Originally Posted by Bass Thumper
Maybe this will help. Here is my crisp definition for "create":

Create: To produce, reveal or give rise to something unique or sufficiently dissimilar from that which came before. To bring into being or awareness a novel thing, idea or relationship which did not exist prior. “Create” and “discover” can be very close cousins.
For the purposes of AI, I consider this to be an insufficiently "crisp" definition, because it lacks intent.

For example, and earthquake can "create" chaos, a flood can "create" ruin, and a river can "create" a canyon.

Yet none of these forces have intent to create any of those things.

Similarly, AI doesn't have the intent to "create" art - it is the result of running an algorithm. The intent was put there by the creator of the AI. The only way that the AI "recognizes" that it's created art is by applying criteria given though training (again, error minimization). But that criteria didn't come from the AI, it came from the people who chose the training data.

Replace the training data with randomly generated values, and the neural network will generate garbage. The program can't discern the difference, but the person monitoring the program will notice that it can't seem to bring the error term down. (It's hard to predict what something should be when the data is random).

Quote
This is important because I believe you have an overly restrictive idea of what intelligence is, but I can't be sure until you articulate [crisply] what that idea is.
I think you can safely believe I've got an overly restrictive idea of what intelligence is. wink

I've already given a key criteria for intelligence. That's not a negative definition, but it's also not a complete definition.

Intelligence isn't a single thing. I'll go back to the chess analogy. At one point, people thought only intelligent things could play chess. There are plenty of elements of the game that require intelligence. But it turned out that playing chess could be encoded into an algorithm. But having computers that could play chess did little to advance AI, other than make computer scientists realize the problem was much more difficult than they thought.

The ability for a tool to "create" useful outputs is a credit not to the tool, but the developer and the data. You're no doubt familiar with Musikalisches Würfelspiel, where music is composed with a lookup table and dice. Would you say the people rolling dice "created" music, and give co-authorship to the lookup tables? (Actually, it's very much like BiaB, but that's a different tangent).

I've been following AI for long enough remember SHRDLU, Shakey the Robot, and ELIZA. Remember when Prolog was going to take over the world? Makes me feel old.

AI has come a long way since then, but no amount of semantics is going to convince me that neural networks are intelligent. There's just too much that's invisibly supplied by people.

-- David Cuny
My virtual singer development blog

Vocal control, you say. Never heard of it. Is that some kind of ProTools thing?