Submitted by “Artificial intelligence is 50 years old this summer, and while computers can beat the world’s best chess players, we still can’t get them to think like a 4-year-old. This week in Boston, some of the field’s leading practitioners are gathering to examine this most ambitious of computer research fields, which at once has managed to exceed, and fall short of, our grandest expectations.”
In fifty years, AI has given us heuristic search. What it hasn’t given us is any insight at all into either intelligence or the working of the brain.
And it gave us heuristic search forty nine years ago.
well, that, and LISP. It definitely gave us lots of inspired sets of parenthesis.
Neural nets are more modern and pretty interesting.
But it is pretty amazing how 50 years ago people thought computers would be world chess champions within a year or two. And that is a fairly simple problem, just requiring brute force in the end.
Actually, neural networks are an old concept, dating back to 1940s. Warren McCullouch and Walter Pitts proposed a neuron model in 1943 where each neuron could be on or off, and would turn on given enough stimulus from neighbouring neurons. In 1950, Marvin Minsky and Dean Edmonds even built a neural network computer (consisting of some 3000 vacuum tubes).
Source: Artificial Intelligence – A Modern Approach by S. Russell and P. Norvig, 2003.
I think what we have learned from artificial intelligence research is what problems are hard and what are easier. Or what problems are solvable with a device that can add two binary numbers together. However, I think the idea that we should make machines act and think like we do is misguided. Late Edsger Dijkstra put it marvellously in EWD936*:
You see, it is not only that the question “Can machines think?” is regularly raised; we can – and should – deal with that by pointing out that it is just as relevant as the equally burning question “Can submarines swim?”
He was also of the opinion that it’s senseless to duplicate the behaviour of an inefficient blob of biomatter (i.e. the brain). Instead, we should build something better.
*) http://www.cs.utexas.edu/users/EWD/transcriptions/EWD09xx/EWD936.ht…
In fifty years, AI has given us heuristic search. What it hasn’t given us is any insight at all into either intelligence or the working of the brain.
Actually, AI gave us a lot of insight into how the brain doesn’t work. And that’s valuable in and of itself.
You can thank John Connor for stopping skynet and saving us all.
<rant> Or you can thank Intel for being anticompetetive. </rant>
AI has not happened yet because the brain does not work like a CPU. The brain does not execute instructions. What the brain does is pattern matching on experiences (sight, sound, smell, touch etc).
The brain’s responses are calculated with parallelism: the brain is a huge parallel machine.
Neural networks simulations proved that artificial brains are possible, but the technology to simulate a system so vast like the human brain has not been invented yet.
The human brain has over 200 billion synapses…the best computers can do is to simulate up to 100,000 synapses.
Exactly. Cpu doesn’t resemble brain in the slightest. It can compute, but will not process data (senses, images etc.) in this huge hyper-parrallell manner. Moreover brain is not a binary system. Every synaps reacts in a different way dependent on environtmental variables (chemical composition, stress etc.).
It’s not even those 200 bln synapsys (I thought there was order of magnitude less of them), but the overwhelming number of possible connections that they can create. Think sth. like factorial of 20bln (when 100! is some billions times the number of atoms in an observable area of the universe). If you want to build a computer behaving like a brain of a living organism, than you better not try to be smarter than mother nature (whith quantum mechanics and billions of year of evolution as her’s allies).
As to the numerical simulations, I think we are only constrained by available memory and … software. It doesn’t have to be real time in order to be intelligent. But I would be more than just inclined, there is need for non deterministic mechanisms incorporated from ground up (sth resembling quantum mechanics implications). Maybe $RANDOM will do, but I don’t really know . This begs a question if we really have free will, but it probably depends on how we define ourselves. And just as a side note – if sth gives results just as a living, thinking (i.e.intelligent) organism would do, doesn’t mean it’s intelligent. Simulation is just it – a simulation, I think the divergence is very important here, although AFAIR Touring’s Test doesn’t define intelligence in this manner.
Anyway, there will be no Skynet any time soon (BTW – I love Terminator teil 1 – watched it .. 15 times or so, Cameron is a Genius), but the future of AI is bright, since we now can acknowledge that there is tremendous work to be done to achive even slightes progress in this matter and hopefully do sth about it.
Edited 2006-07-18 11:25
We are not constrained by CPU power or available memory. How can we be if we do not even know what it is we are simulating?
We do not know how the brain functions. We do not know how the brain encodes information (i.e. the elusive neural code). Is it rate based? Do the timings of individual spikes matter? Do individual neurons matter? Is it noisy? or is it chaotic? Since we do not even know the fundamentals of how the brain processes information, simulating all those billions of synapses isn’t going to help us in anyway. The fundamental problem of neural networks is much much more basic.
> This begs a question if we really have free will
Of course we don’t.
If it all happens by chance, then there’s no free will by definition.
Otherwise, if I decide something about something else in this very moment, whatever decision I take has only two origins:
1) it’s the result of all that happened to me and all other decisions I’ve taken since I was born (notice the recursivity, here), things that have shaped me and made me become the way I am now, and that have, therefore, made me take that decision and not another one.
2) it’s the result of the way my DNA is coded
Either way, I don’t see space for free will.
Of course we don’t.
I can assure you, that nothing is so sure about it.
If it all happens by chance, then there’s no free will by definition.
You said it : “if”. How do you know, that you really don’t have _any_ influence in what you do, if everything you do contradicts this notion. You always make choices, but the questions is: is it really _you_ who makes the decision, or just laws of physics do it for you. I am inclined to believe, that even though on a basic molecular/subatomic level almost everything seems to be ruled by quantum mechanics i.e. probability, but on a bigger scale it’s not that simple anymore. Brain is a much bigger structure, and QM doesn’t apply that simply to it (it doesn’t domintate it’s functioning as a whole). Even the synapses are to big to be directly and relevantly influenced by QM. The question is (again): how is the transfer from the micro world to macro world performed with regards to cognitive processess and ability to think and what is the role of QM in it. I really doubt that I don’t have any real control over my actions (not that I couldn’t live with such a consiousness).
> I can assure you, that nothing is so sure about it.
It’s a matter of logic. Find a flaw in my reasoning, and you may have a point. Otherwise, the logic is on my part.
> You said it : “if”. How do you know, that you really
> don’t have _any_ influence in what you do, if
> everything you do contradicts this notion.
I said “if” in the sense that if that’s the case, then there’s nothing to discuss because the matter is already settled. Then I went on with the other two possible cases.
Let’s recap.
There are 3 cases to be considered, when I take a decision. I take a decision either
1) because of chance
or
2) because my dna is coded that way
or
3) because all my past experiences led me to take that decision and not another one.
This is a recursive definition of “causality of a decision“: 3 is defined in terms of 2 and 1, and 2 and 1 can happen independently.
I can’t see any more cases to consider, can you? And none of those cases contemplates free will.
It’s a matter of logic. Find a flaw in my reasoning, and you may have a point. Otherwise, the logic is on my part.
Be carefull what you wish for
– brain is about 5kg, right ? Quantum mechanics, therefore probablilistic apporoach applies to microworld, and 5kg definitely does not belong to microworld. You can’t just simply apply quantum logic when dealing with “macro” objects. Yes – every “big” object consists of small “micro” objects, which obey general rules of QM, but that doesn’t make them all of a sudden unpredictible (statistics comes to play).
because my dna is coded that way
You are taking this reasoning way to far. DNA doesn’t make you make decisions. It codes information about protein structures and such. Brain is the decision maker, with free will, or without. Of course everything has influence on your decisions, but it does not make those decisions by itself – it only contributes some tiny bit of influence on your central neural system.
In case we are talking about different things (hope this is not the case) consider this “experiment” : put 2 things of whatever in front of you (coions?). Then pick one. Now – was it possible that you would pick another one ? Who made this choice ? You ? Your DNA ? People you’ve grown up with, society, or someone else ? You probably will say, that is all at the same time. But I will argue, that you wrongly interpret ourselves. DNA is part of you. People who you know are partly the same as you are (more so, than with people that don’t know each other). There is no such thing like exclusive/independent judgement/thought/action. But that still does not deprive us of our free will.
Do you wan’t to say, that what we really are beings capable of (abstract) reasoning and we don’t have slightest control on our actions and are in fact unsuscpecting spectators (puppets) ?
> if sth gives results just as a living, thinking
> (i.e.intelligent) organism would do, doesn’t mean it’s
> intelligent.
If it walks like a duck, swims like a duck and quacks like a duck, then it’s a duck.
Really, how do you know the person next to you is intelligent?
This sentence about duck etc. is taken from discussion about programming languages – far from intelligence subject. I guess Turing hadn’t forseen in what way will the computing science -> evolve <- (pun intended). You could code very sofisticated data base of possible answers, which whould mimick human reactions, but that wouldn’t of course be anything that we could call ‘intelligent’.
How do I know if someone is intelligent ? First – I was referring to ‘intelligence’ as a process of ability to perform cognitive processess, to be able to understand at all. Well – I usually assume, that person netxt to me is a human being, a living organism, which resembles in it myself (brain included). And that usually implies some level of intelligence.
If we put some “dumb ass” before computer with a program which is able to make some conversation with humans (text being a medium), then this individual wouldn’t probably have easy task recognising if this computer is intelligent or not.
Moreover some people are obviously able to attribute intelligence to objects which inherently can not possess this “feature” (superstitions, religions etc.). It doesn’t mean that objects of their perceived intelligence are indeed intelligent. It says more about their “state of mind” instead.
> This sentence about duck etc. is taken from
> discussion about programming languages – far from
> intelligence subject.
Actually, it’s the poet James Whitcomb Riley who said it first, and it had nothing to do with computers at all. I used that quote to illustrate what problem we’re dealing with, and how to deal with it.
> I guess Turing hadn’t forseen in what way will the
> computing science -> evolve <- (pun intended). You
> could code very sofisticated data base of possible
> answers, which whould mimick human reactions, but
> that wouldn’t of course be anything that we could
> call ‘intelligent’.
No, you couldn’t, that’s the point. There’s not enough space in the universe to store the infinite amount of possible answers to the infinite amount of possible questions.
> How do I know if someone is intelligent ? First – I
> was referring to ‘intelligence’ as a process of
> ability to perform cognitive processess, to be able
> to understand at all. ù
Right, and of course an intelligent person, by definition, posseses that ability, doesn’t she?
> Well – I usually assume, that person netxt to me is a
> human being, a living organism, which resembles in it
> myself (brain included).
> And that usually implies some level of intelligence.
Sorry, but you assume that an entity that looks similar to you implies that entity has some level of intelligenge? How’s that, then, any different from the duck quote?
> if we put some “dumb ass” before computer with a
> program which is able to make some conversation with
> humans (text being a medium), then this individual
> wouldn’t probably have easy task recognising if this
> computer is intelligent or not.
You’re saying a “dumbass” is not intelligent? So, a person is either intelligent or it isn’t? It appears to me that there are many degrees of intelligence, not just 2. A person not being able to recognize an artificial entity as artificial, implies that the intelligence of that entity is at least on par of the one of the person.
Think of it, and you’ll realize that’s exactly the process you undertake any time you speak with someone on the phone.
> Moreover some people are obviously able to attribute
> intelligence to objects which inherently can not
> possess this “feature” (superstitions, religions
> etc.). It doesn’t mean that objects of their
> perceived intelligence are indeed intelligent. It
> says more about their “state of mind” instead.
Never seen anyone attributing some kind of intelligence to any object, really. Any examples of such “intelligence” you’re talking about?
Actually, it’s the poet James Whitcomb Riley who said it first, and it had nothing to do with computers at all.
I didn’t say who said it first, but assumed you are reffering to the discussion about strong vs weak typing , and since this is site devoted to operating systems and not poetry, this wasn’t that far fetched after all. On a side note : good to know it – thanks.
No, you couldn’t, that’s the point. There’s not enough space in the universe to store the infinite amount of possible answers to the infinite amount of possible questions.
Usually there is more than just one correct answer, if you are not closely fallowing logicaly correct statements. Most questions and answers are probably not the most acurate ones. And show me one person who will be able to ask every single one question out of this infinite. Single man’s knowledge is usually very limited.
Right, and of course an intelligent person, by definition, posseses that ability, doesn’t she?
Yes, but I wanted to know that we are talking about the same thing. What’s wrong with that ?
Sorry, but you assume that an entity that looks similar to you implies that entity has some level of intelligenge?
When it has two legs (not really necessary), two hands (not obligatory, but…) head, and talks to me, than I assume it is capable of thinking … yes.
You’re saying a “dumbass” is not intelligent?
I was trying to say, that “dumbass” will have really huge problem deciding whether he is dealing with intelligent entity or not – he wouldn’t probably know what answers to ask to reveal the true.
It appears to me that there are many degrees of intelligence, not just 2
Now you are REALLY nitpicking. If you really think, that that’s what I wanted to say …
A person not being able to recognize an artificial entity as artificial, implies that the intelligence of that entity is at least on par of the one of the person.
I do not agree. You could apply very complex (in fact it could be infinitely complex) parser (which in itself is NOT intelligent) to the questions being asked. This combined with set of rules “understood” by a program could make it look like intelligent, while obviously being not.
Think of it, and you’ll realize that’s exactly the process you undertake any time you speak with someone on the phone.
No. You assume, that the “person” on the other end of the phone is actually the one which he/she claims he/she is. I assume, but never can be absolute certain – how then can I be sure of the intelligence ? To little information, but judging by the probability of such an event (my interlocutor being not being intelligent being I know of) I conveniently assume, that is not the case. But how can you be sure ?
Never seen anyone attributing some kind of intelligence to any object, really. Any examples of such “intelligence” you’re talking about?
Every “supernatural” thing can be (I think) perceived as an example. The whole religion stuff is also, but I don’t want to go this topic right now if you don’t mind.
Maybe I am wrong, but I have a fealing, that this discussion became more of a flamewar, than it should be.
The human brain has over 200 billion synapses…the best computers can do is to simulate up to 100,000 synapses.
Accoding to Penrose and Hameroff[1][2] it’s even worse. They suggest that every single neuron has roughly 10^7 tubulins involved in quantum computations gving rise to conciousness.
[1] http://www.quantumconsciousness.org/
[2] http://en.wikipedia.org/wiki/Orch-OR
I wouldn’t take Penrose too seriously on this. He’s out of his field and there’s no evidence to support his thesis.
Sir Roger should stick to the mathematics of general relativity.
The brain has a very simple model of work:
1) the sensors create an experience (sight, hearing, smell, touch, etc).
2) the experience is fed to the brain.
3) the brain does pattern matching on the experience to recall a reaction.
4) the reaction has a feedback. The feedback is sent to the nervous system.
For example, when a person hears a ‘hey you’, the brain recalls a reaction: for example, to turn our head towards the sound.
When our body leans from one side, the brain recalls a the balancing reaction.
Through this simple mechanism, consciousness arises. Since the brain has a great storage capability, it is able to store experiences about the process of thinking.
I’ll repeat the Dead Like Me quote used in one of my columns:
“Yeah, when a computer loses it with a meter maid, or kills its self because it thinks it’s too fat; then I will believe in artificial intelligence.”
The brain simply does not resemble a processor in any way. The brain has synapses that are plastic, i.e. they do not always behave the same way. Synaptic pathways can be strengthened and weakened by the things we do, see, smell, hear, taste. In early life, when plasticity is at its peak, they can form completely new synaptic pathways. The brain is so overly complex, even at the synaptic level, it is simply amazing. Take a few years of Psychology like I did, and there they’ll teach you how the brain works. You’ll be amazed– and you’ll also immediatly understand why AI will NEVER even closesly resemble a human being.
The human brain evolved over a few hundred millions of years. The moment the first life was formed, development of our brain started out. Do you really think we will be able to squeeze that process into 50 years?
Ah, but it’s not a matter of whether the brain is like a CPU. It’s a matter of whether or not intelligence has a mathematical model.
This is why, despite it’s fifty year history of overpromising and under delivering, AI is still popular. There are those that hope to settle the determinism versus free will debate once and for all.
I, on the other hand, am of the the opinion that the brain has about as much to do with intelligence as the heart has with emotion.