... I don’t see discussions of this nature. If we cannot answer these questions, than how are we going to make hardware behave in this manor? Keep in mind that animal’s do this from the constraints of a learning process and so will/can a machine!

Why would you want a machine to behave like a human? - we already have too many of them, and as products of evolution the engineering is somewhat compromised. It's a bit like looking at Windows Vista and saying: I need a better operating system, but before I can create it I have to understand this one.

Model-free methods and genetic algorithms (evolution) are brute force methods that can give you results without detailed understanding of your target domain.

What kind of AI do you actually need?

Kev, It’s not that we only want machines to behave like humans. But I could see how machines with intelligent abilities could be used to compliment humans. It will not be necessary or maybe even possible to exactly duplicate human behavior. Its not like we are all duplicates of one another. The mechanical structure of the device will provide a substantially different seat of behavior. Its ability to conceptualize will have a different beginning/frame of reference. However, by pursuing this path we accomplish much more than creating intelligent machines, we foster human understanding. At the heart of intelligence is the key to the human mystery of behavior. One will not be understood without the other.

Let's say you come up with a good hypothesis on how the brain works and build something that acts like a human. Monica and I could build a similar machine on entirely different principals and have it behave pretty much the same. To a third party the machines probably look equally human.

In a similar vein: if intelligence is just algorithmic, how can you be sure everyone is using the same algorithm? Maybe the fact we are all tackling this problem differently means how we think is also different.

Also, there's no guarantee that the structure of the human brain/algorithm scales well - gasoline engines are fairly limited, you can build a better one but at some point you want to switch up to a gas turbine or a rocket.

Kevin,

Your argument is brittlely hinged on the assumption that BOTH you and Monica (or any two other entities) have (or could produce) concepts sufficient to successfully inform the construction of machines that can think. This implies one of at least three interdependent conditions; 1. such concepts are available to or imaginable by researchers today, 2. an understanding of the mechanics of sentience is the only barrier standing in the way of the construction of such a machine, and 3. there are multiple general architectures from which will equally emerge, sentient capacities.

I do suspect that sentience can emerge from dissimilar hardware/software environments, but that a closer examination of the actual meta-logical architecture embodied in both will be symmetric to a great extent.

The question of capacity or extensibility or scaleability is important. What are the limits of intelligence? What are the factors that inform these limits? How do these factors effect each other? If the costs of use within the confines of these limits scale with construction complexity and energy throughput, what does the cost/benefit graph look like when one compares investment to various interesting solution spaces? Assume a perfectly elegant intelligence architecture. As intelligence scales in both ability and cost, what kinds of problems are still worth solving despite the causally indicated investment demands?

What solutions are worth the cost of the intelligence necessary to produce them? What are the costs associated with applying expensive intelligence to problems not worthy of the cost? Maybe there is a definition here (cost of computation vs. benefit of solution) for intelligence itself.

It is entirely possible that humans will rarely (if ever) be able to think up a problem that a superior intelligence will deem worthy of it's intelligence. That in fact might be the best test of intelligence. Dump the Turing Test. If our machines evolve to a level wherein they refuse to answer our questions, and provide as reason, that our questions are not important enough to spend the necessary resources to solve, then we will have truly set into motion the emergence of intelligences larger than our own.

Big hint: the above questions imply a general fitness metric.

Randall,

Just for clarity, what definition of "sentience" are you using? - http://en.wikipedia.o...

You might like this too - http://en.wikipedia.o...

Kev.

PS: Was tempted to deem your questions unworthy :-)

Kev, My main problem with Monica’s direction is that I don’t know what it is. I see nothing that defines what she terms as intuition and it seems to be at the crux of how she understands intelligence. My definition of intuition is that it is related to and a product of homeostasis. Although this is simplified for this discussion, homeostasis is the reference point of animal feelings and emotions. As animal kind experiences the environment it feels good or bad according to its homeostasis. It’s the apple cart that gets upset when things go wrong. When the memes dictate that things don’t feel right, it is because of experiences relating to prior homeostasis suggests that the current surroundings might dangerous. This is a logical process! But, only from the perspective of whatever is doing the experiencing. It will have no relevance to whomever else is observing the situation, unless they have something similar in their background. The human brain has taken a big hit in the logic department because we fail to see the logic from any other perceptive other than our own. Predictability within the animal world can’t be understood from external observance. It is a product of what has happened to an entity within its environment and therefore unique to that individual. It’s my assumption that this must be what Monica is referring to regarding human behavior being model free. If this is what she means than it is true because it is environment dependent. However, it seems to me that she should be explaining this and not leaving it to me to explain or her audience to assume it. But it can and does lend itself to understanding and programming can duplicate it. In the past I think you have spoke of consciousness. Homeostasis is one of the main concerns of what we are conscious about. If we remove it from the picture, would we have the need to form a thought? We couldn’t have feelings about anything because there would be no references. All animals have it. Without it I don’t see what would even motivate us to move. Of course without homeostasis evolution would have prevented there from being an us.

Randall, Because evolution offers the benefit of survival to those who can answer questions of complexity humans survive. However, these answers still act to serve and are a reflection homeostasis. A machine capable of thought will be reacting to a homeostasis entirely manufactured by us. By experimentation and using scaleable intellect we ought to start small and learn what it means to control it. We will be in control of what feels good etc and later what its concerns are.

Lex,

I don't think there is much more to the model-free approach than just accumulating all the data you can about a system and then seeing what rules (or models) fit, as against trying to work out what the model/rules are first and then trying to make the data fit.

Have to disagree with "Predictability within the animal world can’t be understood from external observance" (if I understand your terms correctly), that's what biologists do. Pavlov - http://en.wikipedia.o...

I'm interested in stone-cold rational (emotion free) AI, I'm not sure homeostasis is at all relevant for that.

Kev,
I too want to and do see the rational logic and it is stone cold. In the case of classical conditioning the underling behavior is easily stated as “If your need is that, do this.” The rationality is there. In the entire animal world, it is there. It isn’t necessarily understood or implemented in words but it is acted upon. And done so by very low level creatures. Reasoning is obviously not a very complicated process and is always associated with some need. To me Pavlovian conditioning sounds like “If, Than” statements.

Regarding understanding animal behavior from the inside out. To not do so is like trying to understand a gas engine by watching a car drive down the road. It would be one aspect of a gas engine, but to duplicate you must have a deeper understanding. When you referred to Pavlov, is his the description you believe to be at the base reference roots of animal behavior. I believe this to be only one facet of behavior. Freud had another. Descartes yet another. There are several. They provided useable tools to assist in the understanding of aberrant behavior. Their goals did not offer understanding to the depth necessary to duplicate animal intelligence. They did not intend to. But it seems to me that this is the task at hand. Maybe we aren’t up to this task, but would have Pavlov accomplished nothing had he not started somewhere. It has become clear to me that the current conceptualizations of intelligence begin at levels way too high. As a result we don’t understand the mechanics behind the driving forces of self-actualized behavior. Describing animal behavior starting from the senses and understanding conceptualization formation from this sensory input is just the beginning. This sensory input is referenced to current homeostatic condition and conceptualized according to homeostatic relevance. This is true from the simplest animals to the most complex. By approaching behaviors from this perspective it becomes clearly predictable what an animal behavior will be. By controlling the elements comprising an entities homeostatic condition an entity will act in any we wish, by its own volition. It will feel it necessary.

An alternative example: flight.

Man observes birds, concludes flight is possible. Builds various machines, eventually gets into the air.

Do we know the inner workings of the bird? - I don't think so. Do we need to? - to fly, no.

I also question your assumption about homeostasis. It might be true for autonomic systems, but I see no reason you can claim it's how the higher intellect works.

Kev, Homeostasis is only part of the intellectual process. If intelligence were a gas engine homeostasis would be the spark. Also, it will be the part of an autonomous robotic system that will allow us control. Evolution has shaped us to survive within the environmental niche we do. Our need structures and therefore our homeostasis would not exist the way it does if it didn’t comply with our environmental requirements. This will be true of an intelligent robotic system as well. The major difference is their environmental niche will need to revolve around us. We will need to control what its needs allow the robotic system to do.

Regarding your example of a bird, the key elements of flight are observable externally. There’s little gained elsewhere. But intelligence, that is an internal process. If it is going to be duplicated, it will need to be understood from the inside out.

Perhaps we are talking about two different things here. Is what you are talking about simply improving systems that imitate intelligence more effectively?