Strong AI Thesis - The Question of computerized minds.

[micahrp]

Reading the wiki on Searle Chinese Room (sounds familiar but I dont remember it from class), all that is shown is that learning cannot be done without context which is as true for humans as it is for machines. The english speaking human would learn what the symbols meant if someone else pulled up pictures of the objects/ideas the symbols described (i.e. gave the human some context) and eventually he would be reading both sides. Correlating multiple data sets (usually through interaction and trial and error) into sets of rules and models will be the central task of a true AI.

The Post Office hand writing analysis program was given examples that it visually processed and was given the context of what those symbols represented. Through repetition it created rules for how to recognize the visual with the context. Isn't that exactly how we all learned how to read?

 

[Flamingpenguin]

The definition of "thought" is difficult to define outright, but I believe that computers will pass these tests. I believe that it is a misconception that humans are able to make decisions that are not at the most basic level purely deterministic. As a construct which makes decisions based upon many smaller decisions, it can become simple to forget that the smaller decisions are even being made. I believe we can program computers to be unaware of their simpler decisions (in fact, this is an ability computers have more so than humans) such that feelings become possible to program. For example, a feeling of fright could be represented as the conclusion that many of the possibilities of a task set before the computer would result in danger.

I don't think being deterministic in nature means that computers cannot make decisions like humans. On the contrary, I believe it gives them the ability to act like us. Like creator, like creation.

 

[Heronblade]

The Turing test is in my opinion not a reasonable benchmark for sentience. A program that bests it would certainly be impressive, but simulating human reaction is relatively easy these days, it is only a matter of time until the simulations become fluid and expansive enough to pass the Turing test, all without the program being able to truly think for itself.

At the same time, it is entirely conceivable to have a true AI that can indeed think for itself and is self-aware, but is not gifted enough socially to fool judges into thinking it is human.

In any event, I suspect our best bet for creating the first true AIs will involve mimicry. We have an example of self recursive programming on hand already. Finding a way to simulate some of the core processes and development of the human brain should allow us to "grow" a digital mind in much the same way we raise a child. (although fewer diaper changes and teen pregnancies can be expected)

The primary challenge then would be keeping it sane. For the first few attempts, we would probably have to slow the processing time down to be approximately equal to ours. Otherwise the mind in question would be waiting what seems like millenia at a time for new input on our part.

 

[Da Orky Man]

I'm not the only one who, upon seeing the title, though 'Danyal is back', am I?

Anyway, the way I'm seeing it is that any test a human can succeed at, a sufficiently powerful computer with sufficient programming can as well, even if its a direct emulation of a human mind. Of course, getting a sufficiently powerful computer may be rather expensive, especially if Moore's Law breaks down, and the programming required is far beyond our current skills.

 

[Fluffythepoo]

When computers can process at the same rate as a human brain and software can actually use that power, computers will pass the test.. if you think its not going to happen then go fuck yourself weve only been working on these things for 80 years and the human brain took 80 million, imagine computers in 80 million years... yeah, thats what i thought