Yes it is. Allow be to boast, but over the last 4 years I have implemented automated credit decision systems replacing human credit analysts and event recommendation systems categorizing 3,000,000 events only on the description. The power these new technologies have is on the same scale as fire and electricity.
What's your educational background? I have been following the field closely for the past five years, and I am starting to feel like staying in application development is a dead end unless I start learning AI and Machine Learning concepts at a deeper level.
If you have an affinity for math specifically Linear Algebra and Applied Statistics most of the concepts are quite straightforward to understand. When building credit scorecards, the main point during a project is to find out which characteristics rank.
Is the primary advantage of these systems over human deciders that they make better decisions or that they can be sold as being free of bias in ending?
Absolutely agreed, I have been building stock trading robots for years, but this is not AI, but automation. AI would be the computer to build me one money making robot from scratch...
Maybe adding to that is that is that from a practitioners point of view AI is marketed heavily because most of the large tech companies can increase higher utilization rates of their infrastructure
The application logic is currently smart enough. So long as the vision of AI remain around single, closed, data-oriented, monolithic applications AI will continue to be as impressive as it is now. It will get faster as the hardware gets faster, but it won't be what tech evangelists are hyping it up to be.
If the giant tech companies really want AI to be something more they need to reevaluate what AI is from a very foundational and simplistic approach. Right now, the current approach, AI is just a big smart application, but it can never be smarter than a big single software application.
I think there's an important distinction here in word choice. Sundar said more profound, not a bigger deal. You could imagine that it's more profound in the sort of "what if we make machines that can think / feel / have souls" sort of way, which certainly can't happen with fire or electricity.
That doesn't mean it's more important than electricity. (As a bozo test - how does your AI work without electricity?)
Disclaimer: work at google, in a role completely unrelated to AI.
> (As a bozo test - how does your AI work without electricity?)
You design a system that can take in a store of energy of some sort and converts it into energy through some kind of process. You might call the energy source, say, food and the process of converting into energy could be called nutrition.
That's how you'd do it if you lost electricity I guess. Create a generator that takes not-electricity and turns it into electricity.
If you really have to go without electricity, you could make it run on glucose. We have systems like that running around, just not artificial.
And since we know neurons and brains work using electric pulses, you could say that turning glucose into electricity is also a solved problem.
Of course you could also say that without electricity (as in doesn't exist as a thing) you can't have brains so you'd never be able to come up with AI in the first place.
Altho I wonder if a brain would evolve that runs on something other than electricity if electricity didn't exist as a physical phenomena. That would be interesting.
I wish we knew how to imbue robots with chemical actuation mechanisms that actually worked (as opposed to some proof of concept demo of a simple electrolysis reaction in a bulky 3D printed silicone balloon).
If real muscle cells / limbs serve as a sort of reference, they would be vastly more space-efficient, compliant, and have higher power-weight ratio than any electromagnetically actuated servo we have today.
It makes me wonder if the "modern" androids in Westworld used electricity at all.
I'd be happy if they'd just keep on working on "hey Google" until I can train it to understand me on the first try. It was like magic when I got my first Galaxy S3 but the recognition has only gotten worse.
Is AI still intelligence if it does completely stupid shit that people currently do like racial profiling? Because to me, that's not intelligence. That is fucking scary as shit though that this non-intelligent software can now amplify the worst of human impulses many times over without any oversight or control. If that's what people like like Pichai are talking about--shit software given too much power by humans--then yeah, it's beyond terrifying. But if we're still talking about intelligence, we're not talking about systems like this.
If a neural network (or any other AI algorithm) includes race as one of its input variables, and through its training set optimization utilizes that trait in order to deduce who may be guilty of a crime, and the cumulative error is indeed significantly improved by allowing that trait as an input -- then we have the need to rationally define exactly why this is bad, and what exactly the AI is missing which is needed to address that problem.
I'd imagine the answer is something along the lines of the AI needing to incur a strong penalty when it falsely accuses or falsely casts suspicion, or misses a crime it would've found without that trait distracting it, etc. But the point IMO is that the only way to stop this problem is to understand it without being afraid of talking about it. You talk of the AI as being stupid for this problem -- but the stupidity lies in the axioms we're feeding it, and that only gets fixed by finding out where WE are going wrong in a rigorous and scientific way.
Currently? Almost every published statistic involving people everywhere, includes data that you could call racial profiling. BTW I think your point could be made without the pointless expletives.
fire, electricity, and life/intelligence are result of the same - entropy equalization along the gradient between low entropy region in parameter space and the rest of the space, ie. 2nd law.
AI looks like the next step in that ladder as it isn't limited by physical limitations of a protein based life body, ie. it has all the chances to transcend the current practical limit of 1.5kg of wet live matter blob with 100B elements and 100T connections.
There's also the pseudo bias that we compromise (unknowingly) in favor off what computers are good at.
We take intricate emotions and quantify them into text Memes with a static image. (who was boasting about their AI doing text and images again?)
More and more we communicate over chat (text), taking away non-verbal communication, which has been shown biologically and psychologically to be a very real thing, and even more active in the brain than verbal speech in actual real life conversation.. Which just makes it that more easier for a computer to come across as convincingly sentient if we think a turing test on text alone is good enough.
More and more we organize our lives with things like smart watches and pre-empted structure, and we set goals with clear parameters on a todo list and in a graph. (computers love numbers and graphs creation, walk in the park for the most primitive computer)
We praise computers for having gotten so much better in logical based tasks like math and cross checking peta bytes off data though the decades, that we forgot that the entire existence for computers is that they have always been superior for these tasks in the first place by design
Last week there was a story about a truck driver trying for 20 minutes to fit through a tunnel, eventually getting stuck. When asked why, he said the phone app said it would fit. When the shape sorter is considered a test for monkey's.
People fear AI will take us over..
..
..but perhaps the real fear is us fanboy-ing at their feet.. Religiously if you will tongue in cheek
26 comments
[ 5.4 ms ] story [ 89.4 ms ] thread[1]https://e27.co/innovative-cognitive-services-new-oil-begun-r...
If the giant tech companies really want AI to be something more they need to reevaluate what AI is from a very foundational and simplistic approach. Right now, the current approach, AI is just a big smart application, but it can never be smarter than a big single software application.
That doesn't mean it's more important than electricity. (As a bozo test - how does your AI work without electricity?)
Disclaimer: work at google, in a role completely unrelated to AI.
You design a system that can take in a store of energy of some sort and converts it into energy through some kind of process. You might call the energy source, say, food and the process of converting into energy could be called nutrition.
That's how you'd do it if you lost electricity I guess. Create a generator that takes not-electricity and turns it into electricity.
If you really have to go without electricity, you could make it run on glucose. We have systems like that running around, just not artificial.
And since we know neurons and brains work using electric pulses, you could say that turning glucose into electricity is also a solved problem.
Of course you could also say that without electricity (as in doesn't exist as a thing) you can't have brains so you'd never be able to come up with AI in the first place.
Altho I wonder if a brain would evolve that runs on something other than electricity if electricity didn't exist as a physical phenomena. That would be interesting.
If real muscle cells / limbs serve as a sort of reference, they would be vastly more space-efficient, compliant, and have higher power-weight ratio than any electromagnetically actuated servo we have today.
It makes me wonder if the "modern" androids in Westworld used electricity at all.
They just don't want to let you.
I'd imagine the answer is something along the lines of the AI needing to incur a strong penalty when it falsely accuses or falsely casts suspicion, or misses a crime it would've found without that trait distracting it, etc. But the point IMO is that the only way to stop this problem is to understand it without being afraid of talking about it. You talk of the AI as being stupid for this problem -- but the stupidity lies in the axioms we're feeding it, and that only gets fixed by finding out where WE are going wrong in a rigorous and scientific way.
AI looks like the next step in that ladder as it isn't limited by physical limitations of a protein based life body, ie. it has all the chances to transcend the current practical limit of 1.5kg of wet live matter blob with 100B elements and 100T connections.
We take intricate emotions and quantify them into text Memes with a static image. (who was boasting about their AI doing text and images again?)
More and more we communicate over chat (text), taking away non-verbal communication, which has been shown biologically and psychologically to be a very real thing, and even more active in the brain than verbal speech in actual real life conversation.. Which just makes it that more easier for a computer to come across as convincingly sentient if we think a turing test on text alone is good enough.
More and more we organize our lives with things like smart watches and pre-empted structure, and we set goals with clear parameters on a todo list and in a graph. (computers love numbers and graphs creation, walk in the park for the most primitive computer)
We praise computers for having gotten so much better in logical based tasks like math and cross checking peta bytes off data though the decades, that we forgot that the entire existence for computers is that they have always been superior for these tasks in the first place by design
Last week there was a story about a truck driver trying for 20 minutes to fit through a tunnel, eventually getting stuck. When asked why, he said the phone app said it would fit. When the shape sorter is considered a test for monkey's.
People fear AI will take us over.. .. ..but perhaps the real fear is us fanboy-ing at their feet.. Religiously if you will tongue in cheek
..perhaps human do indeed never change..