The problem of designing an AI to understand what it sees

@humy said
@KellyJay
You clearly haven't been keeping yourself up to speed with modern technology.
There are already clear examples of AIs that are self-taught, i.e do NOT just blindly follow some program their creators made for them, and are now common place.
A good example of that is alphazero;

https://en.wikipedia.org/wiki/AlphaZero

This AI can beat all grandmasters at chess b ...[text shortened]... man can do (possibly excluding having feelings and emotions but even that might eventually change!).

Examples of computers doing what they were made to do isn't AI any more than a calculator doing specific math functions are. A computer will not take it upon its self to do things not programmed into it by will. Either the actions are predictable by design or their useless! We cannot have a computer decided to not do the math properly while running banking balances! Re-enforcing the rule, they don't do what we want, only what we tell them. It makes for great sci-fiction, likely the idea has its roots in evolution beginning at the hands of man, thinking we can do something godlike.

@kellyjay said
Examples of computers doing what they were made to do isn't AI

Don't know what you are talking about; An AI doing something it was designed to do is AI.

A computer will not take it upon its self to do things not programmed into it by will.

I have just given you an example of an AI doing something without 'will' and without being preprogrammed to do it the way it did it.

We cannot have a computer decided to not do the math properly

Why on earth would we won't an AI to NOT do maths correctly? And isn't doing maths incorrectly a sign of LACK of intelligence?

they don't do what we want, only what we tell them.

unless what we tell them to do is what we want; don't know what's your point here.

It makes for great sci-fiction, likely the idea has its roots in evolution beginning at the hands of man

the theory of evolution doesn't say man did it. I'm not even aware of that one in any sci-fiction; I must have missed that episode despite being a sci-fiction fan. In fact, before now, I had NEVER heard of that idea in ANY context. I don't see how you could be so confused by what the theory of evolution is.

@humy said
Don't know what you are talking about; An AI doing something it was designed to do is AI.

A computer will not take it upon its self to do things not programmed into it by will.

I have just given you an example of an AI doing something without 'will' and without being preprogrammed to do it the way it did it.
[quote] We cannot have a computer decided to not ...[text shortened]... at idea in ANY context. I don't see how you could be so confused by what the theory of evolution is.

An AI doing what it is designed to do is not intelligent it’s no different than throwing a rock and watching it fall to the ground. The rock isn’t thinking I will go up and then I will go down, it is only reacting to the universe no thought on how it will react is involved.

@kellyjay said
An AI doing what it is designed to do is not intelligent

Wrong. Its intelligent. Not yet anything like intelligent as us, but intelligent.
Our genes + our experiences determine our brain structure and thoughts and behavior; So we aren't intelligent?

One of the things I plan to eventually research an develop is and AI that, with a genetic algorithm, changes and improves its own program, including the genetic algorithm itself, with experience and trial and error so it ends up with a completely different and much more complex and advanced program with enhanced general learning ability. Of course, I have to give it an initial learning program including a genetic algorithm component else it would have no starting point. I have already done some research on that starting point but needs much more work done on it.

@humy said
Wrong. Its intelligent. Not yet anything like intelligent as us, but intelligent.
Our genes + our experiences determine our brain structure and thoughts and behavior; So we aren't intelligent?

A piss ant as more intelligence than any computer! There is neither knowledge, desires, cares, needs, or any preferences within computers; they are without understanding. In coding, a hash table and arrays are not memory experiences. They are things that we call memory as a description of what we are using to hold information. I can put something in my pocket for use later that too would not mean my pocket has a memory. I do believe we are intelligent; by design, and we were designed by one so far beyond our capabilities, it isn't measurable.

@kellyjay said
There is neither knowledge,

You keep asserting/implying without evidence or any stated premise the obvious falsity that an AI cannot have "knowledge". In what sense an AI cannot have "knowledge"? -and what do YOU mean by "knowledge" that is so different from what everybody else means by "knowledge" that is such that an AI cannot have it?

... desires, cares, needs, ...

-none of which is required for useful AI intelligence. Why would we won't to give an AI those things? All we won't is the AI to do useful things for us, not feel. In fact, it could be argued that, even if we worked out a way to do so, we should NOT allow an AI to feel else that could open a Pandora's box!

A piss ant as more intelligence than any computer!

Even if that was true, its irrelevant to the fact that AIs can (and do) have intelligence. And even then its just a matter of time before an AI will be designed to have greater general intelligence than any human...and any ant. But, and speaking as an AI expert, I'm not even sure its true anyway as already the smartest AIs are starting to master such things as natural spoken language and, have you ever heard of an ant that can talk?

@humy said
You keep asserting/implying without evidence or any stated premise the obvious falsity that an AI cannot have "knowledge". In what sense an AI cannot have "knowledge"? -and what do YOU mean by "knowledge" that is so different from what everybody else means by "knowledge" that is such that an AI cannot have it?

... desires, cares, needs, ...

-none of which is r ...[text shortened]... g to master such things as natural spoken language and, have you ever heard of an ant that can talk?

There is knowledge in books, do you think books are aware?

@kellyjay said
There is knowledge in books, do you think books are aware?

Books cannot appropriately process the meaningful info they contain, AIs can. Therefore AIs have knowledge and "aware" has nothing to do with that as the meaning of "aware" is difficult if not impossible to define anyway. Define "aware" and "knowledge"; I don't pretend to have a complete and philosophically problem-free definition for either (well, none at all in the case of "aware" ) but I say "knowledge" necessarily involves both memory of factually correct info and ability to recall and then information-process that info in a appropriate way by appropriately use it and/or relate it to other info and nothing more; have you got a better definition? Because, if you haven't, I should point out that that means AI can have what you call "knowledge". If what you mean by "knowledge" is different from above then state that difference because, me not being a mind reader, I cannot know that difference.

@humy said
Books cannot appropriately process the meaningful info they contain, AIs can. Therefore AIs have knowledge and "aware" has nothing to do with that as the meaning of "aware" is difficult if not impossible to define anyway. Define "aware" and "knowledge"; I don't pretend to have a complete and philosophically problem-free definition for either (well, none at all in the case of "aw ...[text shortened]... above then state that difference because, me not being a mind reader, I cannot know that difference.

Computers will not either, they will do nothing unless instructed to.

@KellyJay
Tell that to Alpha Go.

@sonhouse said
@KellyJay
Tell that to Alpha Go.

He doesn't seem to either understand or accept, don't know which, that the more advanced AIs in-effect change and evolve their own program (or equivalent to 'program' such as the weights in the neural network etc) and regularly do unpredicted things and make specific unpredicted decisions that they weren't programmed (or hardware equivalent to 'programmed' ) to do by any human, often completely surprising their human creator in the process.

Often the learning process in these more advanced AIs involve repeated random selections of something thus often making the exact outcome of their learning process completely impossible to predict by the creator.
I have already made more than one (rudimentary) AI that then successfully uses what is called the Monte Carlo method (see https://en.wikipedia.org/wiki/Monte_Carlo_method) to learn solutions to some specified problems and that uses a lot of randomly generated numbers to help indirectly derive a correct solution that often surprised me and which I couldn't have predicted in advanced because I didn't tell it (i.e. program) to give that specific solution (or any other specific solution).

@humy said
He doesn't seem to either understand or accept, don't know which, that the more advanced AIs in-effect change and evolve their own program (or equivalent to 'program' such as the weights in the neural network etc) and regularly do unpredicted things and make specific unpredicted decisions that they weren't programmed (or hardware equivalent to 'programmed' ) to do by any human, ...[text shortened]... use I didn't tell it (i.e. program) to give that specific solution (or any other specific solution).

There is nothing about a computer that knows anything, it is all the hardware and software acting as we code. If don't write code to cause something to occur it doesn't occur. The computer will not make a choice it reacts to algorithms, the step by step coding steps we put in place. It will only follow the instructions given it, which as I said before they will do what we say, not always what we want reactions of surprise by coders doesn't alter this fact, it proves it.

You can have a thermostat in your home that is programmable, it isn't aware of anything, but when the designed temps are reached to activate the heater or A/C are reached it will react, by design. If you want to call a computer intelligent because it's programmed to do what it does, so then is your refrigerator, TV, radio, any numbers of equipment that reacts as they were designed to do.

@humy said
There are some potential dangers from the misuse of AI but this is true for any technology. I have no doubt there will be some future misuse of AI as this seems to be like a law that if it can be misused then it sometimes will be. But I don't see any credible special AI danger that could exist any time soon that would make me lose sleep. I once considered writing a whole book ab ...[text shortened]... think its an interesting problem but have so far I have not got around to making any start on that.

In general, I agree with this. Maybe I'm particularly worried about AI tech because I follow Elon Musk on Twitter. He's been writing about advanced and evolving AI "bot swarms" that will soon overtake social media and the internet. It seems there is not currently any regulatory oversight that might prevent this from happening.

https://twitter.com/elonmusk/status/1177099375287271424

@humy said
Right; its the understanding, expressed as certain mental processes, of defined relations (I elaborate on the technical meaning of that on request) of information of what is or what's going on. Why can't an AI be given that? My OP link shows an AI that clearly has some knowledge and understanding of the world because it can look at a picture it has never seen before of the real ...[text shortened]... what sense "diminishing"? To which aspect of life? To any particular experience? Give an example...

The difference I am talking about centers on the words "artificial and intelligence." There is no understanding in the computer; being able to run various complex calculations does not give the computer understanding; it is merely doing what it is designed to do. Current moves through it while things turn on and off. Nothing more than that is going on as it performs all of its human-designed functions, no matter what tasks we assign to it, it is all the same.

When a chess program is running, and it can outplay a human, but it does not understand chess, it is just doing what it was designed to do. It is as aware of the game chess as it is a bird singing in a birdcage in the same room. It has no awareness, no intelligence again a piss ant has more.

Someone brought up morals and having a computer grasp that. Good luck with that because logic alone cannot do that. In order to be given that kind of understanding, moral values need a given, and logic alone will not go there; that is something that transcends logic alone.