The Difference Between AI, Machine Learning, and Deep Learning • Hacker News Board

Re: The Difference Between AI, Machine Learning, and Deep Learning

#101

Post by ktta » Sat, Aug 20, 2016, 5:43 AM UTC

seanwilson wrote: post #95

ktta wrote: post #94

Earlier quoted context omitted.

>I don't think anyone would call an algorithm that in theory can apply to lots of domains intelligent. You need a concrete implementation to demonstrate this How will not applying the same technique/algorithm to lots of domains disqualify from being intelligent? RL and CNNs are being applied in many domains today. Is the complexity of the algorithm is what you're disagreeing with? The human brain (if you strip out st…

Hmm, I think we're misunderstanding each other. Is your view that RL and CNN have been or can be applied to create a machine you'd call intelligent right now? My original comment was about how people shift the goal posts about what AI is. My opinion is that in the same way you'd struggle to call a human intelligent if literally all they could do was play genius level Go, most people would not call a computer program…

>My original comment was about how people shift the goal posts about what AI is. My opinion is that in the same way you'd struggle to call a human intelligent if literally all they could do was play genius level Go, most people would not call a computer program intelligent if all it could do was play Go.

I totally agree with you on that one.

>Hmm, I think we're misunderstanding each other. Is your view that RL and CNN have been or can be applied to create a machine you'd call intelligent right now?

Not really. I was making a point to this:

>I don't think anyone would call an algorithm that in theory can apply to lots of domains intelligent. You need a concrete implementation to demonstrate this.

I was saying that even the human mind is one complex algorithm and we consider that intelligence. Something being understandable and us being able to apply to multiple domains can be intelligence as we see it today. It's just that we don't know what the complex algorithm is.

>If the machine could play many other games and adapt to games it hasn't seen before that's more convincing, but you would expect an intelligent machine to be able to adapt to more varied tasks as well (e.g. having conversations, writing stories, doing maths, recognising objects). Right now, we have AIs that are genius level at one task that cannot even attempt other tasks e.g. genius level at Chess or even a whole category of games but couldn't have a basic conversation.

This is exactly what I'm getting at. If we do find out the algorithm that effective replicates human intelligence, isn't that intelligence according to the definition of intelligence? If we find out the complex human algorithm, then we'd be able to do all the things that you just described

I think where our misunderstanding arises is that I'm saying you're disconnecting the working of human brain and algorithms. I'm saying human brain is one big algorithm and your statement :

>I don't think anyone would call an algorithm that in theory can apply to lots of domains intelligent

Wouldn't hold true in the case where we find out the algorithm for emulating the brain effectively. This disconnect in separating human intelligence from being anything other than an algorithm will make us see human brain as an un-understandable mystery box, which I think it is not.

Re: The Difference Between AI, Machine Learning, and Deep Learning

#102

Post by ktta » Sat, Aug 20, 2016, 5:46 AM UTC

argonaut wrote: post #13

ktta wrote: post #12

Also, an interesting read related to this would 'the AI effect'[0]. A lot of the stuff Deep Learning/Machine Learning is able to do today would be looked at as something that only 'true' AI (whatever consensus on what that means is; I think of it as AGI) would be able to do. But as soon as we are able to solve a problem that we think (feel?) only true AI (AGI) would be able to solve, as soon as we know how it was sol…

I hear this repeated a lot. But I've never actually seen a source from, say, the 90's or the 2000s, actually saying something like "if we're able to classify images better than humans, that's true AI." I suspect it's just hearsay that's been repeated over and over again.

hey argonaut, we meet again. I think you'd be interested in the parent comments of this[0] thread.

[0]https://news.ycombinator.com/item?id=12322536

Re: The Difference Between AI, Machine Learning, and Deep Learning

#103

Post by visarga » Sat, Aug 20, 2016, 8:43 AM UTC

Cybiote wrote: post #76

visarga wrote: post #48

Earlier quoted context omitted.

> once we have a comprehensive AI program that can play many kinds of games Before AlphaGo, DeepMind released a reinforcement learning algorithm that could play many Atari games just from the raw pixels on the screen, in many games surpassing humans. The same algorithm. https://arxiv.org/pdf/1312.5602.pdf Reinforcement learning is a general framework for learning behavior from acting in an environment with the purpos…

> Atari games just from the raw pixels on the screen It's important to distinguish between what sorts of games work well under this method and what sorts do not. Games that are variations of pole balancing, like Pong, fare better than more complex games like Asteroids, Frostbite or Montezuma's Revenge. > Saying that AlphaGo is limited because it only knows to play one game, is like saying that humans are limited beca…

> The neural network is a program that is the result of a search for a function specialized to playing Go. This machine, the program that the parameters across the edges in the graph represent, is logically unable to run any other program. Lee Sedol is a Universal Machine in the Turing sense

I am sure he is using different neurons for playing Go than for playing poker. His Go-related neural net is only able to play go.

Re: The Difference Between AI, Machine Learning, and Deep Learning

#104

Post by visarga » Sat, Aug 20, 2016, 8:45 AM UTC

Cybiote wrote: post #92

Retric wrote: post #88

Earlier quoted context omitted.

That's not really my point. If software X can do handwriting recognition then having software Y call software X is usually fairly easy and has little to do with intelligence. IMO, we already have intelligent AI. It's just not intelligent the way we are used to dealing with. People don't want AI, they want a human brain in a box.

Some of these functions are easy to compose to get something impressive, true. But that is not often the case. Take the case for games and imagine we wanted a meta-algorithm to select an algorithm to apply to each game. The intelligence would then shift into, how does one select the correct algorithm for the current game in the shortest time possible? There was a recent blog post covering this and the difficulties in…

> Take the case for games and imagine we wanted a meta-algorithm to select an algorithm to apply to each game.

I take it you haven't seen the previous accomplishment of Deep Mind before they tackled Go. They used a Reinforcement Learning algorithm to play 50 Atari games - the same algo - with great results. They really created a generic learning algorithm.

Re: The Difference Between AI, Machine Learning, and Deep Learning

#105

Post by visarga » Sat, Aug 20, 2016, 8:51 AM UTC

seanwilson wrote: post #95

ktta wrote: post #94

Earlier quoted context omitted.

>I don't think anyone would call an algorithm that in theory can apply to lots of domains intelligent. You need a concrete implementation to demonstrate this How will not applying the same technique/algorithm to lots of domains disqualify from being intelligent? RL and CNNs are being applied in many domains today. Is the complexity of the algorithm is what you're disagreeing with? The human brain (if you strip out st…

Hmm, I think we're misunderstanding each other. Is your view that RL and CNN have been or can be applied to create a machine you'd call intelligent right now? My original comment was about how people shift the goal posts about what AI is. My opinion is that in the same way you'd struggle to call a human intelligent if literally all they could do was play genius level Go, most people would not call a computer program…

> My opinion is that in the same way you'd struggle to call a human intelligent if literally all they could do was play genius level Go, most people would not call a computer program intelligent if all it could do was play Go.

Let's see, how many skills/domains can a particular human cover? For example, I can't speak Chinese. I haven't learned Chinese. Also, I have no idea about medicine. But people who learned medicine, know a great deal about it. Maybe a human can do 20-100 things, like walking, low level addition and multiplication, speaking a few languages, playing a few games and working in a few domains. Not an infinite list. In the same way RL systems and CNNs can be used for hundreds of different applications, depending on the data they are trained on.

Also, it is not the same neural net in the brain that handles two different skills, we use specialized neural nets for any of our skills too. If you stick together a few neural nets and a controller that selects the right one for the task, you could have a "single" system doing many things, just give it training data to learn those skills. Humans take 20 years to learn the necessary skills to function in society too. Neural nets can do it much faster and often surpass humans. DeepMind started Go a couple of years back and surpassed the best human player - how is it possible to do that, in such a short time span? And it wasn't a case of 'clever tricks' like the chess program Deep Blue.

Re: The Difference Between AI, Machine Learning, and Deep Learning

#106

Post by Cybiote » Sat, Aug 20, 2016, 5:04 PM UTC

visarga wrote: post #103

Cybiote wrote: post #76

Earlier quoted context omitted.

> Atari games just from the raw pixels on the screen It's important to distinguish between what sorts of games work well under this method and what sorts do not. Games that are variations of pole balancing, like Pong, fare better than more complex games like Asteroids, Frostbite or Montezuma's Revenge. > Saying that AlphaGo is limited because it only knows to play one game, is like saying that humans are limited beca…

> The neural network is a program that is the result of a search for a function specialized to playing Go. This machine, the program that the parameters across the edges in the graph represent, is logically unable to run any other program. Lee Sedol is a Universal Machine in the Turing sense I am sure he is using different neurons for playing Go than for playing poker. His Go-related neural net is only able to play g…

That doesn't make sense. The brain is highly interconnected and there's no such thing as a "Go" or poker area. Many of the same structures are recruited for different tasks. For example, the part for recognizing faces will also be recruited for reading letters or playing Go. But the important thing here is that AlphaGo is a fixed program. It's no different than chrome or firefox in that it can no longer do learning based modifications on itself. In a sense, it's actually more limited than a browser in that it cannot simulate arbitrary Turing Machines. As a feedforward network, it's not Turing Complete.

Lee Sedol meanwhile is at least as capable as a Universal Turing Machine and was learning far more per game and modifying himself while also doing the highly complex tasks of vision integrated motion planning.

Re: The Difference Between AI, Machine Learning, and Deep Learning

#107

Post by Cybiote » Sat, Aug 20, 2016, 5:13 PM UTC

visarga wrote: post #104

Cybiote wrote: post #92

Earlier quoted context omitted.

Some of these functions are easy to compose to get something impressive, true. But that is not often the case. Take the case for games and imagine we wanted a meta-algorithm to select an algorithm to apply to each game. The intelligence would then shift into, how does one select the correct algorithm for the current game in the shortest time possible? There was a recent blog post covering this and the difficulties in…

> Take the case for games and imagine we wanted a meta-algorithm to select an algorithm to apply to each game. I take it you haven't seen the previous accomplishment of Deep Mind before they tackled Go. They used a Reinforcement Learning algorithm to play 50 Atari games - the same algo - with great results. They really created a generic learning algorithm.

I'm fully informed about this area of research. Including other research that found simple linear methods could also get good results over a large number of games and DeepMind's recent work where far less computationally involved methods as random projections and Nearest neighbors outperformed Deep Reinforcement learners at the more complex 3D mazes and Frostbite.

But like I keep emphasizing, you can't take a neural net trained on space invaders and have it play Asteroids because each is a task specialized program that was the result of a search. While the search method is more general, the resulting program is not. You can use a single algorithm as simple as linear methods based reinforcement learning and get great results across a wide swathe of tasks but you can't claim to have found a universal learner.