That the author considers the command line an archaic mode of human-computer interaction makes it difficult for me to take his commentary on Google Duplex, let alone AI and HCI design, seriously.
Secondly, I have a hard time thinking design is anywhere close to obsolete. Google Duplex sounds like an adapter that will make reservations for you upon voice command. This will not obsolesce anything except a few forms on most websites. Design encompasses _much_ more than how a form looks or functions.
Same exact feelings bubbled into my head, but as I thought about it, is it really wrong to refer to a cli as archaic today? I posit that that it’s not when discussing application ux/i design.
What do you mean by general purpose computing? I would wager that things like "search a file for a string" or "show the first x lines of the file" count as general purpose and even these simple tasks are easier done on the cli than anywhere else. On windows/macos using the regular guis, you will run into all sorts of problems if the file type is not recognized, you will need to open the entire file in some text editor, click through the interface to find the search feature (or memorize how to use it, which is no different from memorizing the cli command), etc.
Unless you can explicitly enumerate the "small" subset of tasks the cli is simply better at, I don't think that's a valid out. The CLI (specifically, the unix philosophy) is IMO the best platform for general purpose computing. It's the specialized stuff, mostly applications with such a large feature/configuration surface that it would be foolish to try to learn all the commands and arguments, that is best left outside the cli.
General purpose for me tends to be a loose definition of communication /media in any form (web pages, IM, video, photos, games..) either consumption or creation.
CLI works incredibly well for anything related to controling systems (local, web servers), or running any kind of processes at scale (web scraping, photo metadata). Discovery is the biggest problem here, though I do believe it's solvable. If I want to do any one-off task, such as resizing an image, that's incredibly easy to locate and do within the typical operating system GUI, whereas to do it in CLI I'd have to resort to man-pages or search engines to figure out what command to execute.
The inverse is true: there is a small subset of tasks where a GUI can be so good that nothing rivals it (everything that needs the graphical part and has to have a real time feedback).
For everything else CLIs are more consistent, flexible and extendible.
But why not use both? GUIs are amazing for editing a video, command line is amazing. If you want to convert a thousand videos into different formats based on their meta data the CLI will be the only option that doesn’t makes you sit there for a week.
I prefer to think of it as eldritch. It can be difficult to master all it's features, but it's one of the few (pseudo) protocols whose age is actually it's strength.
They don't say archaic, they say "learning to code is beyond the reach of most people", which seems like a fair statement.
My kids learned how to use touchscreens as babies. They learned how to manipulate GUIs around the time they entered school. But they still regard my terminal window as black magic, and I'm not sure how to even start explaining what's happening when I punch in commands.
That said, I agree the article is overstating the impact this will have: computers talking to computers has been a thing for a long time, they just use APIs (which are carefully crafted to be as umambiguous as possible) instead attempting to parse ambiguous human speech and pipe it into arbitrary web pages.
> and I'm not sure how to even start explaining what's happening when I punch in commands.
Though perhaps not trivial, it seems easier than, say, explaining what you're doing when you fix/manipulate things on a car engine.
The terminal uses a form of language to give commands, and the result is the execution of the command and/or some sort of printed output.
`ls -lah` could be explained as 'a fast way of inputting' the equivalent of "Alexa, tell me in detail what files are in this directory". (Yes, you would have to explain something basic about files and directories, but that still seems reasonable.)
That's because finding people unfamiliar with the terminal that are willing to spend an entire hour in front of a terminal learning how to use it is the hard part.
The problem is that CLIs are undiscoverable. You either know what `ls` is or it's just a bunch of random letters. Your average user will never learn a new unforgiving language just to talk to their computer.
You can try to re-experience that frustration for yourself by playing an interactive fiction text adventure game. For example in http://adamcadre.ac/if/905.html, the first few things I tried were: "left", "go left", "move left", "map", "where am i", "help", "?", "tell me what I can type", "fuck you", "exit" (you've now exited the room and are in the living room)
Sounds like a lazily written cli, typing help in bash gives usefull information and also refers to info. So you get an overview of the available commands and a description of the basics with minimal effort.
Meanwhile I can play hide and seek with the tools I need in Gimp for hours.
Indeed. What if the cli had both natural language syntax, large alias/synonym matching, and command prediction (eg similar to Gmail's recent feature)? That might reduce the non-discoverability issue to about the same as most direct manipulation ui (where you have to look through menus for the "right" item to click on).
>and I'm not sure how to even start explaining what's happening when I punch in commands.
When you see your kids do something with a GUI ask them to write down a list of instructions so that you can do it too. That list of instructions is basically what using a CLI is, it's just less discoverable.
Agreed, though design will evolve to facilitate these sessions. My guess is that it will become more artistically / brand driven and less technical. A designer won’t need to think about the fidgety aspects of the interface and more about the presentation.
Sensationalist headline. This totally ignores the fact that not all of web traffic is transnational; a sizable percentage of it is—surprise surprise—for leisure, and in those cases there is no substitute for interacting with an interface directly.
At each stage, the most expert remaining users are left behind. Many professionals extensively use the CLI. The majority of the remaining professionals use a traditional GUI on a traditional OS. Most people using smart phone UIs are not professionals, but some of them are fairly expert in their interactions. Those using AI UIs are generally just crossing their fingers to avoid pulling their phone out.
Maybe someday AI will be a legitimate user interface, but I don't think we know what that looks like yet.
> Duplex is making websites redundant. Designers like me are now faced with the possibility that we could ‘optimize’ the experience by simply removing it altogether and have the AI interact with the server instead.
This was one of the original points of HTML and the original design of the web. The Semantic Web meant that 3rd party and automated assistants would be able to control it. Burners-Lee described it like so:
> I have a dream for the Web [in which computers] become capable of analyzing all the data on the Web – the content, links, and transactions between people and computers. A "Semantic Web", which makes this possible, has yet to emerge, but when it does, the day-to-day mechanisms of trade, bureaucracy and our daily lives will be handled by machines talking to machines. The "intelligent agents" people have touted for ages will finally materialize.
Everything goes full circle, and there is nothing new under the sun.
Where voice assistants are concerned, this is a largely unexplored UX space. Voice assistants in the future are not going to be the same as they are today, because our design trends are going to evolve as we learn more.
If those trends move towards predictable interactions over interpreted ones, and if we decide that there are times we want to use assistants without physically talking to them, will we see a resurgence of text-based assistants? Will we eventually go full circle all the way back to command lines, just under a different name?
I think today the technology is different, there are more qualified people, and the futurists and the like will make it stick.
Automation and people trying to have everything 'cloud-based' and 'secure' will probably rely on services not just like this, but like time-saving and squared-away solutions like this.
Businesses had no idea of this back when HTML was invented, and now that we understand how that works and that has become such a task that we would benefit greatly from implementing something to get rid of it, who knows!
I think it's worth a shot to automate something like this, which is already tech focused and surrounded by talent, that automating other things will become closer in scope.
That's what it's all about really, time-saving and automation.
This isn't necessarily on-topic, but how is Duplex going to deal with Captchas -- particularly when its doing 100% automated things like remembering to book an appointment in the future?
Are we going to see an industry shift away from the belief that a physical human needs to be at a website in the future? Is there going to be some kind of back-door where Duplex won't need to solve Google Captchas?
It's hard for me to look at Duplex without thinking that it's something of an admission that bots and automated assistants are a legitimate way to interact with the web, and that we should be trying to block behaviors, not agents.
Something seems off here, the google AI was probably trained on these designed forms, to be able to fill them.
If designers then stop creating these forms, then what will the AI use? Seems like some kind of API would be needed, but typical REST API is not detailed enough to support this.
So the API would somehow need to be created (potential job for UX person) or inferred from something.
I absolutely hate to use voice-controlled assistants. Not because they are annoyingly dumb, not because they can only be good in areas their developers prepared them, but because this way of communication is much slower and absolutely not private, at all. Maybe most of the population currently is living lonely, but I have family and I'm not ready to speak out my web surfing to them. Even more stupid it would be to do in office or train, plane. Only good place for voice-controlled assistants is a car, I think. And a spaceship.
(Seems i can't link the actual image but in the article there's a chart that breaks down the speed/information density of different types of communication.)
The "punch card, command line, GUI, touch screen" chart makes no sense in "facilitating HCI". Touch screens are simply an input method for GUIs. Why not include voice input, eye tracking or gestures in that case?
Wasn't 2016 supposed to be the year of the chatbot? Whatever happened to that?
These things have been around for years. There was a MIT Media Lab demo long ago. It's basically Amazon Echo for services, right? Google might make this work by insisting that services offer an API that they can call to get business from Google users. Of course, Google will want a cut of the revenue.
They'll probably get it going for food delivery and car services, then bail on anything that isn't basically online ordering. Airline reservations, maybe. Doctor appointments, probably not. Appointments with important individuals, unlikely.
An interesting article. An ‘existence proof’ of (reasonably) effective voice UI interactions is the Apple Watch. I find it liberating to leave my phone at home and still stay connected with my watch.
To use the example in the article, I can imagine making the request to Siri on my watch, having Siri read back a reservation, followed by “is this OK?”
42 comments
[ 4.7 ms ] story [ 93.8 ms ] threadSecondly, I have a hard time thinking design is anywhere close to obsolete. Google Duplex sounds like an adapter that will make reservations for you upon voice command. This will not obsolesce anything except a few forms on most websites. Design encompasses _much_ more than how a form looks or functions.
Unless you can explicitly enumerate the "small" subset of tasks the cli is simply better at, I don't think that's a valid out. The CLI (specifically, the unix philosophy) is IMO the best platform for general purpose computing. It's the specialized stuff, mostly applications with such a large feature/configuration surface that it would be foolish to try to learn all the commands and arguments, that is best left outside the cli.
CLI works incredibly well for anything related to controling systems (local, web servers), or running any kind of processes at scale (web scraping, photo metadata). Discovery is the biggest problem here, though I do believe it's solvable. If I want to do any one-off task, such as resizing an image, that's incredibly easy to locate and do within the typical operating system GUI, whereas to do it in CLI I'd have to resort to man-pages or search engines to figure out what command to execute.
For everything else CLIs are more consistent, flexible and extendible.
But why not use both? GUIs are amazing for editing a video, command line is amazing. If you want to convert a thousand videos into different formats based on their meta data the CLI will be the only option that doesn’t makes you sit there for a week.
My kids learned how to use touchscreens as babies. They learned how to manipulate GUIs around the time they entered school. But they still regard my terminal window as black magic, and I'm not sure how to even start explaining what's happening when I punch in commands.
That said, I agree the article is overstating the impact this will have: computers talking to computers has been a thing for a long time, they just use APIs (which are carefully crafted to be as umambiguous as possible) instead attempting to parse ambiguous human speech and pipe it into arbitrary web pages.
Though perhaps not trivial, it seems easier than, say, explaining what you're doing when you fix/manipulate things on a car engine.
The terminal uses a form of language to give commands, and the result is the execution of the command and/or some sort of printed output.
`ls -lah` could be explained as 'a fast way of inputting' the equivalent of "Alexa, tell me in detail what files are in this directory". (Yes, you would have to explain something basic about files and directories, but that still seems reasonable.)
You can try to re-experience that frustration for yourself by playing an interactive fiction text adventure game. For example in http://adamcadre.ac/if/905.html, the first few things I tried were: "left", "go left", "move left", "map", "where am i", "help", "?", "tell me what I can type", "fuck you", "exit" (you've now exited the room and are in the living room)
Meanwhile I can play hide and seek with the tools I need in Gimp for hours.
When you see your kids do something with a GUI ask them to write down a list of instructions so that you can do it too. That list of instructions is basically what using a CLI is, it's just less discoverable.
Maybe someday AI will be a legitimate user interface, but I don't think we know what that looks like yet.
This was one of the original points of HTML and the original design of the web. The Semantic Web meant that 3rd party and automated assistants would be able to control it. Burners-Lee described it like so:
> I have a dream for the Web [in which computers] become capable of analyzing all the data on the Web – the content, links, and transactions between people and computers. A "Semantic Web", which makes this possible, has yet to emerge, but when it does, the day-to-day mechanisms of trade, bureaucracy and our daily lives will be handled by machines talking to machines. The "intelligent agents" people have touted for ages will finally materialize.
Everything goes full circle, and there is nothing new under the sun.
Where voice assistants are concerned, this is a largely unexplored UX space. Voice assistants in the future are not going to be the same as they are today, because our design trends are going to evolve as we learn more.
If those trends move towards predictable interactions over interpreted ones, and if we decide that there are times we want to use assistants without physically talking to them, will we see a resurgence of text-based assistants? Will we eventually go full circle all the way back to command lines, just under a different name?
Automation and people trying to have everything 'cloud-based' and 'secure' will probably rely on services not just like this, but like time-saving and squared-away solutions like this.
Businesses had no idea of this back when HTML was invented, and now that we understand how that works and that has become such a task that we would benefit greatly from implementing something to get rid of it, who knows!
I think it's worth a shot to automate something like this, which is already tech focused and surrounded by talent, that automating other things will become closer in scope.
That's what it's all about really, time-saving and automation.
Are we going to see an industry shift away from the belief that a physical human needs to be at a website in the future? Is there going to be some kind of back-door where Duplex won't need to solve Google Captchas?
It's hard for me to look at Duplex without thinking that it's something of an admission that bots and automated assistants are a legitimate way to interact with the web, and that we should be trying to block behaviors, not agents.
I believe this is the entire point, and how they end up capturing an entire market that nobody else can compete with.
Recaptcha requires you to prove you're a human. Unless you're a google bot, then you're fine.
Too bad I hate wearing watches.
No thanks
If designers then stop creating these forms, then what will the AI use? Seems like some kind of API would be needed, but typical REST API is not detailed enough to support this.
So the API would somehow need to be created (potential job for UX person) or inferred from something.
(Seems i can't link the actual image but in the article there's a chart that breaks down the speed/information density of different types of communication.)
These things have been around for years. There was a MIT Media Lab demo long ago. It's basically Amazon Echo for services, right? Google might make this work by insisting that services offer an API that they can call to get business from Google users. Of course, Google will want a cut of the revenue.
They'll probably get it going for food delivery and car services, then bail on anything that isn't basically online ordering. Airline reservations, maybe. Doctor appointments, probably not. Appointments with important individuals, unlikely.
To use the example in the article, I can imagine making the request to Siri on my watch, having Siri read back a reservation, followed by “is this OK?”