# Will A.I. reach human intelligence?

Discussion in 'Science & Technology' started by Narz, Apr 24, 2010.

?

## Will A.I. reach/surpass human intelligence?

15.6%

26.7%

24.4%

11.1%

15.6%

6.7%
1. ### SouronThe Dark Lord

Joined:
Mar 9, 2003
Messages:
5,947
Location:
(GMT-5)
This point depends on the definition of square root. If you define the square root as the set of real numbers that can be multiplied by themselves once to get the specified number, you would be right. But you can just as easily define square root as the function represented by the radical sign, that maps a positive number to it's positive root, and a negative number to the root with a positive imaginary component. I'd agrue that this is the more common. If you what the other one, you can ask Wolfram to solve for x, x*x=501.

I agree, but this is not an indication of lacking intelligence. Certainly not a lack in the capability of AI, as WolframApha does this as desired.

That's what fuzzy logic is for. The Sims series are a good example of fuzzy logic in action.

I disagree. Mathematics does not always approximate nature. Often it matches spot on. For example, if there are two people on a bus, and two get on, there would now be four people on the bus. No approximation present.

I agree that, just as electricity in circuits is sometimes the flow of electrons, and other times the flow of holes, so in the brain signals are transmitted by different medium. It's still a logical signal, though that could be encoded in the flow of electrons, or whatever other way you choose to represent it.

And so Bremermanns-limit applies just as much to computers as to the brain.

2. ### mdwhChieftain

Joined:
May 14, 2005
Messages:
3,565
And how many humans would spot this point? I think it's rather reasonable to interpret "the square root" to mean the principal square root. It's also reasonable to interpret it as the square root function (i.e., as indicated with the radical symbol) - and this does only have one answer, i.e., the positive root.

From http://mathworld.wolfram.com/SquareRoot.html - whilst it says that there are indeed two square roots, it also states: "In common usage, unless otherwise specified, "the" square root is generally taken to mean the principal square root."

If this is meant to be a more general AI, rather than a straightforward calculator, I'd argue that producing these kinds of results is more expected.

3. ### SouronThe Dark Lord

Joined:
Mar 9, 2003
Messages:
5,947
Location:
(GMT-5)
Sure, but who uses them? Nobody actually thinks with this particular prepositional logic, so it not being understood is not a hindrance to AI.

So what's the challenge? You just have a different threshold for different context, and represent the likely hood of being true as a real number from 0 to 1.

So obviously computers can only do processes. Any inability to describe irrational numbers is not relevant.

In any case the description of most irrational numbers are not given with a '...'. For instance, e, the inverse of the natural logarithm, can be described as the real function that is its own derivative evaluated at 1.

But you are right, there exist processes that we cannot describe, but that we know exist. Also there a functions that are known to be uncomputable. So my statement that anything in math can be done by a computer is wrong. But the majority of mathematics is concerned with actually computing things, which a computer certainly can do. Take a precise process, and mathematics is by nature precise, and it is possible to program a computer to follow it.

4. ### SpoonwoodGrand Philosopher

Joined:
Apr 30, 2008
Messages:
4,791
Location:
Ohio
Most fuzzy logic texts that I've seen basically imply that we can't do this.

If we talk about numbers and collect them into a set, such as {1, 2, 3, 4} we've basically homogenized all the objects as the same. That works fine with abstract numbers. But, if we start talking about people, we have some respects in which all people come as the same, and some respects in which they don't. The bus example treats all those people as the same, while ignoring any possible aspects in which those people differ. In other words, such an analysis only considers part of what goes on in the real world, and only with that in mind can one speak precisely about the number of people on the bus. I don't think natue acts in such a way.

I don't know. But, I don't see how such bears on assessing artifical intelligence. Care to explain what you see as the connection?

But, see with logics like that, it becomes clear we don't (fully) understand logic, because one can form as many of those logics as there exist real numbers. On top of this, we may well actually think with logics like that. How do you know what sort of logics we actually think with?

It doesn't come as a likelihood. Usually could mean that 78% doesn't fit at all, 82% fits exactly well, 84% doesn't fit at all, while 80% fits to degree .5, 83% fits to degree .5.

I think you mean that the exponentional function, the inverse of the natural logarithmic function, can get described as the real function which equals its own derivative. Then exp(1)=e. But, that doesn't describe e as rational, nor irrational. It just tells us how to compute e as close as we can.

I agree that describes mathematical practice well, and that gives us useful information about how to do mathematics. It has the advantage of making things, in principle, transparent and readily checkable. However, the precision of mathematics basically implies that we'll only explore a small portion of mathematical objects that way. Why? Well say we just talk about the real numbers. There exist far more many uncomputable real numbers than computable ones. The same holds for complex numbers since we can represent them by the form a+bi, where a and b denote ANY real numbers. The same holds for logics, since we can extend logic from {0, 1} to the interval [0, 1], and there exist as many real numbers in [0, 1] as there exist in any closed and bounded interval of the reals [a, b]. And so on...

5. ### mdwhChieftain

Joined:
May 14, 2005
Messages:
3,565
Well that was my point - I don't see how your criticism of the answer has any connection to whether we consider it AI or not.

Well, one has to prove that it's irrational from the definition. See http://en.wikipedia.org/wiki/Proof_that_e_is_irrational .

Joined:
Apr 4, 2007
Messages:
21,810
Location:
Liverpool, home of Everton FC
e is also transcendental to boot.

7. ### SouronThe Dark Lord

Joined:
Mar 9, 2003
Messages:
5,947
Location:
(GMT-5)
You don't need to know all the ways to describe "usually". You just need to pick one fuzzy membership function.

It is people who categorize things into groups like that. So if AI does it, then it is not being less human.

Further, the categorization is implied when asking the question "How many people are on the bus?". If a computer is to answer a question in this framework, then it needs to apply the model of people as being comparable units.

AI is about making machines appear human. Mdwh is Saying that it is more human to provide the principle root.

We think with the same kind of logic that we use for debate. Which is well studied. We do not think with the kind of logic you describe.

Can you describe what you see as the limitation more clearly? I am not understanding your point.

I did not mean to say that mathematics is well understood, rather that what is understood is understood precisely (for the most part). It is the precise nature of mathematics that makes it computable.

8. ### SpoonwoodGrand Philosopher

Joined:
Apr 30, 2008
Messages:
4,791
Location:
Ohio
I certainly don't believe that we think dialectically. Most of thinking happens in perception. When we perceive a rock, we don't just respond to sensory information, our brains structure the information that comes in such a way. Even academics who look at books most of their working day still spend a significant amount of time sleeping, traveling between places, and engaging with objects they see, hear, feel, taste, touch (they eat, talk, listen, move, etc.) Even to read requires sensory input, which requires a perception to get inerpreted by the brain. For writing we have to have sensory of input of where to write. So, even in supposedly "abstract" processes, we STILL have perception at some level present. Thus, perception happens throughout our processes, and thus most of thinking involves perception.

We don't engage in debate-style logic while perceiving something. We might, but only in rare cases. So, I simply don't see us as thinking with that type of logic. Any sort of logic of perceptive processes doesn't come as really all that well understood. And no doubt, there exist vastly many of them.

[ It is the precise nature of mathematics that makes it computable.]

I don't see how mathematics writ large can come as computable, even due to the precise nature of mathmetics, since most mathematical objects don't come as computable.

9. ### SouronThe Dark Lord

Joined:
Mar 9, 2003
Messages:
5,947
Location:
(GMT-5)
Perception is converting input into facts. It's not the process of reasoning.

I agree that perception is something that computers still struggle with.
I did not mean to say that all mathematical entities can be mimicked by a computer. All I meant to say was that if you want a precise description look to mathematics. And a precise description of a process is what you need to write a program.

Joined:
Apr 4, 2007
Messages:
21,810
Location:
Liverpool, home of Everton FC
11. ### Miseisle of lucy

Joined:
Apr 13, 2004
Messages:
28,495
Location:
London, UK
At the risk of making an "I agree" post, I agree with Sauron and mdwh.

12. ### SS-18 ICBMOscillator

Joined:
Mar 5, 2006
Messages:
14,769
Location:
Here and there
Taking the slightly longer bet. Human intelligence is just as irreducible as the "vital force".

13. ### Ayatollah Sothe spoof'll set you free

Joined:
Feb 20, 2002
Messages:
4,387
Location:
SE Michigan
Philosophically, those are different, but in practice, they will all be solved at about the same time. (The second one will take longer, I think, due to the need for neural-compatible inputs and outputs, and the greater need to understand human psychology.) All three versions of intelligence basically amount to having a crapload of computing power and a versatile structure for employing it.

Bingo. It worked once, with no help at all from truly intelligent design. It will work again, but more easily, because we'll "cheat" (use intelligent design to avoid local-but-not-global optima, and such).

14. ### SpoonwoodGrand Philosopher

Joined:
Apr 30, 2008
Messages:
4,791
Location:
Ohio
[Ayatollah_So] All three versions of intelligence basically amount to having a crapload of computing power and a versatile structure for employing it. [/QUOTE]

That may actually understate things.

I've found some of this discussion somewhat disturbing. Not that artifical intelligence could happen, but it seems that hardly anyone realizes how complicated even the brain of simpletons end up as. I ask that you guys take a look at this search. Just looking through a few of them, the estimates vary from 10 billion to 1 trillion. This doesn't include any other sort of glial cells, such as astrocytes which communicate with neurons. It doesn't include blood vessels. But, let's ignore that and just concentrate on the neurons in the human brain. Let's go with the smallest estimate for the number of neurons in the brain 10 billion.

Say one uses neural network models to mimick the human brain, which comes as probably the most accepted method in AI. How many neurons do neural networks use? From what I've seen, and a specialist in this area probably has a better idea, it ranges between about 5 and on the order of 1000. No one has an artificial neural network which has 10,000 or more artificial neurons at work. In order for that, we need 10 times as large a system. Just to get to 10^10 (10 billion) neurons from 1000, we need on the order of 10,000,000 times the computational power of the most complex neural networks in existence. Does anyone, with these approximations kept in mind, really believe build a computer will ever do THAT many computations? And what happens when we consider how we think and act in real-time? When we consider other structures besides neurons which influence our intelligence? What if we consider problems like the traveling salesman problem, where the number of possibilities blow up VERY quickly?

Hopefully this gives us a sense of why there exist so many approaches in AI, and how it has such a bad reputation for overstating what it will do. Hopefully, it also indicates why there exist so many approaches, and why only approximating human intelligence comes as a much more reasonable goal.

15. ### SS-18 ICBMOscillator

Joined:
Mar 5, 2006
Messages:
14,769
Location:
Here and there
Given enough nodes and enough connections, there will be enough possible paths to calculate many possibilities. It is only a matter of time and resources.

16. ### Miseisle of lucy

Joined:
Apr 13, 2004
Messages:
28,495
Location:
London, UK
Just so everyone's aware, the SyNAPSE project recently simulated a "cat-scaled brain" consisting of 1.6 billion neurons. Obviously though that's not a simulation of a cat brain by any means.

17. ### peter grimes...Moderator

Joined:
Jul 18, 2005
Messages:
13,143
Location:
Queens, New York
Yes, and what's more - I really do think that it will happen within the next 25 years, actually. Familiar with a 'hockey stick' graph? I think we're in the crook, so to speak.

18. ### SouronThe Dark Lord

Joined:
Mar 9, 2003
Messages:
5,947
Location:
(GMT-5)
Using this site http://www.willamette.edu/~gorr/classes/cs449/brain.html
Brain has 10^14 synapses operating at 100Hz.
That's 10^16 operations per second.
A modern CPU has 10^8 transistors per chip, with a 10^9 clock. Lets say that a synapse like computation may actually take 10 transistors. Then a chip can be made to do the same 10^16 operations per second.

Now that doesn't take into account interconnections, and the problem that you have to make the computer behave as if it were not clocked. On the other hand, clock much faster than 10^9 clock would be possible if only 10 transistors were used per operation.

So, A hardware neural net with the same size and speed as the human brain, if someone wants to build it, is right on the doorstep!

19. ### SpoonwoodGrand Philosopher

Joined:
Apr 30, 2008
Messages:
4,791
Location:
Ohio
Souron,

Using that site, even granting for a moment hardware with the same size and speed as the human brain, the style of a digital computer significantly differs from that of a human brain. Fault tolerance also changes (which more-or-less means the underlying logic has to change).

20. ### SouronThe Dark Lord

Joined:
Mar 9, 2003
Messages:
5,947
Location:
(GMT-5)
That site is trying to compare a conventional computer to a human brain. I'm using the same data to imagine a computer that was designed like the human brain, ie as a neural net. Such a thing would be easy enough to design and cast in hardware.

As for fault tolerance, the reason computers are not fault tolerant is because faults are so rare. Transistor don't individually fail to work.