So in short, I believe intelligence is actually an extremely broad spectrum of things. It’s not, and of course you can specify more things like an individual intelligence for instance, or fish intelligence or search motor intelligence or something like that, and then it might imply something slightly different.
It would be a very, very primitive sort of artificial intelligence. Yes.
Max Welling: Thanks you very much.
I’m with you [on] all of that, but maybe because human intellect is organizing information, it’s planning beforehand, machines are doing something different just like search engines and that. Maybe I should ask this question: What’s isn’t intelligence? I mean at some point, doesn’Can it lose all its meaning if it’s enjoy it’s kind of… plenty of things? I mean like exactly what are we talking about when we when we return to intelligence? Are we speaking about problem solving? Are we speaking about adaptation or what? Or is that so meaningless that it has no definition?
So the technique we’re using to earn a whole lot of improvements in artificial intelligence, now that computers is machine learning, I guess it’s really a very simple idea. Let’s study data. Let and make forecasts to the near future. How successful is that approach … what exactly did you believe are the inherent limitations of that specific means of gaining building and knowledge intelligence?
Okay. If intelligence isn’t something that’s easily defined in one sentence. I believe there’s a whole spectrum of intelligence, and actually in systems we’re starting to see quite different sorts of intelligence. For example you may think as being smart in some way, however it s kind of intelligence right?
Fair enough. And then going back centuries before I read the first vending machines, the very first coin operated machines managed to dispense holy water and you would drop a coin in a slot and the weight of the coin would weigh down something which would start a valve, then dispense some water then, as the water had been dispersed, the coin could fall out and it would shut off again. Is that a really, very crude artificial intelligence?
But… in synthetic systems it could be something quite different. In an artificial system, you could still sense information, you could compute and process information in order to meet your clients –that is similar to providing them with anything or greater search results like this. The same phenomenon is underlying, although so that ’ s a goal that is different.
Well yeah, it is dependent on how broad you wish to specify it. I mean you could ask yourself if there is a fish intelligent. And I think a fish to some degree is intelligent as you know it has a mind, information is processed by it, it adjusts a little bit to the surroundings. So even a fish is smart, but obviously it s a whole lot less intelligent than a person.
Today, and also you mentioned learning and adaptation, so that I think those are matters which are super important pieces of being smart. So a system that can adapt and learn from its surroundings and out of experiences is a system that could keep advancing itself and consequently become better or more intelligent at its task, or accommodate if the environment is changing.
These are really important parts of being intelligent, but not necessary because you could envision a self-driving car as being completely pre-programmed. It doesn’t adapt, but it still behaves intelligently in the sense that it knows when things are happening, it knows when to overtake other cars, it understands the way to avoid collisions, etcetera.
Listen for this one-hour episode or see the full transcript at www.VoicesinAI.com
Permit ’s first do computer vision. People have many hand coded features that they would attempt to spot on the image. Right. And from there they’d make predictions or for there’s some whether a individual was in the picture or something like this. But then we essentially said,”Well let’s just throw all the pixels, all the raw pixels at a neural nets. This is a convolution of net and let the nets figure out what are the features that are right. Let this neural net learn what the right features would be to attend to when it needs to perform a certain task.” And it works a whole lot better, again because there’s many patterns that it learns to have a look at which individuals didn’t think about to look in –they seem to look at these items.
So in the event that you think about discovering some… let’s say if someone is suffering from Alzheimer’s out of a brain MRI, well you may have a look at like the size of your hippocampus also it’s known that that thing shrinks–that manhood shrinks if you’re beginning to suffer with memory problems which are correlated with Alzheimer’s. So a person put this in as a guideline and can consider this, but it ends up that there’so many, many more far more subtle patterns in that MRI scan. And if you sum all those up, then actually you can find a outlook.
Now another instance is that the Alpha Go, maybe. In Alpha Move something happened. Humans come up for how to play the sport and have analyzed this game. But Alpha Go figured things out that people can’t understand, it s too intricate. But it created the game is won by the algorithm.
In other words, it was all up to people to figure out what are the important things to examine, to feel, and also how to respond to them and if you make enough of these, actually a system like that looks like it’s ’s behaving very intelligently and really still I think nowadays, self-driving automobiles… a huge component of those cars is made from lots and lots of these rules that are hardcoded from the computer system. As they act intelligently, and if you’ve got many, many of these crude parts of intelligence together, they might look.
Concerning this Episode
So I’d say it’s a brand new paradigm that goes well beyond trying to hand code individual invented features into a system and therefore it’s far more powerful. And this is also the way of course humans work. And I don’t find a real limitation to this, right? Therefore, if you pump more data through it, in principle you may learn a lot of things–or well everything you want to learn so as to become smart.
Byron Reese: This is Voices in AI brought to you by GigaOm, and I am Byron Reese. My guest is Max Welling. He’s the Vice President, Technologies in Qualcomm. He holds a Ph.D. in theoretical physics in Utrecht University and he’s done postdoc work at Caltech, University of Toronto and other places as well. Welcome to the series!
So anything I’d say that has the intent of sensing–sort of obtaining information from its environment, computing from that information to its own benefit. In other words, to live better is the aim or to reproduce maybe is the penultimate goal. And so essentially, after you’ve obtained any info and you compute then it’s possible to act–use this information. You can act on the planet in order to bring the entire world in a state that’s beneficial for you, right? So that you can survive reproduce much better. So anything that processes advice, I would say in order to reach a target, to be able to achieve a particular goal which is reproducing or surviving.
The new paradigm I would say, that is:’Well, why are we really trying to hand code each these things which we should sense in there by hand since basically you can only do so to the level of what the human imagination really can develop with, right?”
I don’t know. I mean you can drive these items. Certainly this is some type of mechanism. I guess when there is currently sensing and this can sense, there’s a bit because it ’ s feeling the weight of a coin, of sensing and it has a response to this. It’s just like kind and a response of response that is automatic, and humans have a number of these reflexes. Should you hit on your knee with a hammer, using a paddle of a hammer such as the physician does, your knee jerks upward, so that’s actually being done via a nervous system that goes to… doesn’t actually reach your brain. I believe it’s down here somewhere on mind in the back of your backbone. Therefore it s very, very primitive, but nevertheless you can argue that it senses something and it behaves. It does something, it computes something and it acts. Therefore it s just like the very most form of intellect. Yeah.
I always like to start with the query [on] first principles, which is: What’s intelligence and why is artificial intelligence artificial? Is it not smart? Or can it be? I’ll start with that. What is intelligence and why is AI artificial?
So there is human intellect and I guess that’s the ability to plan ahead and to analyze the world, to organize information–these kinds of things. But artificial intelligence is artificial because it’s kind of in machines not in human minds. This ’s the reason we call it’artificial.’ I don’t think there is any reason why artificial intelligence couldn’t be the same or very similar to human intellect. I feel that ’s a very restricted set of intellect. And we could imagine having a whole wide array of intelligence.
But humans, they wouldn’t be able to even see those subtle patterns because it’s just like if this brain area and this brain region and this brain region, but maybe not which brain area, would kind of have this specific pattern. You know that this is a little bit of proof in favor of and such as Alzheimer’s hundreds and hundreds of these things. So that people lack the type of their capacity or the imagination to think of all these rules. And we found that allow the machine itself figure out exactly what these principles are rather than trying to hand code them and just supply a data collection. And this really is the huge change for instance with learning [as] in computer vision and speech recognition.
Listen to the one-hour episode or read the full transcript in www.VoicesinAI.com
Episode 82 of Voices in AI features host Byron Reese and Max Welling talking the nature of intelligence and its relationship with intuition, development, and need.
WellI believe it’s kind of interesting if you take a look at the history of AI. So in the previous days, there was lots of AI that was programming principles. About what would be the most of the eventualities which you may 15, so you would think. And for every of these, you would sort as an automated response to those of program that a response. And those systems did not necessarily examine data from massive amounts from which they learn to respond and would learn patterns.
How far down in simplicity could you extend that? Therefore, in case you have a pet and you have a food bowl that refills itself if it becomes empty…it’s got a weight sensor, and when the weight sensor reveals nothing in there, it opens up something and then fills it. It has a goal which is: keep the cat happy. Is a primitive sort of artificial intelligence?