Artificial Intelligence

[I_am_not_mafia]

I think we're arriving slowly at the same point but from different directions Rhythm. One thing I would like to add though is that while we may be able to optimise an artificial brain and make it more efficient in terms of the processing power and memory required to perform the same functionality as that which we can find in real brains, we actually need several orders of magnitude more processing power. This is because we also need to perform artificial evolution (or some other kind of search technique) to configure our artificial brains.

When writing any adaptive controller there are always a load of parameters that I can't even guess at so I use an evolutionary algorithm to select it for me. Even something as simple as three layers of biologically plausible neurons will take me weeks of non-stop processing. An entire evaluation of a single neural network itself will take about a second or two.

This problem is worse if you want to create an intelligence that is situated in the physical world because mechanical movement is orders of magnitude slower. And unless it is a heavily constrained environment then it will need repeated evaluations otherwise you cannot tell the difference between a good solution that was tested in a hostile environment and a poor solution that was tested in an easy environment. There are tricks and techniques you can do to get around this to some extent, but doing so makes it even more important that an adaptive function works as generically as possible. This is why we need strong AI that adapts to its environment or the use that it is put to rather than aiming to superficially emulate behaviour.

[The Grand Nudger]

Again I would ask how you know that we need several orders of magnitude more processing power, how have you determined a number even in the general?  There's no shortage of examples where a human brain has, itself, been reduced - in some cases the reduction may have been almost unbelievably substantial - and this is at least one example of a system capable of presenting the effect desired which we both point to (ignoring any other troubling detail regarding intelligence, of which there are many). Perhaps it's not the size (or density) of your system that's stalling you, but the manner in which you're using it? How have you ruled that out, particularly in light of the examples we have with regards to the human brain. How have you ruled out problems with your criteria for success, as another example? Perhaps there are -already- intelligent machines, just as there are already intelligent plants and animals (in the case of the former, no neural architecture -at all-)....we're simply looking for "human" machines, and failing to find them..deciding that there are no intelligent machines? Finally, how have you determined that our own intelligence wasn't arrived at by "emulating behavior" over successive generations, with the failure conditions being death and/or extinction? Those of us remaining being those that "emulated behavior" in a successful manner given a particular environment or task? The way we think (and the subjects of our thoughts), for example, would fail a flower.....as powerful as we perceive our intelligence to be, and as powerful as it is, it's largely useless to a creature (or system) vastly different to ourselves. Hell, my intelligence might fail me in your environment, at your task and vv - and we're both human. Just how adaptive is -our- method or manner of intelligence?

I think that you're making a great many assumptions about life and intelligence, not computing, that are impossible to justify set against each other and other assumptions stated or implied.

As I mentioned a couple of posts back, it's not your comp sci I'm questioning.

[The Grand Nudger]

Here's an interesting thought experiment.  Imagine that you are an alien from a distant planet comprised of lifeforms that human beings would consider machines, mechanical.  You see the organisms here and conclude that some portion of them are thinking, are intelligent, but according to your alien standard, artificial.   Human beings down to bacteria, where do you, as that alien species, draw the line for strong ai within this set?  What is the creature that is, essentially, the floor for "strong ai in bioautomotons".  Now, perhaps even more illuminating, the next creature down the rungs, still an intelligence, but not a "strong" intelligence, what would that be, and on what grounds does the one fail to qualify as a strong bioautomaton intelligence, where the other succeeds?

[I_am_not_mafia]

Here's an interesting thought experiment.  Imagine that you are an alien from a distant planet comprised of lifeforms that human beings would consider machines, mechanical.  You see the organisms here and conclude that some portion of them are thinking, are intelligent, but according to your alien standard, artificial.   Human beings down to bacteria, where do you, as that alien species, draw the line for strong ai within this set?  What is the creature that is, essentially, the floor for "strong ai in bioautomotons".  Now, perhaps even more illuminating, the next creature down the rungs, still an intelligence, but not a "strong" intelligence, what would that be, and on what grounds does the one fail to qualify as a strong bioautomaton intelligence, where the other succeeds?

Both the biological life and mechanical alien life in your thought experiment are strong AI because they are both wholly self organising.

Nothing has to deliberately plug values into their brains or meddle with the internals from outside. A rat that is having its pleasure centres stimulated by a scientist in a lab is no longer a completely self organising system but the two of them together are. And they together are a small part of a larger self organising system.

[I_am_not_mafia]

Again I would ask how you know that we need several orders of magnitude more processing power, how have you determined a number even in the general?  
There's no shortage of examples where a human brain has, itself, been reduced

I explained. You need to perform a search to configure your values. If you disagree then produce a massively complex, generalisable strong AI that scales and which does not need to somehow involve artificial evolution or an extensive search to configure and then we can talk. Yeah we didn't always use genetic algorithms, but the cost was still there. Scientists tried to design complex systems manually. Genetic algorithms have massively expanded what we are capable of producing.

An adaptive system that can cope with the myriad of complex situations in a generalisable and scalable way will require complexity. If it didn't we would have had managed it by now. After 50 years of using your approach we haven't.

Perhaps it's not the size (or density) of your system that's stalling you, but the manner in which you're using it?  How have you ruled that out, particularly in light of the examples we have with regards to the human brain.

I'm free to listen to any alternatives which you may have come up with, preferably in the form of a published paper.

 How have you ruled out problems with your criteria for success, as another example?  

All the time. I'm a scientist.

Perhaps there are -already- intelligent machines, just as there are already intelligent plants and animals (in the case of the former, no neural architecture -at all-)....we're simply looking for "human" machines, and failing to find them..deciding that there are no intelligent machines?

I have never said that intelligence is specifically human yet you keep assuming that this is what I am arguing.

I earlier gave a working definition of intelligence as the ability to adapt to an unknown environment.

Finally, how have you determined that our own intelligence wasn't arrived at by "emulating behavior" over successive generations, with the failure conditions being death and/or extinction?

I never said it hadn't. But if you create a wholly self organising system that can emulate general behaviour that it observes then I will call it strong AI. Most people don't try to do this though. They emulate it themselves. They don't do it for a reason. Because it's insanely difficult to do with a completely self organising system. I would love to be able to do this myself but I'm starting off even simpler.

As I mentioned a couple of posts back, it's not your comp sci I'm questioning.

But yes it is. Because the computation is the same whether you perform it on silicon or with wetware. There is nothing special about either that will give you functionality that the other won't. But silicon is faster and cells have greater connectivity and can act in parallel so it is more practical to perform different algorithms with each.

[The Grand Nudger]

Both the biological life and mechanical alien life in your thought experiment are strong AI because they are both wholly self organising.

Nothing has to deliberately plug values into their brains or meddle with the internals from outside. A rat that is having its pleasure centres stimulated by a scientist in a lab is no longer a completely self organising system but the two of them together are. And they together are a small part of a larger self organising system.

Here again I'm not sure if you're misunderstanding my question or I'm misunderstanding your response.  Would grass, for example - to you, qualify as strong ai? In your example of the rat, I'm not sure why you would rescind your designation of intelligence in the event that some outside agent was stimulating it's pleasure centers.  Outside agents stimulate my pleasure centers all the time, and yet I remain an example of intelligence, strong intelligence even -while- that's occurring (though my wife may beg to differ).  What would that have to do with the dividing lines between non-intelligence, intelligence, and strong intelligence?  

I was, more simply, asking you for some off the cuff examples of which extant systems here, on earth, you feel fall on each side of that line.  As in (as an example) People on down to rats are strong intelligence, rats down to mollusks are intelligence, and beneath that the rest is non-intelligent.  This example doesn't describe my thoughts on the matter, it's not representative -at all-, merely giving you a better idea of what I was asking.

[The Grand Nudger]

Wrote out a lengthy and detailed reply, realized that it didn't matter, because we clearly weren't having the same conversation, binned that, went with this.

-You're aiming for a number "x", a measure of processing power whereby you feel that strong AI would be possible.  

-Human beings are an example of a strong intelligence.  So some number "x" possessed -by us- is sufficient. That number "x", however, cannot be higher than the lowest number "x"  in a human being while still presenting the effect, though it may be lower, neither of us knows.  We can agree to that much between us, yes?  

-Now, granting that human beings are an example of a strong intelligence doesn't grant that we are the floor for that designation, does it?  A system may have an even lower number "x" than the lowest number "x" in human beings, and still be a strong intelligence?  Can we agree to that between us (and would you be willing to suggest a candidate system or organism which presents the effect)?

[I_am_not_mafia]

I shall try to demonstrate what I mean when I say that if you want your solution to be able to adapt to a wider variety of situations then it will by necessity be more complex. I really am trying to explain things here so please do not start to get angry with me which your latest post suggests.


Imagine that you are given the task of creating a black box which performs a specific function. Being a black box the user does not know how it works. They cannot see inside. All they see is a single input and the ten outputs.  The input allows the user to tell the black box whether it has chosen the correct output. The reason why the user wants this black box is to plug into an external system. The blackbox chooses an action and the external system returns a measurement as to how successful that action was.


Any computer programmer will tell you that this is a trivial problem. You can come up with a solution that is efficient in terms of memory and processing power. So this is what the user chooses.  They could have asked for a really complicated system using biologically plausible neural networks, or even real live cells, or some other completely self organising system. Both solutions adapt but the computer program below will be smaller, faster and more accurate than the complex system. There, well done, compared to using a live human, you've chosen something that performs better and more efficiently.

Here is some pseudo-code:

---

For each action
 Perform action
 Check input signal
 Record success of this action
end for


Create a variable to record the highest success
Create a variable to record the winning action
For each action
 Check success of action when previously performed
 if success is greater than current highest success
   current highest success = sucess;
   set winning action to current action
 end if
end for

Perform winning action forever

---

But now imagine the user takes their black box and plugs it into another application. Over time the system changes and the outputs initially chosen by the black box no longer correspond to the best choices. The black box is now useless for this task and needs to be reprogrammed.

So the user takes it back to the creator of the black box who add in a loop to periodically check all the actions before deciding on which is the best one. But how often should this check be made? The programmer decides to write a check to see whether the actions being executed bring in the same expected reward each time. But at what point should it start exploring alternatives rather than exploiting what it already does? The programmer examines the system the user wants to put it into and decides on a decent error threshold before the program needs to explore other actions.

The small and fast system has now become slightly more complicated, slightly slower and cannot now act optimally because it needs to explore alternatives.  It also now requires some values to be configured in order to operate.

The user then happily plugs this black box back into their second system and it works well although not optimally.

Now say that the user plugs it into a third system. This system suffers from a lot of noise and the best action that the blackbox needs to choose changes at irregular intervals.  The computer program sometimes works well and sometimes doesn't because of the values chosen by the programmer. The user then takes the black box back to the programmer who adds in a load of statistical techniques for filtering out noise and tracking average error thresholds over time. The programmer then needs to decide on even more values to hard code into the blackbox. Now the program has become significantly more complex in order for it to be able to adapt to a noisier more variable environment.

Now imagine the user plugs the black box into yet another system. This time the black box shouldn't decide on a single action but choose the relative strength of each output. The program is not up to this task at all and needs to be completely re-written from scratch. On the other hand, if the user had chosen the second black box to begin with, the one contained the evolved self organising neural network, then they could have re-used it each time without needing to take it back to the programmer.


Computers are very fast idiots. They follow explicit instructions. The reason for developing Artificial Intelligence is to allow them to adapt so we do not have to specify everything explicitly. At each stage the initial solution chosen by the user became more complex, required more values to be specified and in the end did not scale.

[I_am_not_mafia]

Here again I'm not sure if you're misunderstanding my question or I'm misunderstanding your response.  Would grass, for example - to you, qualify as strong ai?

We're getting into the murky problem of trying to create a definition for a spectrum of emergent phenomena. Such questions are typical, like what is life for example. Is a seed alive? I would say grass not intelligent for several reasons, others may disagree. I am working on a formal definition of intelligence that would exclude it but I am not going to jump off into these murky depths in this thread considering how badly we seem able to communicate with each other.

In your example of the rat, I'm not sure why you would rescind your designation of intelligence in the event that some outside agent was stimulating it's pleasure centers.  

Let's think about a classical neural network, a back-propagation network. You need to look at the outputs and then go through the network changing the weights yourself. Neither myself or anyone in the field would call this self organisation. The network is being changed by a third party. The same with the rat brain. A self organising system settles into a stable, or meta-stable state by itself.

[The Grand Nudger]

I binned the majority of my reply in an edit, I think that the above encapsulates what I;m hoping to learn from you. I'm going to have to remind you here that no explanation of how we model systems, or what that entails for the system is capable of clearing up the issue that I have with your claim. Not a bit. I'm having trouble accepting your claim in principle..at a fundamental level. Not in it's practical achievement. I would -need- to see your range of "x" to be able to accept that claim as true, and though neither you nor I could point to a demonstrable machine with strong AI between us, we both ought to be able to point to existing organisms with strong intelligence, or else we have no means of comparison...and thus your claim -could not- be true.

General definitions of intelligence allow for plants and even bacteria to be intelligent, even strong intelligence. In our conversation, more specificity is required. If we're using general behavior, or some observed effect in a range of systems as the measure for whether or not a system is intelligent, and how strong that intelligence is - then this needs to be bluntly laid out as the basis for that conversation. It cannot proceed in the absence of that declaration, and so we could not make claims regarding the minimum processing power as a requirement. We have to be able point to a system and say "at least that much". Once we have, we could discuss how much "that much" -is-.

(lol, @ "fast idiots" I used a variation of that alot myself back when I was in the PCB and component manufacturing business, telling people that computers are, almost by definition, retarded - so maybe they shouldn't expect them to shore up their mistakes. As a bit of background, hopefully to ameliorate any need on your part to explain how these machines work to me, I used to be a process engineer for cisco, qlogic, and DoD boards, before dropping out and getting into production management instead with the same firm. I assure you, I'm familiar with the construction and operation of comp systems at a fundamental and practical level - but we're talking about intelligence here, so that only gets me so far - as your comp sci only gets -you- so far. I -could- drown the conversation with lengthy explanations from my new career path, explaining how auxin receptors work and what it might take to make a functional or practial ai modeled out of an entirely disparate, biological, and chemical architecture...but this wouldn't be any more indicative of the relationships we're discussing than your explanations of the various models currently available within comp sci.)