just re-enforced my point; I've been using Linux seriously for about a year now, and I didn't know about 'apropos'. Bash never mentioned it whenever I got stuck, and my learning travels on the internet has never referenced it.
Honestly google is a better tool unless you already kind of know what you're looking for.
Google "linux how much space on drive" and the first hit tells you exactly what you need.
Another question someone might ask is "how do I change my password". Google for "linux change password" and there is a special help box right at the top of the search results.
apropos password and you get this enormous list:
apg (1) - generates several random passwords
chage (1) - change user password expiry information
chgpasswd (8) - update group passwords in batch mode
chpasswd (8) - update passwords in batch mode
cpgr (8) - copy with locking the given file to the password or gr...
cppw (8) - copy with locking the given file to the password or gr...
cracklib-check (8) - Check passwords using libcrack2
create-cracklib-dict (8) - Check passwords using libcrack2
crypt (3) - password and data encryption
crypt_r (3) - password and data encryption
des_read_2passwords (3ssl) - Compatibility user interface functions
des_read_password (3ssl) - Compatibility user interface functions
doveadm-pw (1) - Dovecot's password hash generator
endpwent (3) - get password file entry
endspent (3) - get shadow password file entry
EVP_BytesToKey (3ssl) - password based encryption routine
expiry (1) - check and enforce password expiration policy
fgetpwent (3) - get password file entry
fgetspent (3) - get shadow password file entry
fgetspent_r (3) - get shadow password file entry
getpass (3) - get a password
getpw (3) - reconstruct password line entry
getpwent (3) - get password file entry
getpwnam (3) - get password file entry
getpwnam_r (3) - get password file entry
getpwuid (3) - get password file entry
getpwuid_r (3) - get password file entry
getspent (3) - get shadow password file entry
getspent_r (3) - get shadow password file entry
getspnam (3) - get shadow password file entry
getspnam_r (3) - get shadow password file entry
git-credential-cache (1) - Helper to temporarily store passwords in memory
gitcredentials (7) - providing usernames and passwords to Git
grpconv (8) - convert to and from shadow passwords and groups
grpunconv (8) - convert to and from shadow passwords and groups
grub-mkpasswd-pbkdf2 (1) - generate hashed password for GRUB
lckpwdf (3) - get shadow password file entry
login.defs (5) - shadow password suite configuration
lppasswd (1) - add, change, or delete digest passwords.
Net::LDAP::Control::PasswordPolicy (3pm) - LDAPv3 Password Policy control object
Net::LDAP::Extension::SetPassword (3pm) - LDAPv3 Modify Password extension ob...
pam_pwhistory (8) - PAM module to remember last passwords
pam_unix (8) - Module for traditional password authentication
passwd (1) - change user password
passwd (1ssl) - compute password hashes
passwd (5) - the password file
passwd2des (3) - RFS password encryption
putpwent (3) - write a password file entry
putspent (3) - get shadow password file entry
pwck (8) - verify integrity of password files
pwconv (8) - convert to and from shadow passwords and groups
pwunconv (8) - convert to and from shadow passwords and groups
seahorse (1) - Passwords and Keys
setpwent (3) - get password file entry
setspent (3) - get shadow password file entry
sgetspent (3) - get shadow password file entry
sgetspent_r (3) - get shadow password file entry
shadow (5) - shadowed password file
shadowconfig (8) - toggle shadow passwords on and off
smbpasswd (5) - The Samba encrypted password file
smbpasswd (8) - change a user's SMB password
ulckpwd...
$ apropos -a change password
chage (1) - change user password expiry information
kpasswd (1) - change a user's Kerberos password
passwd (1) - change user password
PKCS12_newpass (3ssl) - change the password of a PKCS12 structure
$ apropos -a space disk
df (1) - report file system disk space usage
df (1p) - report free disk space
The fact that the search defaults to "or" instead of "and" is a design flaw IMHO. By far the most common case when you have multiple arguments is that you want the item that contains all of the keywords.
Nothing says user friendly like making the user read the manual for his manual so he can memorize the nine different categories (plus perhaps custom categores like TCLs) to get the manual page he wants.
I use man and apropos all of the time, but I have no illusion that they're new user friendly. They're amazing when you just need to look up some options or get the syntax for a system call, but if you've not assimilated the Unix way of thinking they're quite obtuse.
I think people would complain that apropos doesn't do its job if it were named help.
Apropos would be a powerful part of a "help" command, but in 50 years nobody has ever written a help command that gained enough traction to be included in the standard toolchain.
The name “help” is, unfortunately, already taken by a shell built-in command. But one could argue that this help command should be very much more helpful than the restricted functionality it currently implements.
I feel like the real problem here is that bash is still the default and bash defaults/UX are straight up awful. The fish shell gets a lot of things right in regards to a sane out of the box experience, though there are a few things it could do better.
The tools themselves leave a lot to be desired from a UX experience too. Things like nonsense names, bad defaults and/or no inherent intelligence in the program, means you have to specify numerous options manually for things that should be easy to automate.
The whole experience would benefit greatly by telling people about things like man/apropros on first startup and other one time tutorials.
There are some days where I feel tempted to write small helper scripts whose sole purpose would be to rename/reorient the default experience/flags of various CLI apps, so that it is not a jumbled mess full of historical accidents that can no longer be changed.
At least with a cli you can usually invoke help and visually scan the text for what you want. With audio chatbots, listening to a list of options quickly becomes tedious.
With CLI, you usually can type help or add a -h to the command or something like that and you'll be presented with options and examples.
You really don't have this option with voice. On the other hand, when I visited my mother I was surprised that she figured out that Google Assistant can provide her with nutrition information. She would have never figured out by herself on a CLI or menu because it's very intimidating.
This somewhat contradicts my previous comment about how much I dislike voice interfaces but they probably have some strengths.
In my experience, these "commands" are also quite vague. Instead of expecting the "help", they should just put (x) and show the menu where I can click on things and be done with it.
On voice interface, your best chanse is to connect to an actual intelligent being as soon as possible because getting the options is a frustration by it's own.
My experience is that they tend to repeat their intro dialogue, which is often something like "tell me a few words about why you're calling, for instance you could say 'upgrade account' or 'billing issue'". Unlike keypress phone trees, they virtually never enumerate all the options.
If I'm saying 'help' it means I've tried that and failed, so it's not clear what the benefit is.
> With CLI, you usually can type help or add a -h to the command or something like that and you'll be presented with options and examples
And you get to enjoy fumbling around trying to guess whether this particular program's magic incantation to summon help is "help", "-h", "--help", "?", "/?"...
Note that mobile GUIs have often re-invented the "undiscoverable" interface, where things like swipe or long-press do things but it's not obvious where you can safely do them.
CLIs, though, serve a different market. I put them in the 'professional' box, alongside Autocad and such; yes, these tools are built with the assumption that the people using them have experience using them.
IVR, is supposed to be 'intuitive' or 'natural' or however one's preferred marketing dialect describes it. The assumption is usually flipped - that tools are supposed to assist those ignorant of their use[1].
It has been a long time since normal users have been expected to use a CLI. If you're looking at one, you choose to. This is not at all true of IVR/Sirlextana.
[1] Of course, they don't serve the caller, they serve the robot's owner, so interests are only aligned to the extent the robot efficiently helps the caller with their goal, which adds an additional layer of opportunity for frustration.
The point of a CLI is to be able to write programs to handle complicated interactions between applications with ease, and by default. A gui doesn't have a pipe.
As a tech oriented site we're nearly all accustomed to unix/unix-inspired general purpose CLIs. However that's not all that exists. There exist more specialized systems, such as those designed to be used over teletype machines, that present the user with a pre-defined menu prompt at each step in the interaction. However unlike with voice-oriented systems, the user can sit there and read the entire list of options for as long as they'd like. You connect, and a menu prints out. You press a key to make a selection and a new menu prints out, or the system prompts you to enter a different sort of data.
Think back to the earliest programming courses people take:
What's your first name: Jack
What's your last name: Ch
Hello Jack Ch!
Unintelligible grunting works too. After a few "I'm sorry, I didn't get that" it always transfers me to a human.
The most annoying part of the experience to me is the menu items are always ~25 things I can trivially do on the website. No, I'm not calling to check my balance, pay a balance, update payment information, etc. I'm calling because your website specifically said that function isn't available online and I need to speak to a representative.
Also annoying is never knowing how deep the menu tree is. I have to write down (or hold my fingers out) what the best option/number at the time, because its likely to change as the resolution of options refines.
Including circular paths in phone menus ought to be some kind of professional misconduct, at least when they're not invoked by "go back" or "start again". There's nothing quite like getting seven levels in and discovering you swapped trees somewhere and are back where you started.
my guess is that this is not the case, but it's still worth it to handle any situation they can with a robot, or as a consolation, an impatient customer.
That often depends on your definition of easy – most large companies have poorly designed or buggy websites, lousy account management, etc. The problem is that the phone acts as a safety valve for those problems but when management sees the cost they react by trying to have fewer humans answering the phone rather than getting serious about UX.
My father in law, who at 71 is fairly tech savvy, still defaults to calling when he needs help with something. He's always saying "I was on the phone with Apple for 3 hours yesterday" or "I was on the phone with Xfinity all day yesterday."
I think if you're from a certain era, you just pick up the phone first and want to talk to somebody, even if it might be pretty easy to find what you're looking for online.
Must depend on eras, yes, because personally I avoid using the phone at all costs. I'd rather take my eyes out with a spindle gouge than talk to someone over the damn phone
I usually go with loud noises or swearing. That gets me through to a real person. To make up for it, I always say "please" and "thank you" to Siri when im at home.
And I think most of us are in the same camp there. Usually, when I am calling a business, it's because of some kind of unusual or complicated inquiry or special situation because I generally don't want to call if I can do or get what I need by logging in online to the company's site to manage my account or do whatever I need there as that's often the fastest. But the online account management only does so much so when a weird edge case regarding doing business with the company arises, I will call, and in those situations the AI/voice prompts never handle the situation I am calling about such that I need a human agent/rep every single time. I wouldn't be calling for something simple!
As someone for whom English is not the native language it's the worst. It's incredibly humiliating when the stupid machine can't understand what I am saying - and 100% of the ones I encountered fail to understand my surname and I have to repeat it 5 times like an idiot before it finally gives up and just tries something else as an authentication method.
The dirty secret is that there are thousands of humans listening in on these phone bot conversations. They have the power to press a button to override the endless loop.
Amazing what they are spending to keep up illusions.
On the other hand, I can think of at least two prominent elected officials (one at my state's level and one at the national level) who no longer speak in sentences that I can parse with my own mental model of English. The one from my state released a statement yesterday that seemed identical to the kind of thing one would get from a late '90s IRC chatbot that was trained on a big dataset of energy company press releases.
I presume that the national level figure was Trump. Whose "nuclear sentence" has been much discussed:
“Look, having nuclear — my uncle was a great professor and scientist and engineer, Dr. John Trump at MIT; good genes, very good genes, OK, very smart, the Wharton School of Finance, very good, very smart — you know, if you’re a conservative Republican, if I were a liberal, if, like, OK, if I ran as a liberal Democrat, they would say I’m one of the smartest people anywhere in the world — it’s true! — but when you’re a conservative Republican they try — oh, do they do a number — that’s why I always start off: Went to Wharton, was a good student, went there, went there, did this, built a fortune — you know I have to give my like credentials all the time, because we’re a little disadvantaged — but you look at the nuclear deal, the thing that really bothers me — it would have been so easy, and it’s not as important as these lives are — nuclear is powerful; my uncle explained that to me many, many years ago, the power and that was 35 years ago; he would explain the power of what’s going to happen and he was right, who would have thought? — but when you look at what’s going on with the four prisoners — now it used to be three, now it’s four — but when it was three and even now, I would have said it’s all in the messenger; fellas, and it is fellas because, you know, they don’t, they haven’t figured that the women are smarter right now than the men, so, you know, it’s gonna take them about another 150 years — but the Persians are great negotiators, the Iranians are great negotiators, so, and they, they just killed, they just killed us.”
It's fascinating how big a quotient the persuasiveness of his speech lies on the vocal level, and is almost nonexistent when written out, abstracting away that detail. It almost makes me think that the current level of chatbots are enough, if only they could nail the text-to-speech component; to really surf whatever it is that makes speech sound powerful and influencial.
I'm pretty right leaning and I honestly think the funniest thing is listening to him talk, I never have any idea what he's talking about no matter how long I listen to his speeches. Ideologically he and I agree on some things, but I just never know because I can't understand a single thing he says unless he tweets it. I think he actually suffered when twitter upped the char allowance bec the brevity forced him to be succinct.
The thing is, that's a very hard problem. He's probably got either the luck to be in a favorable space or some sort of native intuition that makes his speeches resonate with a certain group of people.
It's basically something very fuzzy, historically the hardest thing to emulate with computers :)
Fuzzy would imply it's hard or impossible to model it, mathematically. But ask any good biologist and I bet you'll hear quite a bit of confidence that the keys behind powerful speech can be understood. For an engineer without domain expertize, yes it's fuzzy. Armed with the right research papers? Highly likely something persuasive could be implemented.
I think some context would be helpful to parse the specific things he's referencing, but, here's my partial translation/paraphrase:
"I'm smart. My uncle was smart too. I went to Wharton, he went to MIT. He was a nuclear scientist. You know, I feel like people don't give he and I the credit we deserve. We have good credentials but people don't pay attention to that. They think I'm stupid because I'm Republican. But if I ran as a Democrat, they'd think I'm smart.
Anyway, my smart uncle told me that nuclear energy is really powerful and efficient. And that was 35 years ago, so I imagine it is even more powerful and efficient now.
Now I'm going to tell you about three or four prisoners. Some are male and some are female, I think. The female ones are smarter.
But anyway, the Persians/Iranians are great negotiators. The US should have negotiated harder with the Iranians. But I feel the Iranians ended up getting a good deal regarding our nuclear negotiations. And I feel that the United States' economic and security interests have been harmed by the deal we made with the Iranians, who are smart."
It's easier than that. "My uncle and I are both smart and understand the power of nuclear. The media deny how smart I am because I'm a conservative. I wasn't there when we negotiated the Iran deal- someone weaker was, who by the way didn't recognize the intelligence of females, and that resulted in additional American prisoners and the US getting a bad deal.
What was he trying to accomplish with that speech? Educate a classroom of college students? That's what lecturers are for. What was he trying to accomplish? He was trying to win a political race. He succeeded. It's annoying when people act like Trump is an idiot. (Or GWB before him. It was EXACTLY the same pattern.) Hate him a little or hate him a lot, he wouldn't have become president if he were an idiot. One could strongly argue part of his success is due to people underestimating him so much.
I was trying to keep this as a discussion about the technology and not about the politicians involved, but it apparently didn't stop most of the thread from ending up dead anyway, so the statement I had in mind (you can find more on the source if you're interested) was:
> "Our electric grid system is not stable. If it can be down or you have rolling brown outs or black outs and we know that is possible. If you don't have base load that means something, some type of energy that will run 24/7, rain, shine, no matter what happens is uninterruptible. The only two things you have that does that on a uninterruptible basis is coal and nuclear"
This isn't the first time I've wondered at the applications of natural-language processing software in political [speech]writing. I'm not sure where best forum is for people to discuss these ideas, but I apologize if this was too political, regardless, for this site.
Chatbots basically killed a startup I was working with a couple years ago. We had really nice apps, with a realistic exit strategy. Then came "The Pivot", and we were all working on chatbots. Didn't take long for me to leave that company, who went out of business not long after.
The PTBs at my last place of employment wasted an enormous amount of cycles on chatbots, despite pleading that the available platforms to use all sucked.
Sounds like the company was dead anyway and you are mistaking causation. Companies that are "alive" don't pivot. Companies that are dying people to avoid death. That the company had to pivot suggests the owners decided the exit strategy wasn't actually realistic.
A pivot is like a "Hail Mary" pass... risky and unlikely to work -- something that only makes sense when time is running out and you have nothing left to lose.
Or... if your company pivoted when the end wasn't near, the leadership was greatly incompetent, which means you pretty much never had a chance in the first place.
Interesting to note: the poster read the signs and made the right move at the right time in any case.
I always found the chatbot idea odd, it felt like a step backwards in terms of interaction. We started out with very basic input methods to computers, like punch cards. Then we moved onto a command line interface where you could type in words. Then we eventually got GUIs, graphics, websites, all sorts of complex and nuanced ways to interact with a computer.
To go back to interfacing with a program using written language seemed like an odd step. It's never been the most efficient way of doing something, and it requires very advanced technology to accurately understand what people are trying to say, in whatever slang, shorthand, or bad spelling/grammar they use.
Besides, it's not really dead, the tech just moved to "voice assistants" rather than "chatbots" - really just spoken word rather than written word. And I'm not convinced that's the "revolution" most people are expecting either. I'll stick to clicking buttons and typing things into my terminal.
The best place to look for cool and innovative (and free!) IF games is the Interactive Fiction Competition [0].
Anything by Andrew Plotkin [1] (aka Zarf) is guaranteed to be interesting. Other indie authors of renown to look for are Emily Short and Adam Cadre. But also look for new authors! Mind you, modern IF focuses less in puzzles and more in narrative or exploring the boundaries of the medium.
One of my personal favorites is "Spider and Web" by Zarf, because I love its Cold War-esque setting. Mind you, it can be difficult! The best IF games also explore the console interface itself, such as in "Fail Safe".
An example of a particularly innovative game is "Rematch" [2]: it's a single move game (i.e. you win or fail in a single input, which can be quite complicated and shows off what modern parsers can do). It's sort of a "Groundhog Day" where you must prevent a disaster in a single move, and if you lose you replay it again, and again, and again, till you get it right.
Many of these games can be played in a browser, without installing anything.
The Infocom games were actually after a lot of earlier games like Colossal Cave (aka "Adventure") and the Scott Adams (not the Dilbert guy) adventures. As such, their parser was actually pretty advanced compared to the 1970s games and understood whole sentences rather than the traditional two words.
The human-machine communication issues are the same in both cases. When a video game doesn't understand what you're trying to do, nothing would happen or you get a negative, non-specific feedback. You think it is better because you get distracted by the colors, lights and sounds but the communication channel is as narrow, if not narrower actually.
By the way, what OP says may be true of a game like Dunnet. I don't know if it is no more maintained or if it is kept that way because nostalgia, but more recent text-based games (that is, one that could have been programmed by your father instead of your grandfather) do way better than this. Just try some popular MUD (the MMO version of text adventures), I'm sure you'll be surprised.
> It's never been the most efficient way of doing something, and it requires very advanced technology to accurately understand what people are trying to say
It seems like a lot of times when people talk about chatbots, they really mean these phone trees in text form, in which case I would agree with your sentiment.
However, be cautious in conflating chatbots with CLIs. I would say a CLI is not (always) an intuitive interface, but for a lot of problems, they are quite efficient.
The CLI style chatbots tend to be much better since they are basically CLIs in an easily accessible location (e.g. in an app on your phone).
That's a great way of putting it. Also, GUIs attempt to use human natural reasoning (about objects, space, movement, etc.) and have some kinesthetic components (the mouse, pointing, dragging).
It's interesting to think about. In some ways GUIs are more primitive than text interfaces. The scathing
characterization of GUI as "caveman's point and grunt interface" isn't entirely wrong. But at the same time, chat bots are spoken word. Text is a later invention, which was created for a reason. It has much higher information density, allows you to look at multiple tings at the same time and skip irrelevant detail, etc.
To me it was a way for engineers to make interfaces without having to worry about any are design. An engineer-turned-marketing-person would call it "CLI 2.0". The problem was the non-designers still don't know UX at all and natural language is hard so you end up with a sub par, poorly design interface that doesn't understand you.
I keep seeing references to design/UX, and IMO they are way off the mark. The problem is far more fundamental than that. If the only problem were design/UX, then proper chatbots would've taken off by now.
Understanding natural language is much more than just a design problem. It's a grand challenge and core subfield of Computer Science. It's the original Turing Test.
My pet theory is that there are a bunch of investors somewhere salivating at the thought of dominating a market for selling products to baby-boomers who are becoming impaired with age.
The biggest manifestation of this is the self-driving car hype for when grandpa can't legally/safely drive himself, however voice-assistants also fit that mold: Something to sell to grandpa when they don't want to learn/buy a new thing and his eyesight is bad and arthritis makes typing hurt.
The two biggest flaws with building chatbots are a) the lack of good tools to express chat flows at a higher level (no, modeling it as a graph isn't good enough) and b) no Natural Language Generation that's good enough to trust in production.
The first one leads to subpar bots that can't respond to most things, the second means humans have to fill the 'long-tail' of potential responses by hand, which is impossible.
I think the biggest flaw is chat is just a black hole of a UI. Until the machine is a full AI, you have to be able to put boundaries and parameters on expected inputs. Otherwise, it's like putting a casual user in front of a Linux terminal and asking them to use a website via curl. Unless you give them very visible paths to interaction, you're done. The fact is, we have had that kind of thing for ages and it's called IVR.
One of the best insights we had at Ozlo was putting prompt bubbles on our UI. They served two main purposes. First they showed what type of questions you could ask; and second, they showed how to ask a question.
They provided guard rails so to speak. If you don’t really strong hints about what you can do — and at least recognize intents you can’t handle — you’re screwed. There is always some asshole who sees “Hi, I can tell you about restaurants and entertainment. What are you looking to do?”, and answers “What’s the atomic weight of boron in nanofirkins?”
I think all people with some technical insight could've told you that 3 years ago. But still I'm happy the world finally realizes it. Now it only(!) takes another 5-10 years until my employer figures it out. Some of these bigger companies just start to bet on chat bots, but of course will fail at the same hurdles as everybody else.
> On the input side, it’s easier and faster to click than it is to type.
I disagree with this. It may be easier, but it is in no way faster. For those who don't use computers very often I can see it being slower. But for people who use computers on a daily basis, for mice to still be a commonly used input device is disappointing. With constant context switches, moving an object across the desk with you wrist, while keeping your fingers in the same place to click, and carefully putting the cursor just in the right place isn't nearly as fast as the same interface using typing. Or comfortable for that matter.
I'm not saying I disagree with the article as a whole.
I think that a combination of GUI with a keyboard can be great too. I think CLIs have stuck around for so long for their consistency. You type something and press return. The web on the other end of a spectrum is full of surprises and therefore requires something to be able to handle it's unpredictability.
Who knows what is really going to happen if I press Tab here? Will it really bring me to the next field or some random link in the other side of the page?
I wouldn't doubt that once the web finally settled down from this hype, people will find ways to get the full potential out of both GUI and the keyboard in a consistent way.
The web is the only reason I need a mouse during work anymore.
It’s just like why all voice interfaces are shitty: no one has any idea what the thing can or cannot do. They have a hidden user experience but the interface makes it feel like you’re talking to a human, but it’s so far from being a human.
These interfaces are almost like a dark pattern because of how bad they are.
For me, these AI solutions always feel like I'm talking to a very stupid person who has no idea what's doing but has a booklet in possession that may or may not have the information that I'm looking for. Just give me that booklet and I will figure it out,geeez... What a frustration to deal with these smart machines.
With Siri or Google Assistant, you soon figure out few things that this very low intelligence person can do(like telling the weather or setting an alarm) and stick with it.
This is also why I'm excited about iOS12 with all these Siri shortcuts, instead of pretending that we are talking to a smart being let's have a concrete list of things that can do.
On the other hand I do believe that these voice interfaces have some potential, just the technology is not there yet.
The audio interfaces are so bad I resort to one-word answers to every question, to get me to a human as fast as possible.
"Hi, in a few words, what can I help you with today?"
> "billing"
"It sounds like you have a question about your bill. I can help you with that! If you can give a few words to describe the reason you are calling, I can help you with your bill."
> "billing"
"OK, let me get you to a representative who can help!"
... instead of spending 10 minutes wrangling with the vapid AI, I can actually move on with my day after speaking to a human. Was this the future we envisioned in the 90s? I think not. Some systems let you spam 0 (zero) and it transfers to a human, but more and more are requiring you to interface with the system in some way, even if disabled or impaired.
'Agent' or 'representiative' will usually do it faster.
But I hear you. I recently wanted to change the ownership of my cellphone account. "Change ownership", "It sounds like you want to change cellphone plans, is that correct?", "Agent", "Okay, let me get you to a representative".
The only tasks these things are equipped to do, are the tasks that I can do via the company's online portal, and in a much less frustrating manner. I wonder who is actually using these things.
It's likely that these are for company itself. Ain't no one got time to deal with questions that are already explained on the website. This is made to filter mindless or lazy drones who don't want to search and its totally fair.
I've noticed that companies without IVR lines have switched from hold music to recordings that repeatedly list all the things you can handle online instead.
It's annoying when my issue isn't on the list, since I'll have to hear it 50 times. And it's incredibly aggravating when my issue is on the list, but the website doesn't actually work right. But it's not hard to understand why it happens.
Oh my Goddess, there is nothing worse than hearing: "It's on the website".
Yeah, that's unhelpful if "the website" is a sprawling, vast wasteland. Finding a specific thing on an obese, convoluted site is like trying to find the bucket of ice cream in Siberia. I always check the site first, so if I am calling... it's because your website is pretty bad.
So if I do find the thing I need and it is broken, then I call and dodge the idiot AI, and the beleaguered agent or offers the pathetic advice of: "It's on the website". I tell them it doesn't work, and this has led to two outcomes, either the agent tries to do it and it is so broken and fucked up that they tell me that's it's down and I have to call back or try the website days or weeks later, or I am lucky and the agent has to do a tedious task that I would have preferred to do myself online.
I was on hold with my insurance to file a claim, the IVR loop said for the third time "Did you know? You can file your claim online!" so I gave in, hung up, went online and all the form asks for is my contact information (phone + email) and my preferred method, so I enter email.
I get an email "Hey, I need to contact you so we can proceed, what method would you prefer?" I reply "Email, please" provide the information I assume the rep needs and which the online form should have been able to ask me and offer "What else do you need".
"Is there a time I can call you and [go over the script and fill out a form on my computer, as if you'd just stayed on hold hours ago]?"
This is always annoying because the only reason I would ever call is if I've exhaustively tried every way to use the website and encountered some sort of roadblock/edge case where they tell you to call in.
This is stacked against power user just like anything else in this world. Technology is dumbed down for the average user whereas I believe we should strive to make the average user smarter instead. Not happening tho it seems.
Bashing # several times quickly also works on a lot of systems. I always give it a go first time around - when it works, the voice bot typically says 'I didn't understand that, I will connect you with a.. [human]'.
Failing that I also resort to answering in one word answers. Bots can't handle any sort of ambiguity. Even when they manage to solve the language-parsing problem, people will still just bark one word answers at anything that looks or sounds remotely non-human[1] because that's how the early AI's trained us. The future-concept of people chatting away to robots is incredibly unlikely because to do that you need a level of empathy for the person/object you are talking to. We've clearly shown so far that we just don't have it for bots.
--
[1] Like the 19th century British Explorer shouting English at the natives.
>> The only tasks these things are equipped to do, are the tasks that I can do via the company's online portal, and in a much less frustrating manner. I wonder who is actually using these things.
Old people. I'm not kidding. I used to do PCI compliance work with call centers. I've listened in on many a call to verify procedures. Call centers deal with an unending stream of older people who cannot or will not use "computers". They want to talk to a person on a phone. The people who are comfortable with the online interface only call when the online interface fails them. Old people go to the telephone first.
And there is a growing subset of old people who are functionally illiterate. I don't mean they don't know how to read, I mean their eyesight is diminished to the point that using the computer/phone/tablet is uncomfortable. Or their fingers don't move well enough for a keyboard/touchscreen. The phone is simple and reliable, not requiring either visual or physical dexterity.
I agree with everything you've said, and I feel really badly for how tech's dark patterns, crappy UX, and removal of humanity have treated the elderly.
Even non tech items like opening a bag of chips or box of crackers. I estimate that some packaging must require 40+ lbs to open. And I think of the "how it's made" show and how advanced manufacturing plants are. So they use all this tech to make a box of crackers as cheaply as possible, they have cameras fast enough to find and remove a single dark grain of rice in a torrential stream, but put zero thought into the UX experience and how the user can open the box.
It sounds like the simplest "tech fix" for this phone menu is once their account is found,
if (age > 60 )
sendToPerson ();
Furthermore, you could do "soft" account verification with their number. So if they call from a number that matches an elderly customer, just route it straight to a human, even though numbers can be spoofed it's not going to harm the customer to just assume it's them for that step only and fully verify their acct later.
You think old people don't just hit zero, pound, or say 'agent', to go to a person, to do the things the phone tree can do?
If they don't, is that because they don't know any better, or because they actually want to use the IVR?
My guess is either they go straight to a person, or they don't know that's an option. So if I'm correct (and I think the onus would be to prove that old people think differently than everyone else in this regard), these things exist to take advantage of the ignorance of old people to save a company some money. Greaaaat.
I worry the human operators will hold it against me. The quality of assistance they provide may be biased by how irate and short tempered they think I am, even if I was just putting on a show for the robot.
Hah, I really do, if ever so slightly, feel this way. Surprised by the down votes. Just disapprove because you frown on humor, which it wasn't intended as, or somehow did I offend you?
Reminds me of a tech talk I went to last year, where the speaker/developer was trying to get on the Alexa hype train by building a "virtual doctor" application of some sort. He was very excited about the possibility of reducing physicians' administrative overhead, but my thought the whole time was, "seems like one of those terrible phone navigation systems..."
A colleague of mine worked on the automated call center systems in the early 80s. The fastest way to get to a human is to cuss it out. This has worked for almost every system I’ve worked with so far (not sure if they’re using the same underlying software or if this behavior has been ported over). I only opt for this trick when I have no idea where to go given the menu read out to me because I like to think that support menu systems are designed to make it easier to navigate an organization’s own bureaucracy primarily, not necessarily to help me with my issues.
This happens way too often. I wish I knew why systems ended up like this. Seems to be a business opportunity to fix them once and for all. This is 2018, we can do better
Vendors do this because they know they can get away with it, i.e. no perceived counter party risk. However what they don't understand, is that there is a HUGE counter party risk.
If its a charge on my credit card or checking, that's what I do, I give them this counter party risk. I call and do a charge back. Basically after 20 minutes with a vendor on hold, I declare that they are being unreasonable and they are unable or unwilling to refund me, I call the bank to do this charge back.
Now, what you tell the bank is critical. For whatever reason, vendors have been able to un-do my chargeback. So instead, I tell the bank that I dispute the charge -- but that I am also disputing the method of payment. So if they argue that the charge is valid in the investigation, I say that is fine -- however, the bank is not authorize on my behalf to pay them, and if they think the charge is authorized they must contact me to arrange a different method of payment. And under no circumstances can they use the bank to make payment. So yeah if they argue the charge is legit, that's fine, contact me (I wish I could put them on hold haha), and I'll pay them -- but the bank can't. The bank likes this because they are off the hook and I like it because they don't get to screw me. Worked every time.
If everyone did this, and made this a huge country party risk, these vendors would stop putting people on hold and ignoring or purposely inadequately addressing their greivences with their product/services.
how can you dispute payment method especially when they tell that you used your card with OTP/signature giving your consent to use this payment method? Just trying to understand so can use this technique if situation arises :)
A solution for me is to avoid dealing with shitty companies like this and seek online-only alternatives (with a good web UI), and failing that, if a shit company screws up and I can’t fix it quickly through their awful customer support I stop paying and let them call me.
It’s funny how they never keep their promises of calling you back until you actually stop giving them money. They also magically start paying more attention to what you say, and complaint letters that went unanswered for months and presumed “lost” suddenly become found and answered.
The reason why is that the phone tree/IVR and back-office systems were developed by someone like IBM, Accenture, Oracle, etc instead of actually competent developers. The budget being blown on the aforementioned companies, they have no choice but to stick with it.
USAA really impressed me recently with a really slick feature: I was in their mobile application, and realized that what I wanted to do was a little too complex for what the app could handle. So I called support, and they (1) automatically identified me based on my number and (2) knew I was just using the app, and asked if I'd like to transfer directly to a person!
It is probably not the most complex integration technically, but it made the experience so smooth.
USAA can be pretty great. In some ways they are ahead in tech, and have been for a long time. I do wish they'd add support for virtual credit cards tho.
The app also has a useful feature to launch the website with session tokens, so it keeps you logged in if you were logged into the app.
So if you know where the thing you're looking for is on the website, you can just launch the website and use that.
Super simple feature but it seems like for so many companies, if you go online, it redirects you to their mobile site which has been "phased out" and tells you to the get their app, which you have and would have used if it had the feature you needed.
For such a "post computer" world it seems ridiculous how often I throw my hands up and find my laptop when trying to complete an online chore from my mobile.
On the flipside, think of the joy of realizing you've finally been connected to a smart, competent person who is empowered to solve your problem. It is like finding a call center unicorn.
That's not a terribly uncommon position, not everyone wants to live forever. Arguably it's a healthy attitude to take, since death will come for you eventually whether or not you make peace with that inevitability.
I had this experience once and it was the result of finally getting a local call centre with better trained and empowered employees (most calls were being farmed out to overseas call centres which followed a very simple script).
It really depends on what sort of organization you're dealing with. My credit union has always been a pleasure to deal with because their phone people are all hired and trained locally. On the other hand they only take calls during business hours which can be a little awkward when I'm in another timezone.
By and large, voice user interfaces still fail to abide by Grice's Maxims[1]. In your example, that's the maxim of manner:
> when one tries to be as clear, as brief, and as orderly as one can in what one says, and where one avoids obscurity and ambiguity
In a sense, this is not unlike an unskilled and nervous attendant who, faced with a request that they don't understand, starts to chatter more and more in hopes of eliciting the information they really need.
> "more and more are requiring you to interface with the system in some way, even if disabled or impaired."
In a sort of sick and very selfish way, I find myself somehow glad that blind and deaf people exist and are protected by accessibility legislation. I've often found accessibility features, particularly those in software, to be exceptionally useful even though my vision and hearing are fine.
For many of these systems, you actually are talking to a human. The person listens to what you say then directs your call as (hopefully) appropriate. They don't have a mic, so clarifying questions are limited to buttons they can push on a glorified soundboard -- in your case it sounds like there were several departments that handle different aspects of billing and they were trying to sort out where to route your call.
The guy you were talking to eventually just routed you to someone with the same job but who has a mic, because you weren't giving him anything to go on and all he'd have been able to do is push the button again to ask you the exact same question (and having a person do it who can at least vary the inflection of the question is less infuriating than seemingly being stuck in a chatbot's infinite loop).
--
Source: have a friend who used to have this job. He has thankfully moved on to something less soul sucking than being a human literally pretending to be a robot.
That seems utterly nonsensical and dehumanizing. If you have a human greeter/operator, why not expose that to the customer? You are already paying for the human to be in the loop, why not benefit from the impression of a friendly human touch as well?
I'd love to know more about these operations, thanks for mentioning this.
Sadly I know very little more about it, other than being able to confirm that it really is dehumanizing. You're essentially being paid for your capability of performing simple natural language processing. The amount of courtesy that call-center people get is already pretty low, and bundle in that the frustrated person thinks they're talking to a robot that literally doesn't have feelings... he had some Bad Days (though he did say that you could let a lot of things slide by just reminding yourself of how comically nonsensical the whole situation is).
I'm just speculating here, but I suspect it means that the employee needs less training. All he needs is a quick flow-chart to route the calls. No training for how to talk to customers, no liability regarding things he might say, etc.
> For many of these systems, you actually are talking to a human.
Is this practice widespread at all? I've worked at several contact center software companies with many customers and never once heard of this. It's always been automated IVR software controlling this. This sounds to my ears like maybe an outlier of a particularly awful contact center? I'd certainly pause before claiming this is how it works for "many of these systems".
That's actually a good question -- I don't know. My only window into this world was through him, and from his point of view it seemed widespread. I'll update my comment accordingly.
Wow, I'm sorry your friend had to go through that. This is the first time I've heard that this job actually exists!
It seems very strange to have a human who can listen but can't speak. I guess these companies think that by removing speech as a possibility, they're able to pay the employee less or something?
I think it's less about having to pay them less and more about less "customer interface" training, less concern about liability with regards to what the person might say, not having to worry about the potential stigma associated with a foreign accent, etc. But I'm just speculating here.
There's also a rise of interfaces where talking to a human is not an option, and it's actually very challenging to break out of the default set of options. AirBNB is an example of this. This might be OK for the majority of calls, but when you need something outside of the normal set of problems, good luck. They're almost trying to not document the problems that they haven't already documented.
Note that a lot of the time, the point of those interfaces isn't to direct your call, so much as it is to collect information ahead-of-time, and transcribe it with voice-recognition, so that when you do begin speaking to a representative, they already have your question in front of them on the screen.
And yet for some reason I have to tell them my account number after I have already entered it on a keyboard. This frustrates me beyond belief as it says to me they have disparate software systems that don't talk to one another.
I've noticed a sharp increase over the last year or two maybe in the number of companies that will bypass the awful classifier and connect you to a human immediately if you yell the word "fuck" into the phone because the machine thinks it's got an angry customer on the line. IME it weirdly has to be the word "fuck" and it has to be loud. I don't pretend to know all the ins-and-outs of IVR systems, but it seems like maybe one of the bigger service providers these companies use have this one weird trick programmed into them?
Oddly, my problem with Google is that it's too flexible. Most of the time, it understands me, and that's great. But when Alexa doesn't understand, there's usually an error. When Google doesn't, it seems to call people, cancel navigation, and do all sorts of loony things I didn't ask for.
80% success and 20% no action might be usable for an assistant, but adding in 5% random behavior makes it drastically worse than useless.
I find voice interfaces absolutely infuriating to use! I invariably have to put on an American or English accent in order to get them to understand anything I say, and even then it takes several attempts.
It's not unlike a programming language that attempts to resemble written English. Sure it looks like English at first, but really there is a rigid API there that you must adhere to, and that breaks any resemblance to natural language; at which point you wonder, why make it look like English in the first place?
There's an issue of trust, as well. When I call a support number, and I get a robotic voice that assures me I can speak to it as if it were a person in plain language, I simply don't believe it.
It's just not true, either. So I end up trying to figure out how to structure my query so the robot on the other end will understand what I want, instead of just saying what I want.
In the end, I just repeat "human" and mash the 0 button over and over until I get a real person to talk to.
Some are actively designed to make you give up on the call, at least that is what it seems like to me.
How many times have you called and the prompts go:
Press 1 to talk with sales
Press 2 to talk with marketing
....
Press 9 to talk to tech support, the only reason anybody dialed this number
Then:
Speak your 18 digit account number, being sure to pause between each digit to make sure the computer records it correctly.
Then:
Speak your phone number
Then:
Speak your 24 digit hexadecimal product code
Finally you get through to a person and 100% of the time they ask you for all of that information again so they can type it in (and get it wrong).
And even when you get a person on the line they make you go back and do all of the stuff you already tried before finally transferring you to someone with half a clue.
I suspect most of the time the agents hate the systems like you describe as much as the callers. They are actually trying to make things work better and faster by asking for this info, so when you have to repeat it it's a failure rather than a deliberate design decision...
Time spent talking to a robot is free, people are expensive. So the company prefers 1hr with robots + 1minute with human, over 10 minutes with each. It's customer-hostile though, raising the cost of the service in a hidden way.
So they can know who they are talking to. IDK in the US, in Spain most services use DNI (national identity document 00000000A) to identify a client and the telephone number as most easy way to know which location is the customer asking info about (people might have more than one landline, DSL or whatever service).
Spanish law also require a second ID verification to disclose some information (let's say your wifi password), in our case it's the las 4 digits of the bank account number.
Also to pipeline you into different call centers, because it won't be reasonable to demand agents to know all the tech and commercial info about the company, it's just too much.
Walgreens prescription refill has a good voice input system. It doesn't pretend to be intelligent. But you can read a long prescription number to it, speaking rapidly, and it gets it right. It's better than humans for that.
They hide key functions behind a 3D Touch, but there's absolutely no discover-ability. So you're either left trying to 3D Touch everything to see what works, or actively researching 3D Touch tricks.
As phone gestures become more popular, they'll have the same issue.
This is a huge problem with phone interfaces in general. There's a big push to make them "clean" by removing obvious widgets and basically hoping the user guesses correctly that they need a horizontal half screen swipe to bring up the search bar or tap and hold on an item to bring up options. You're basically just jabbing at the thing until it does what you want. Eventually people figure out various developer's favorite tricks and try them first and it isn't quite so bad, but for new users smartphone interfaces can be quite daunting.
Seriously, watch someone on a new iPhone and it's just painful.
Oh, I created this contact for fun, now how do I delete it. Tap it? Nope. Tap and hold? Nope. Double tap? Nope. Is there a menu somewhere? Nope. Maybe I slide it over? Nope, that brings me back to the previous page. Maybe I pinch? Nope. Guess I'll pull up the help, oh there isn't any. Off to Google then. Oh, I have to slide it over from the middle, not from the edge.
In other words: they were never a thing. They were a media-driven and fad-investor-driven hype wave. They never should have been a thing. Good riddance.
And I don't even want to blame the interfaces entirely. To me the real problem is people building and deploying things when there is a giant gap between hype and reality.
The Web was legitimately the next big thing. And after that, mobile. Both of them have changed our lives in deep and lasting ways. But we as an industry are absurdly hungry for the next, next big thing.
How many dumb-ass voice and bot and AI and blockchain projects are there out there now? That basically don't work, but have been shipped anyhow? How many millions of dollars have been wasted? And really I should say billions. Theranos alone burned through $1.2 billion of hype. And there was the wave of "Uber for X" companies, busily failing to replicate the business model of a company whose success still isn't a given.
I should be clear that I'm not opposed to trying new stuff. I'm all for it! But I think if we explore technological possibilities with less flagrant waste, we'll learn more. And be able to explore more.
There are a lot of smart and motivated people putting effort into voice interfaces. It's the past (we've been speaking for over a hundred thousand years) and it is the future. It will take some time, but I'm pretty confident that computers interacting with our auditory cortex will replace small slabs of glass that we look at and touch for many tasks.
I think that's highly unlikely. We've had radio for 100 years. The written word is still thriving and TV gets twice as much time from people as radio does. Audio's fine for some things, but it's so very limited.
And as an aside, we haven't really had voice-only interfaces for 100k years. Really, they've only existed since the telephone. What existed previous to that was humans, whose in-person interactions are almost always far more than voice. People have different estimates of the amount of information conveyed in a conversation through expression, gesture, posture, glance, and the like, but it's never a small amount.
You did say "replace small slabs of glass that we look at and touch", so I think "voice-only" was a reasonable interpretation.
As I said, I'm all for trying new stuff. We should look at the extent to which computers can usefully leverage that channel. But I don't think we should presume that it will be particularly useful.
I think voice is greatly underrated but that it will be the next generations (our kids) will use them without feeling weirded out.
When I look how my kids interact with Google home it's becomes fairly obvious to me that to them this is completely natural. Google Home is almost like a pet to them not just a tool.
We are finally at a stage where voice recognition starts to become powerful enough to understand nuances now the next question is what to connect them to. One thing that I really like is that it allow us to retrieve information without having to look at a screen. It feels like having a 5th person at the table.
At First Principle, we built a little a voice app that allows you to ask Google Analytics or Salesforce for data (and potentially whatever you want to connect with) for meetings so we can ask instead of having to look up. It becomes a natural part of the conversation and everyone have access to the data.
That's where I think it will first make an impact. In meetings with relevant data.
Chatbots proposed to solve a problem that current technology is bad at: understanding natural language. Very few experts claimed that algorithms could master the combinatorics of natural language beyond a few very narrow domains. On top of the technical risk, the voice and chat UX does not give a lot of visibility to users, as this post points out. What happened was: algorithmic advances in other areas lead to analogous reasoning by non-experts whose self-promotion aligned with the media's need for something new, big and understandable. And then it failed.
The irony, of course, is that research is making strides in NLU, it's just too late for the last wave of chatbots. Here are two recent papers from DeepMind:
It's a funny coincidence. I've just gotten off a phone with a bank officer about a simple question "Is my card there yet?". It took 29 minutes, including a few minutes at the beginning talking to a bot before getting to a human. I'm being unfair and emotional, but right now I'm not sure if humans are really better.
I was trying to get in contact with my bank to allow a large transaction through that they were blocking. Was impossible to talk to a real human, super frustrating. I finally decided to say I wanted to "open a new account"; was immediately connected to a human that helped me from there.
Humans are bad interfaces too. That’s why we have computers. People only like humans because they are also human. Nobody will miss the long lines at McD, but they might miss the smile. If a robot smiles at me, I won’t be happy, I’ll be scared.
What's really sad is that this has come around multiple times. I remember hearing about intelligent agents coming anytime to MSN Messenger or whatever it was called at the time back in the early to mid 2000s. And people have had this delusion since ELIZA that natural language interfaces are just on the cusp. A good demo is very convincing, but reality creeps in as soon as you use these things for more than a few rote interactions. The only real innovation of the latest generation of natural language interfaces is their ability to somewhat reliably understand the actual words you use. But even that is highly context dependent and limited to straightforward constructions. And it's all built on just having huge massive datasets against which to compare what's being said. But "natural" language doesn't depend on having past analogues. "Natural" language is constantly finding new ways to say things. And I've not seen any evidence we're any closer to that now than we were in 1966.
Very interesting point, I never thought about why I hate voice interfaces like that so much.
The tipping point for me was when the AT&T small business line I used to use changed to a voice interface and included fake keyboard taping sounds after each interaction. That just felt so damn insulting.
Lately I just say ridiculous shit with these interfaces to see what happens.
A few days ago I was using one for Delta that couldn't tell the difference between "Yeah" and "Yes". Sigh.
I'd prefer randomized 'hard disk seeking' noises; not the click of death, the sort of soft muted chirping you used to get from very large and slow magnetic media as you actually heard the heads seek back and forth.
In an alternate pinch, licence some item get sounds from vidoegames and do A B testing on customer satisfaction.
Interestingly, before making this change the line did use a simple "enter account your number" prompt that was followed by a consistent combination of little clicking sounds that seemed to indicate something was happening. I found that much more pleasant than fake key taping.
I don't want to talk to it like a human. The main advantage is that you can just say "temperature outside" or "turn light off". And I expect short answers too, not the Google/Amazon boring long sentence shit.
These chat and voice interfaces are basically the public's version of a command line interface. It has the same flaws and virtues. The main difference is marketing.
Of the bots that are marketed to be more human with lots of machine learning. From my experience, they feel no more better than the original ELIZA (https://en.wikipedia.org/wiki/ELIZA) despite the leap in tech.
Depends. As the article states, the hybrid approach is becoming more common. This gives users the ability to use various applications from facebook messenger, and the UI capabilities of messenger is pretty much good enough to achieve anything.
Oh and this comes along with a modern website that can execute all those use cases too.
But then you throw in the natural language, enabling users to write complex queries in English. That and great funded teams focussing on niches.
My experiences with bots are becoming outstandingly good.
I think it's equally wrong to just write off chatbots. They are a User Interface tool that works in some areas and doesn't in others (and recently has been dramatically misapplied).
Chatbots work well as an input when your hands are otherwise occupied:
- driving/directions (Google Maps telling you where to go)
- cooking (reciting a recipe)
They work well when you're requesting a specific thing:
- "Play Everlasting Light by the Black Keys"
- "Add Eggs to my shopping list"
- Responding to Answers in Jeopardy: "What is Syracuse?"
They can work as an alternative CTA in certain narrow areas where they function like a traditional "wizard". I've seen some ecommerce stats for things like "Are you shopping for a Fathers Day Gift? Does your dad like sports? Does your dad like gadgets? Want to see some popular gift ideas?"
Where they don't work:
- complex NLP dependencies
- data entry
- when there is an expectation set that you're talking to a human.
Yes, I did. What I was responding to was more the comments in the thread so far which mostly felt like dunking on how bad chatbots were, etc.
And the article was overall hopeful about chatbots (presuming they got better NLP+AI), it didn't really get into examples, etc. that I was hoping to elicit from the HN audience.
Actually, they may not be the "next big thing", but they're still a pretty huge deal. Especially for IT and ecommerce support. The support sector is estimated to a value of a few trillions of USD / year in the coming years, because of the progressive shift of large cohorts of customers to the cloud.
The point is not to have chat-bots handle all cases, but if they can handle 80% of cases, which are generally trivial, then it'll good enough.
Of course I hate talking to a bot just like anyone else, and being cornered into a dead-end discussion with one, or to be in an infinite loop, or in a situation where the bot has no answer and does not offer another path to resolution. But, from a business perspective, it's a rather sound approach to have people go through a bot first rather than have every single complaint clog an inbox. It doesn't scale so well.
HR support within organisations, too. I'm aware of some large organistaions currently investing in chatbots right now to help people find out how many vacation days they have left, find policies relating to their job etc.
Though this issue might equally be resolved by intranet search engines no longer being painfully bad (e.g. ones powered by Microsoft Graph).
And they're often essentially an alternative for a first-level support person who is reading off a script and may not speak English very well. So even if chatbots are a fairly sucky experience most of the time, so is frontline support a lot of the time.
Bank of America, GEICO, and other mainline financials have gone this route for most of their common questions (or so I've noticed - and for some reason they're all named random women's names).
It is extremely frustrating to use them because I usually only reach out when I require help with something complex, but I could see it being useful if your metrics indicate that question X is 50% of your phone time.
I think what you're referring to are distinct from chatbots. Chatbots are meant to take input from a user and do a wide variety of tasks like a customer service representative would do, Twitter bots usually do just one thing.
With the chatbot hype, there was a healthy number of skeptics throughout: people who wondered about the technical realities of delivering satisfying UX, ones who worried about unanswered questions about business models, and ones who figured out that this future would be most beneficial to gatekeeper-platforms who'd then act as discovery facilitators for users to select from among competing bots, rather than for botmakers themselves.
It's no accident that Amazon and Google are currently leading in the consumer voice assistant market: they had large, pre-existing base of users and enough intrinsic first-party functionality to bootstrap their assistants into rudimentary usefulness, but then they built platforms where third-parties could compete for users the same way it happened for apps. In the app boom, aside from a few runaway hits, the only ones who reliably got rich were the platform-owners.
Having a chat with someone can be enlightening because the conversation can be engaging, can grow organically and the ensuing organic path can lead to insight.
Chatbots, to me, just don't feel very organic - it's very much a request/response paradigm, often with a limited set of useful responses (perhaps I have just encountered the basic ones). If that's all we're getting, I'd much rather just hit buttons on a screen, than type an entire question in the hope that the bot understands it (when it doesn't, it's just an annoying guessing game).
Plus, as other comments have mentioned, it's very hard to see what the bot is actually capable of. With a good screen based UI, the capabilities are much more obvious. If it's not fit for your purpose, you can just quickly move on.
525 comments
[ 2.0 ms ] story [ 204 ms ] threadThat was never going to be the case for sufficiently complicated tasks with enough decisions to make. Same reason Google Home/Echo are awkward interfaces for almost anything other than getting music to start playing.
GUIs on phones/desktop are tremendously productive because they're information rich and trivial to use. I can search, choose, customize, and purchase an item on Amazon in like 15-30 seconds flat. A bot isn't going to beat that for the general case.
Yeah, right now the home/echo/siri/etc seem completely worthless to me for all but the simplest tasks, but the idea is for the use cases to keep increasing.
Tools do get better and better. Automation and AI will continue to improve. It's more a question of why chatbots haven't gotten there yet and what needs to be done differently.
Egg timers too.
But stupidly, Siri only lets me have one simultaneous timer.
They used to say that GUIs were toy interfaces and only CLIs could offer real productivity. I don't know where chatbots are going but I wouldn't expect the current state of the art to be anything like where they end up.
It was possible and commonplace to invoke parameter-oriented CLIs from another process, and use the CLI as a perfectly adequate API, while other interface paradigms like TUIs and GUIs never quite achieved such ease of programmatic manipulation.
Chatbots and NL interfaces in general are thus far a failure because they neither improve on these purposes nor provide a third, unique benefit. I could see them being useful for consolidation of services (instead of comparing eight puzza places' sites, go to one third-party pizza chatbot), but there's a very, very long way to go.
The most obviously useful technology, usually is only obvious in hindsight.
The basic tech dates back to SHRDLU and ELIZA. In some ways this is similar to the partially-self-driving-car problem: if it's a mechanical interface where the human is in control it's fine, and if it's a hypothetical human-equivalent AI it's fine, but in the middle people forget to adapt to the limitations of the system. And the limitations are very severe as soon as you go outside the lines.
At that point I just push 0 and repeat "speak to a human."
Why, oh why, would any rational person think that this kind of technology was about to suddenly take over everything?
This is a typical example of groupthink delusions.
It's not as though the tones are harder to recognise!
But seriously, it's the same reason so many sites won't accept a credit card number with spaces in it. Someone writing from scratch the same thing that's been written a thousand times, but only thinking about the way they would use it (or the way they were told others would use it).
They already have those online, in the form of FAQs and knowledge bases that don't tell you anything new.
Voice-based systems currently combine all of the limitations of human conversation (e.g., low bandwidth, strongly linear, possibilities hidden) with all the limitations of computers (e.g., not very bright, highly literal, inflexible).
I think the main reason they're so popular for support is that people are bad at accounting. If I install an IVR system, I see that calls to human agents go down, saving me easily measured cash money. But I probably haven't measured the time and cognitive load burden shifted to customers, or the value lost by suboptimal use of whatever they're supporting.
I think customer service departments find them boring and want to automate them away.
In a sense, the Google search bar is a type of chatbot, but we don't converse with it in grammatical English. It doesn't present as a human, so there's no uncanny valley effect. What gets typed into Google is a sort of lingua franca that we've all collectively learned through 20+ years of increasingly capable search engines. What we need is that level of lingua franca, but for a full, state-change-driving conversation, instead of just a one-step search.
It's not the machines that need to learn, it's us.
Particularly, of course, they no longer care whether results include your required words.
One has to learn and re-learn how to get the best from these systems.
Aside, I'm impressed with Alexa on FireTV, but until it enters data in to the apps, and searches within media libraries, it remains a novelty.
The uncanny valley effect, in my opinion, is due to the fact that 99% of our speech and writing is not dictated by content or necessity, but instead driven by social tendencies.
If I ask my friend to lunch, I'll text him "lunch?". But if I ask my boss out to lunch, it's going to be something like a paragraph, explaining what I'd like to cover during the lunch.
Most of what we say is a kind of dance to ensure other people that we aren't stupid. We don't need this formality when dealing with non-human entities, so speaking in grammatical language when chatting with a bot feels incredibly stupid.
Of course one can say that this is not an exclusively AI issue. Call centers, especially offshore ones, have the same problem.
Those strings are piped through a Switch statement of static responses, or fall out into the Default response. This was the situation in 2013 when I created a speech bot in Powershell, and it sounds like the state of the art here hasn't progressed far.
What's needed is logic to dynamically build the Switch statements, or otherwise better parse human entry and build responses. There has been much work on a few different fronts, but I'm not aware of any which were wildly successful.
Cloud services backed by big data sets tend to be better although I admit I haven't tried a local copy of Dragon Naturally Speaking for a long time.
In any case, at least assuming fairly mainstream American/English accents, the voice recognition isn't really the problem any longer. Sophisticated NLP and responses are. We're a long way from virtual assistants that can do anything sophisticated.
This is also true for CLI interfaces. It illustrates the popularity of GUIs - much to the dismay of CLI enthusiasts everywhere.
Edit: Yes, a -help or "help" command can be used to list the menu. But then this command has to be known beforehand. What if a clever designer decides to use -assist or "assistance" instead?
Linux Newbie:
"I need to find out how much space I have left on this machine."
(looks it up on the internet) ... AND that's why GUIs are preferred by most non-technical computer users.EDIT: Maybe I should have said 'CLI Newbie' - I am in no-way singling out Linux here.
“I need to find out how much space I have left on this machine.”
Now what?
"Oh there's an entry called My PC" (click)
"Aha, there's my disk and little bar showing the free space"
It's two clicks if you have your wits about you, maybe a few more if its not that obvious.
I don't know why you singled out Windows. It's just as easy in Gnome or KDE etc - the point here is that CLIs can never be as intuitive as GUIs. As the original article said - Humans are very visual people. Seeing things in a visual space is much easier for us.
Personally I think the experience is very similar for noobs.
But then again, why wouldn't the Linux guy search google in the first place?
Or if it is a total newbie, look for the disk icons, the information is there as well.
Google "linux how much space on drive" and the first hit tells you exactly what you need.
Another question someone might ask is "how do I change my password". Google for "linux change password" and there is a special help box right at the top of the search results.
apropos password and you get this enormous list:
man -k -S1 password
I use man and apropos all of the time, but I have no illusion that they're new user friendly. They're amazing when you just need to look up some options or get the syntax for a system call, but if you've not assimilated the Unix way of thinking they're quite obtuse.
(Seriously though, reading the manpage for man is where I first learned about apropos.)
Apropos would be a powerful part of a "help" command, but in 50 years nobody has ever written a help command that gained enough traction to be included in the standard toolchain.
Apropos doesn’t do it’s job because it isn’t named — or integrated with — help.
The tools themselves leave a lot to be desired from a UX experience too. Things like nonsense names, bad defaults and/or no inherent intelligence in the program, means you have to specify numerous options manually for things that should be easy to automate.
The whole experience would benefit greatly by telling people about things like man/apropros on first startup and other one time tutorials.
There are some days where I feel tempted to write small helper scripts whose sole purpose would be to rename/reorient the default experience/flags of various CLI apps, so that it is not a jumbled mess full of historical accidents that can no longer be changed.
then it's a really bad one.
You really don't have this option with voice. On the other hand, when I visited my mother I was surprised that she figured out that Google Assistant can provide her with nutrition information. She would have never figured out by herself on a CLI or menu because it's very intimidating.
This somewhat contradicts my previous comment about how much I dislike voice interfaces but they probably have some strengths.
On voice interface, your best chanse is to connect to an actual intelligent being as soon as possible because getting the options is a frustration by it's own.
If I'm saying 'help' it means I've tried that and failed, so it's not clear what the benefit is.
And you get to enjoy fumbling around trying to guess whether this particular program's magic incantation to summon help is "help", "-h", "--help", "?", "/?"...
swipes to the right
> Message archived.
No wtf.
undo
Guess it doesn't.
IVR, is supposed to be 'intuitive' or 'natural' or however one's preferred marketing dialect describes it. The assumption is usually flipped - that tools are supposed to assist those ignorant of their use[1].
It has been a long time since normal users have been expected to use a CLI. If you're looking at one, you choose to. This is not at all true of IVR/Sirlextana.
[1] Of course, they don't serve the caller, they serve the robot's owner, so interests are only aligned to the extent the robot efficiently helps the caller with their goal, which adds an additional layer of opportunity for frustration.
Think back to the earliest programming courses people take:
Spamming the # key repeatedly seems to work pretty well for this too.
The most annoying part of the experience to me is the menu items are always ~25 things I can trivially do on the website. No, I'm not calling to check my balance, pay a balance, update payment information, etc. I'm calling because your website specifically said that function isn't available online and I need to speak to a representative.
I think if you're from a certain era, you just pick up the phone first and want to talk to somebody, even if it might be pretty easy to find what you're looking for online.
Because the boardroom got a huge boner when realizing they can do major cost cutting in the human department.
Amazing what they are spending to keep up illusions.
“Look, having nuclear — my uncle was a great professor and scientist and engineer, Dr. John Trump at MIT; good genes, very good genes, OK, very smart, the Wharton School of Finance, very good, very smart — you know, if you’re a conservative Republican, if I were a liberal, if, like, OK, if I ran as a liberal Democrat, they would say I’m one of the smartest people anywhere in the world — it’s true! — but when you’re a conservative Republican they try — oh, do they do a number — that’s why I always start off: Went to Wharton, was a good student, went there, went there, did this, built a fortune — you know I have to give my like credentials all the time, because we’re a little disadvantaged — but you look at the nuclear deal, the thing that really bothers me — it would have been so easy, and it’s not as important as these lives are — nuclear is powerful; my uncle explained that to me many, many years ago, the power and that was 35 years ago; he would explain the power of what’s going to happen and he was right, who would have thought? — but when you look at what’s going on with the four prisoners — now it used to be three, now it’s four — but when it was three and even now, I would have said it’s all in the messenger; fellas, and it is fellas because, you know, they don’t, they haven’t figured that the women are smarter right now than the men, so, you know, it’s gonna take them about another 150 years — but the Persians are great negotiators, the Iranians are great negotiators, so, and they, they just killed, they just killed us.”
It's basically something very fuzzy, historically the hardest thing to emulate with computers :)
"I'm smart. My uncle was smart too. I went to Wharton, he went to MIT. He was a nuclear scientist. You know, I feel like people don't give he and I the credit we deserve. We have good credentials but people don't pay attention to that. They think I'm stupid because I'm Republican. But if I ran as a Democrat, they'd think I'm smart.
Anyway, my smart uncle told me that nuclear energy is really powerful and efficient. And that was 35 years ago, so I imagine it is even more powerful and efficient now.
Now I'm going to tell you about three or four prisoners. Some are male and some are female, I think. The female ones are smarter.
But anyway, the Persians/Iranians are great negotiators. The US should have negotiated harder with the Iranians. But I feel the Iranians ended up getting a good deal regarding our nuclear negotiations. And I feel that the United States' economic and security interests have been harmed by the deal we made with the Iranians, who are smart."
> "Our electric grid system is not stable. If it can be down or you have rolling brown outs or black outs and we know that is possible. If you don't have base load that means something, some type of energy that will run 24/7, rain, shine, no matter what happens is uninterruptible. The only two things you have that does that on a uninterruptible basis is coal and nuclear"
This isn't the first time I've wondered at the applications of natural-language processing software in political [speech]writing. I'm not sure where best forum is for people to discuss these ideas, but I apologize if this was too political, regardless, for this site.
Like, was the company in financial distress, and the first product wasn't doing well, so they said "I know! We'll make a chatbot!"?
A pivot is like a "Hail Mary" pass... risky and unlikely to work -- something that only makes sense when time is running out and you have nothing left to lose.
Or... if your company pivoted when the end wasn't near, the leadership was greatly incompetent, which means you pretty much never had a chance in the first place.
Interesting to note: the poster read the signs and made the right move at the right time in any case.
They can pivot into being a company that is alive, though. Remember that Slack was a pivot from an online game.
You can also see a piece of the game on Slack's 404 page! http://slack.com/404
https://news.ycombinator.com/item?id=10662128
To go back to interfacing with a program using written language seemed like an odd step. It's never been the most efficient way of doing something, and it requires very advanced technology to accurately understand what people are trying to say, in whatever slang, shorthand, or bad spelling/grammar they use.
Besides, it's not really dead, the tech just moved to "voice assistants" rather than "chatbots" - really just spoken word rather than written word. And I'm not convinced that's the "revolution" most people are expecting either. I'll stick to clicking buttons and typing things into my terminal.
>get sword
I don't know what "sword" is.
Some of the current interactive fiction (as the genre is now called) would surprise you.
Anything by Andrew Plotkin [1] (aka Zarf) is guaranteed to be interesting. Other indie authors of renown to look for are Emily Short and Adam Cadre. But also look for new authors! Mind you, modern IF focuses less in puzzles and more in narrative or exploring the boundaries of the medium.
One of my personal favorites is "Spider and Web" by Zarf, because I love its Cold War-esque setting. Mind you, it can be difficult! The best IF games also explore the console interface itself, such as in "Fail Safe".
An example of a particularly innovative game is "Rematch" [2]: it's a single move game (i.e. you win or fail in a single input, which can be quite complicated and shows off what modern parsers can do). It's sort of a "Groundhog Day" where you must prevent a disaster in a single move, and if you lose you replay it again, and again, and again, till you get it right.
Many of these games can be played in a browser, without installing anything.
----
[0] https://ifcomp.org/
[1] https://en.wikipedia.org/wiki/Andrew_Plotkin
[2] http://ifdb.tads.org/viewgame?id=22oqimzgf8snv002
You can play it at https://classicreload.com/zork-i.html or http://textadventures.co.uk/games/view/5zyoqrsugeopel3ffhz_v...
https://en.wikipedia.org/wiki/Zork
By the way, what OP says may be true of a game like Dunnet. I don't know if it is no more maintained or if it is kept that way because nostalgia, but more recent text-based games (that is, one that could have been programmed by your father instead of your grandfather) do way better than this. Just try some popular MUD (the MMO version of text adventures), I'm sure you'll be surprised.
It seems like a lot of times when people talk about chatbots, they really mean these phone trees in text form, in which case I would agree with your sentiment.
However, be cautious in conflating chatbots with CLIs. I would say a CLI is not (always) an intuitive interface, but for a lot of problems, they are quite efficient.
The CLI style chatbots tend to be much better since they are basically CLIs in an easily accessible location (e.g. in an app on your phone).
There is only so much input you can give and subsequently evaluate if you show a GUI with 3 buttons.
Now if you restrict the chat or voice input to 3 options than the interface feels unnatural, annoying even.
Understanding natural language is much more than just a design problem. It's a grand challenge and core subfield of Computer Science. It's the original Turing Test.
The biggest manifestation of this is the self-driving car hype for when grandpa can't legally/safely drive himself, however voice-assistants also fit that mold: Something to sell to grandpa when they don't want to learn/buy a new thing and his eyesight is bad and arthritis makes typing hurt.
The first one leads to subpar bots that can't respond to most things, the second means humans have to fill the 'long-tail' of potential responses by hand, which is impossible.
They provided guard rails so to speak. If you don’t really strong hints about what you can do — and at least recognize intents you can’t handle — you’re screwed. There is always some asshole who sees “Hi, I can tell you about restaurants and entertainment. What are you looking to do?”, and answers “What’s the atomic weight of boron in nanofirkins?”
I disagree with this. It may be easier, but it is in no way faster. For those who don't use computers very often I can see it being slower. But for people who use computers on a daily basis, for mice to still be a commonly used input device is disappointing. With constant context switches, moving an object across the desk with you wrist, while keeping your fingers in the same place to click, and carefully putting the cursor just in the right place isn't nearly as fast as the same interface using typing. Or comfortable for that matter.
I'm not saying I disagree with the article as a whole.
EDIT: typo
Who knows what is really going to happen if I press Tab here? Will it really bring me to the next field or some random link in the other side of the page?
I wouldn't doubt that once the web finally settled down from this hype, people will find ways to get the full potential out of both GUI and the keyboard in a consistent way.
The web is the only reason I need a mouse during work anymore.
Yup. Spreadsheets proved this many years ago.
http://dangrover.com/blog/2016/04/20/bots-wont-replace-apps....
These interfaces are almost like a dark pattern because of how bad they are.
With Siri or Google Assistant, you soon figure out few things that this very low intelligence person can do(like telling the weather or setting an alarm) and stick with it.
This is also why I'm excited about iOS12 with all these Siri shortcuts, instead of pretending that we are talking to a smart being let's have a concrete list of things that can do.
On the other hand I do believe that these voice interfaces have some potential, just the technology is not there yet.
Here: https://news.ycombinator.com/item?id=17265683
When Siri first hit the Iphone and I was trying to figure out if it was useful for me (no), I named my phone Searle[1].
[1] https://en.wikipedia.org/wiki/Chinese_room#Chinese_room_thou...
"Hi, in a few words, what can I help you with today?"
> "billing"
"It sounds like you have a question about your bill. I can help you with that! If you can give a few words to describe the reason you are calling, I can help you with your bill."
> "billing"
"OK, let me get you to a representative who can help!"
... instead of spending 10 minutes wrangling with the vapid AI, I can actually move on with my day after speaking to a human. Was this the future we envisioned in the 90s? I think not. Some systems let you spam 0 (zero) and it transfers to a human, but more and more are requiring you to interface with the system in some way, even if disabled or impaired.
But I hear you. I recently wanted to change the ownership of my cellphone account. "Change ownership", "It sounds like you want to change cellphone plans, is that correct?", "Agent", "Okay, let me get you to a representative".
The only tasks these things are equipped to do, are the tasks that I can do via the company's online portal, and in a much less frustrating manner. I wonder who is actually using these things.
It's annoying when my issue isn't on the list, since I'll have to hear it 50 times. And it's incredibly aggravating when my issue is on the list, but the website doesn't actually work right. But it's not hard to understand why it happens.
Yeah, that's unhelpful if "the website" is a sprawling, vast wasteland. Finding a specific thing on an obese, convoluted site is like trying to find the bucket of ice cream in Siberia. I always check the site first, so if I am calling... it's because your website is pretty bad.
So if I do find the thing I need and it is broken, then I call and dodge the idiot AI, and the beleaguered agent or offers the pathetic advice of: "It's on the website". I tell them it doesn't work, and this has led to two outcomes, either the agent tries to do it and it is so broken and fucked up that they tell me that's it's down and I have to call back or try the website days or weeks later, or I am lucky and the agent has to do a tedious task that I would have preferred to do myself online.
I get an email "Hey, I need to contact you so we can proceed, what method would you prefer?" I reply "Email, please" provide the information I assume the rep needs and which the online form should have been able to ask me and offer "What else do you need".
"Is there a time I can call you and [go over the script and fill out a form on my computer, as if you'd just stayed on hold hours ago]?"
10/10
Failing that I also resort to answering in one word answers. Bots can't handle any sort of ambiguity. Even when they manage to solve the language-parsing problem, people will still just bark one word answers at anything that looks or sounds remotely non-human[1] because that's how the early AI's trained us. The future-concept of people chatting away to robots is incredibly unlikely because to do that you need a level of empathy for the person/object you are talking to. We've clearly shown so far that we just don't have it for bots.
--
[1] Like the 19th century British Explorer shouting English at the natives.
Usually you'll get it try to say stuff, give up and then say "Thanks, I'll put you through to one of our team"
The people using the automated phone services are the people without internet or even computers.
Believe me, they exist ;)
"Okay, let me get you to a representative."
"I'm sorry, it looks like you want to talk to a representative. You need to answer a few questions first."
Very HAL-like.
Old people. I'm not kidding. I used to do PCI compliance work with call centers. I've listened in on many a call to verify procedures. Call centers deal with an unending stream of older people who cannot or will not use "computers". They want to talk to a person on a phone. The people who are comfortable with the online interface only call when the online interface fails them. Old people go to the telephone first.
And there is a growing subset of old people who are functionally illiterate. I don't mean they don't know how to read, I mean their eyesight is diminished to the point that using the computer/phone/tablet is uncomfortable. Or their fingers don't move well enough for a keyboard/touchscreen. The phone is simple and reliable, not requiring either visual or physical dexterity.
Even non tech items like opening a bag of chips or box of crackers. I estimate that some packaging must require 40+ lbs to open. And I think of the "how it's made" show and how advanced manufacturing plants are. So they use all this tech to make a box of crackers as cheaply as possible, they have cameras fast enough to find and remove a single dark grain of rice in a torrential stream, but put zero thought into the UX experience and how the user can open the box.
It sounds like the simplest "tech fix" for this phone menu is once their account is found,
if (age > 60 ) sendToPerson ();
Furthermore, you could do "soft" account verification with their number. So if they call from a number that matches an elderly customer, just route it straight to a human, even though numbers can be spoofed it's not going to harm the customer to just assume it's them for that step only and fully verify their acct later.
If they don't, is that because they don't know any better, or because they actually want to use the IVR?
My guess is either they go straight to a person, or they don't know that's an option. So if I'm correct (and I think the onus would be to prove that old people think differently than everyone else in this regard), these things exist to take advantage of the ignorance of old people to save a company some money. Greaaaat.
"We are sorry our website didn't work, let me get you a human..."
Unpacking it...
Haiku: http://en.wikipedia.org/wiki/Haiku_%28operating_system%29
Penguin doll: http://en.wikipedia.org/wiki/Tux
Bearded dude with swords: http://blog.xkcd.com/2007/04/19/life-imitates-xkcd-part-ii-r...
which is a recursive reference to XKCD: http://xkcd.com/225/
Shibboleet: Shibboleth from Judges 12
http://en.wikipedia.org/wiki/Shibboleth
combined with "leet"
http://en.wikipedia.org/wiki/Leet from hacker culture
- Me" "REPRESENTATIIIVE"
- Phone: Sorry, I didn't get that. Say "representative" to talk to an agent
- Me" "REPRESENTATIIIVE"
- Phone: Sorry, I didn't get that. let me transfer you to a representative
(This specific one happens on United calls, every single time.)
Always having to ask for permission is tiring.
I need a machine to take all of the measurements and make decisions entirely without any kind of subjective judgement on my part.
Please, validate my health the same way automated testing and other such diagnostics checks work.
> one two three etc
"Please hold"
"Hello this is agent a, Can I verify your X"
> It's XXXX
"Looks like you have <totally unrelated subject to what you called"
> Yeah but, I called for Y
"Oh ok"
> Y this, Y that, I need Y to do Z
"Let me forward you to an agent that can help you"
"Hello this is Agent Q"
> I need to do Y! damn it I've been on hold and transfering for 20 minutes
"I can help you with Y, but first, I need your account info
> FREAKING A 1234
"Can you verify XYZW?"
> YES GOD DAMN IT YYYY
"Ok sir, I'll need you to FAX it in"
> What, that's technologies from the 1920s
"I'm sorry sir, is there ANYTHING ELSE I can help you out with <condescending voice>"
> PLEASE KILL ME NOW
"Would you like to take a survey?"
a week later, 7pm you get a robo call
"You had a call with Agent X how did that call go"
If its a charge on my credit card or checking, that's what I do, I give them this counter party risk. I call and do a charge back. Basically after 20 minutes with a vendor on hold, I declare that they are being unreasonable and they are unable or unwilling to refund me, I call the bank to do this charge back.
Now, what you tell the bank is critical. For whatever reason, vendors have been able to un-do my chargeback. So instead, I tell the bank that I dispute the charge -- but that I am also disputing the method of payment. So if they argue that the charge is valid in the investigation, I say that is fine -- however, the bank is not authorize on my behalf to pay them, and if they think the charge is authorized they must contact me to arrange a different method of payment. And under no circumstances can they use the bank to make payment. So yeah if they argue the charge is legit, that's fine, contact me (I wish I could put them on hold haha), and I'll pay them -- but the bank can't. The bank likes this because they are off the hook and I like it because they don't get to screw me. Worked every time.
If everyone did this, and made this a huge country party risk, these vendors would stop putting people on hold and ignoring or purposely inadequately addressing their greivences with their product/services.
It’s funny how they never keep their promises of calling you back until you actually stop giving them money. They also magically start paying more attention to what you say, and complaint letters that went unanswered for months and presumed “lost” suddenly become found and answered.
It is probably not the most complex integration technically, but it made the experience so smooth.
So if you know where the thing you're looking for is on the website, you can just launch the website and use that.
Super simple feature but it seems like for so many companies, if you go online, it redirects you to their mobile site which has been "phased out" and tells you to the get their app, which you have and would have used if it had the feature you needed.
For such a "post computer" world it seems ridiculous how often I throw my hands up and find my laptop when trying to complete an online chore from my mobile.
Can't they see what I just typed before? Seriously, how bad can this be.
> when one tries to be as clear, as brief, and as orderly as one can in what one says, and where one avoids obscurity and ambiguity
In a sense, this is not unlike an unskilled and nervous attendant who, faced with a request that they don't understand, starts to chatter more and more in hopes of eliciting the information they really need.
[1]: https://www.sas.upenn.edu/~haroldfs/dravling/grice.html
In a sort of sick and very selfish way, I find myself somehow glad that blind and deaf people exist and are protected by accessibility legislation. I've often found accessibility features, particularly those in software, to be exceptionally useful even though my vision and hearing are fine.
The guy you were talking to eventually just routed you to someone with the same job but who has a mic, because you weren't giving him anything to go on and all he'd have been able to do is push the button again to ask you the exact same question (and having a person do it who can at least vary the inflection of the question is less infuriating than seemingly being stuck in a chatbot's infinite loop).
--
Source: have a friend who used to have this job. He has thankfully moved on to something less soul sucking than being a human literally pretending to be a robot.
I'd love to know more about these operations, thanks for mentioning this.
I'm just speculating here, but I suspect it means that the employee needs less training. All he needs is a quick flow-chart to route the calls. No training for how to talk to customers, no liability regarding things he might say, etc.
He was glad when he found another job.
Is this practice widespread at all? I've worked at several contact center software companies with many customers and never once heard of this. It's always been automated IVR software controlling this. This sounds to my ears like maybe an outlier of a particularly awful contact center? I'd certainly pause before claiming this is how it works for "many of these systems".
Edit: ahh I can no longer edit it :|
It seems very strange to have a human who can listen but can't speak. I guess these companies think that by removing speech as a possibility, they're able to pay the employee less or something?
Not perfect but good enough that it provides value.
I now use voice a lot because of the quality with Google tech.
80% success and 20% no action might be usable for an assistant, but adding in 5% random behavior makes it drastically worse than useless.
Except that with text adventures I was willing to overlook an obtuse parser if I was enjoying the game. I never call my bank just for the fun of it.
It's just not true, either. So I end up trying to figure out how to structure my query so the robot on the other end will understand what I want, instead of just saying what I want.
In the end, I just repeat "human" and mash the 0 button over and over until I get a real person to talk to.
How many times have you called and the prompts go:
Press 1 to talk with sales Press 2 to talk with marketing .... Press 9 to talk to tech support, the only reason anybody dialed this number
Then: Speak your 18 digit account number, being sure to pause between each digit to make sure the computer records it correctly.
Then: Speak your phone number
Then: Speak your 24 digit hexadecimal product code
Finally you get through to a person and 100% of the time they ask you for all of that information again so they can type it in (and get it wrong).
And even when you get a person on the line they make you go back and do all of the stuff you already tried before finally transferring you to someone with half a clue.
a) The backoffice is really shit, broken, and doesn't even display any info (down, badly designed, etc)
b) Backoffice and/or VPN connection from outsourced call center to ISP is really slow, so they work faster by asking info.
c) CRM/Customer database has no consisten quality information, so agents do not trust it.
I may forget something, but those are the most common ones.
Spanish law also require a second ID verification to disclose some information (let's say your wifi password), in our case it's the las 4 digits of the bank account number.
Also to pipeline you into different call centers, because it won't be reasonable to demand agents to know all the tech and commercial info about the company, it's just too much.
They hide key functions behind a 3D Touch, but there's absolutely no discover-ability. So you're either left trying to 3D Touch everything to see what works, or actively researching 3D Touch tricks.
As phone gestures become more popular, they'll have the same issue.
Seriously, watch someone on a new iPhone and it's just painful.
Oh, I created this contact for fun, now how do I delete it. Tap it? Nope. Tap and hold? Nope. Double tap? Nope. Is there a menu somewhere? Nope. Maybe I slide it over? Nope, that brings me back to the previous page. Maybe I pinch? Nope. Guess I'll pull up the help, oh there isn't any. Off to Google then. Oh, I have to slide it over from the middle, not from the edge.
The Web was legitimately the next big thing. And after that, mobile. Both of them have changed our lives in deep and lasting ways. But we as an industry are absurdly hungry for the next, next big thing.
How many dumb-ass voice and bot and AI and blockchain projects are there out there now? That basically don't work, but have been shipped anyhow? How many millions of dollars have been wasted? And really I should say billions. Theranos alone burned through $1.2 billion of hype. And there was the wave of "Uber for X" companies, busily failing to replicate the business model of a company whose success still isn't a given.
I should be clear that I'm not opposed to trying new stuff. I'm all for it! But I think if we explore technological possibilities with less flagrant waste, we'll learn more. And be able to explore more.
And as an aside, we haven't really had voice-only interfaces for 100k years. Really, they've only existed since the telephone. What existed previous to that was humans, whose in-person interactions are almost always far more than voice. People have different estimates of the amount of information conveyed in a conversation through expression, gesture, posture, glance, and the like, but it's never a small amount.
My point is that voice is a powerful and ancient channel, and it's about time that computers leverage that channel.
As I said, I'm all for trying new stuff. We should look at the extent to which computers can usefully leverage that channel. But I don't think we should presume that it will be particularly useful.
When I look how my kids interact with Google home it's becomes fairly obvious to me that to them this is completely natural. Google Home is almost like a pet to them not just a tool.
We are finally at a stage where voice recognition starts to become powerful enough to understand nuances now the next question is what to connect them to. One thing that I really like is that it allow us to retrieve information without having to look at a screen. It feels like having a 5th person at the table.
At First Principle, we built a little a voice app that allows you to ask Google Analytics or Salesforce for data (and potentially whatever you want to connect with) for meetings so we can ask instead of having to look up. It becomes a natural part of the conversation and everyone have access to the data.
That's where I think it will first make an impact. In meetings with relevant data.
The irony, of course, is that research is making strides in NLU, it's just too late for the last wave of chatbots. Here are two recent papers from DeepMind:
Learning to Follow Language Instructions with Adversarial Reward Induction https://arxiv.org/abs/1806.01946
Relational inductive biases, deep learning, and graph networks https://arxiv.org/abs/1806.01261
The tipping point for me was when the AT&T small business line I used to use changed to a voice interface and included fake keyboard taping sounds after each interaction. That just felt so damn insulting.
Lately I just say ridiculous shit with these interfaces to see what happens.
A few days ago I was using one for Delta that couldn't tell the difference between "Yeah" and "Yes". Sigh.
In an alternate pinch, licence some item get sounds from vidoegames and do A B testing on customer satisfaction.
Of the bots that are marketed to be more human with lots of machine learning. From my experience, they feel no more better than the original ELIZA (https://en.wikipedia.org/wiki/ELIZA) despite the leap in tech.
Oh and this comes along with a modern website that can execute all those use cases too.
But then you throw in the natural language, enabling users to write complex queries in English. That and great funded teams focussing on niches.
My experiences with bots are becoming outstandingly good.
Chatbots work well as an input when your hands are otherwise occupied:
- driving/directions (Google Maps telling you where to go)
- cooking (reciting a recipe)
They work well when you're requesting a specific thing:
- "Play Everlasting Light by the Black Keys"
- "Add Eggs to my shopping list"
- Responding to Answers in Jeopardy: "What is Syracuse?"
They can work as an alternative CTA in certain narrow areas where they function like a traditional "wizard". I've seen some ecommerce stats for things like "Are you shopping for a Fathers Day Gift? Does your dad like sports? Does your dad like gadgets? Want to see some popular gift ideas?"
Where they don't work:
- complex NLP dependencies
- data entry
- when there is an expectation set that you're talking to a human.
And the article was overall hopeful about chatbots (presuming they got better NLP+AI), it didn't really get into examples, etc. that I was hoping to elicit from the HN audience.
As in, not just reciting but allowing me to ask how much of X is needed or what the next step is? That would be great
The point is not to have chat-bots handle all cases, but if they can handle 80% of cases, which are generally trivial, then it'll good enough.
Of course I hate talking to a bot just like anyone else, and being cornered into a dead-end discussion with one, or to be in an infinite loop, or in a situation where the bot has no answer and does not offer another path to resolution. But, from a business perspective, it's a rather sound approach to have people go through a bot first rather than have every single complaint clog an inbox. It doesn't scale so well.
Though this issue might equally be resolved by intranet search engines no longer being painfully bad (e.g. ones powered by Microsoft Graph).
That seems like it should be something that could be a trivial CRUD website, with far less complexity.
He used the chatbot to ask what his 2016 salary was going to be after taxes :/
It is extremely frustrating to use them because I usually only reach out when I require help with something complex, but I could see it being useful if your metrics indicate that question X is 50% of your phone time.
It's no accident that Amazon and Google are currently leading in the consumer voice assistant market: they had large, pre-existing base of users and enough intrinsic first-party functionality to bootstrap their assistants into rudimentary usefulness, but then they built platforms where third-parties could compete for users the same way it happened for apps. In the app boom, aside from a few runaway hits, the only ones who reliably got rich were the platform-owners.
Chatbots, to me, just don't feel very organic - it's very much a request/response paradigm, often with a limited set of useful responses (perhaps I have just encountered the basic ones). If that's all we're getting, I'd much rather just hit buttons on a screen, than type an entire question in the hope that the bot understands it (when it doesn't, it's just an annoying guessing game).
Plus, as other comments have mentioned, it's very hard to see what the bot is actually capable of. With a good screen based UI, the capabilities are much more obvious. If it's not fit for your purpose, you can just quickly move on.