You couldn't without AGI.

If computers do EXACTLY what you tell them to, then how would you make the more flexible and adaptable?

Well quantum computing is slowly coming, They sort of have it but when Ai uses it, That might open up something quite powerful.

Normal computing is like a straight 2D line, Quantum computing is like a 3D line. (an example)

Simple: you write the programs to respond to changes.

Computers are like that because they're made that way, a specific command is linked to a specific action by the computer because it was useful to have it that way. Computers don't have to be this way though, for example chess computers that learn and get better at playing chess have been around for a long time now.

引用自 shoopy：

Computers are like that because they're made that way, a specific command is linked to a specific action by the computer because it was useful to have it that way. Computers don't have to be this way though, for example chess computers that learn and get better at playing chess have been around for a long time now.

... which are also just executing their program.

引用自 Kargor：

... which are also just executing their program.

I find it funny that the 1982 TRON movie played with that premise. Namely, that the "Master Control Program" was basically a chess program before, but somehow managed to grow into something bigger, to the point it determined its OWN directive eventually and didn't care for the directives given to it by its users. I guess the real question is, is an AI program being flexible when it solves a problem a different way, or is it actually no longer a typical AI program because it would be considered "self-aware" and "sentient" at that point?

引用自 Iggy Wolf：

引用自 Kargor：

... which are also just executing their program.

I find it funny that the 1982 TRON movie played with that premise. Namely, that the "Master Control Program" was basically a chess program before, but somehow managed to grow into something bigger, to the point it determined its OWN directive eventually and didn't care for the directives given to it by its users. I guess the real question is, is an AI program being flexible when it solves a problem a different way, or is it actually no longer a typical AI program because it would be considered "self-aware" and "sentient" at that point?

Chess programs from the '80s onwards are flexible in the sense that they don't know the exact way to solve a problem, which meets the criteria of being flexible and not needing to be told exactly what to do. (if you have to tell the chess computer exactly how to play, then there's no point in having it)

It doesn't have to be self aware to be flexible. Typically they would look at their own games they played and remember which moves led to wins or losses, which is going to make the computer develop different habits depending on who it played against.

individual chess computers could sort of develop a "personality," because even if they start with the same code, they won't play the same games so they all learn a bit differently, plus if they have different hardware it will also affect how it plays and therefore how it learns, because faster multithreaded CPU's with more memory can "think" quicker and more deeply and are more likely to find optimal moves right away, where as a computer with weak hardware will do what it can to find a solution quickly which isn't usually the best solution, it's just the best one it could find within the time constraints of a game of chess. Eventually it has to stop thinking and make a move which may not be the best one.

引用自 Iggy Wolf：

I always found it funny that a lot of sci-fi movies with computer bad guys are the result of giving them a directive and then being surprised that they follow it to a T, even if it's impractical and actually goes against what they "wanted". But, if computers are programmed to follow a directive to the letter, then how would one program it to be more flexible and adaptable and accept a change in the directive, without basically BSODing itself trying to?

With Javascript.

-JS is considered very forgiving.
Too forgiving for some.
It will adapt to a certain degree to make the program work where it is possible.

The drawback is potentially it making a calculation you do not want it to while it tries to work through the users poor coding (or intentionally bad coding).