Hacker News Reader

Comment by dougmwne | Hacker News Reader

This feels like a bit of light speculation at the moment, but I do agree with the point and think it will have an interesting effect on collective human behavior. I like the example of AlphaGo discovering new moves that then became popular because this is an example of an AI discovery, not a another human writing the rules of engagement that other humans follow.

So to take this further, what if Imagen discovers a new perspective style that catches on in human art? What if GTP-5 popularizes a new twist on the detective procedural? What if CONGA-9 kicks off a new wave of AI synth world fusion? And what if we get to the point where our artistic trends are mostly derived from AIs Instead of humans artists? Is it still human culture?

Anyway this was all super unexpected to me. Sci-fi had me trained that AIs would be intelligent, incapable of lying, humor or art. All GTP-3 does is lie to me.

View on HN

In chess this is nothing new as humans have been using chess engines to discover new opening tactics for the last couple of decades.

Chess engines used to be at forefront of Artificial Intelligence research but now it’s not fashionable to even call them “AI” anymore - they are just “engines” - tools.

AI is only in fashion for technology once it isn't bleeding edge. Also, MuZero (the latest in similar game engines that were started from AlphaGo) can learn the game by itself without knowing the rules or having any opening books or Go data which is very different than how traditional Chess engines operate.

This "Zero-knowledge" style program, be it AlphaZero or MuZero and whatever, is a bad meme.

Adding in little bits of knowledge, such as liberties counting, ladder-reading, and other tricks, allows KataGo to absolutely crush LeelaZero in training speed and practical play in the game of Go. Its no comparison.

These days, no one studying the game of Go is seriously using LeelaZero anymore. Its an obsolete methodology compared to KataGo.

A more familiar example might be fuzz testing. You could fuzz a program by just passing the fuzzer totally random data to mutate, and it's true that this might find a crash that wouldn't be found any other way. However, most of the time you will have much better results, and achieve them much more quickly, if you pass the fuzzer something that looks like an actual expected input to the program.

Zero-knowledge ML is still interesting if your goal is to indisputably demonstrate the act of learning, just not if your goal is to make the smartest program in the least time.

> can learn the game by itself without knowing the rules

In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6.

“What are you doing?”, asked Minsky.

“I am training a randomly wired neural net to play Tic-Tac-Toe” Sussman replied.

“Why is the net wired randomly?”, asked Minsky.

“I do not want it to have any preconceptions of how to play”, Sussman said.

Minsky then shut his eyes.

“Why do you close your eyes?”, Sussman asked his teacher.

“So that the room will be empty.”

At that moment, Sussman was enlightened.

http://www.catb.org/jargon/html/koans.html

Also, https://en.wikipedia.org/wiki/TD-Gammon

> TD-Gammon is a computer backgammon program developed in 1992 by Gerald Tesauro ... It explored strategies that humans had not pursued and led to advances in the theory of correct backgammon play...

> For example, on the opening play, the conventional wisdom was that given a roll of 2-1, 4-1, or 5-1, White should move a single checker from point 6 to point 5. Known as "slotting", this technique trades the risk of a hit for the opportunity to develop an aggressive position. TD-Gammon found that the more conservative play of 24-23 was superior. Tournament players began experimenting with TD-Gammon's move, and found success. Within a few years, slotting had disappeared from tournament play

I had an AI professor that described artificial intelligence as an agent that operates independently in some environment and reasons probabilistically / mathematically to determine actions, so I made a trading bot that used some simple time series techniques to dictate decision making. She ended up telling me that it wasn't an AI because the algorithms wasn't complex enough. OK... well, then why didn't you just say an AI is really just defined by whether or not the 'agent' is using the newest, shiniest algorithms? I was a little salty about that one lol

> I made a trading bot that used some simple time series techniques to dictate decision making

They were quite right. In no sense is what you are doing AI. People have been using time series techniques to analyze and attempt to predict data since before the computer.

Alright, so when does an algorithm become AI then?

> Sci-fi had me trained that AIs would be intelligent, incapable of lying, humor or art.

You must have been reading a different set of books from me. Even Asimov's robots constrained by the three laws manage to lie.

I've mentioned this before - I'm using the melodies I've created together with my AI songwriting assistant as input for further training. This requires human judgement and I only have around 100 melodies that I've judged as good. But the training set is small, so together with "bad" melodies, this already makes a difference. I don't know if I'm the first to do this for AI art and be able to say that there is a difference (lower loss on the validation set of human melodies). Over time, it could result in style drift and increased originality and I don't see why it couldn't be done for AI-generated images as well with some more effort (because the training set is much larger).

You mention “further training” but im curious for what: Is this expanding a data set for AI or is it for you as a musician?

I meant for the AI. I do currently need to be involved creatively to create these melodies but I figure that it might not even be necessary with some more advances.

"a new perspective style" - I'd love to see 32x32 icons becoming popular again!

"AI synth world fusion" - count me in for new music!

"artistic trends are mostly derived from AIs" - Art is in the interpretation. "ceci n'est pas une pipe". while (1) { printf("this is not an infinite loop"); }

"All GTP-3 does is lie to me" - Yes. But the lies inspire creativity, because you can do better, right? Also, I'm very grateful for Katalin Karikó and the other researchers who created a beautiful mRNA "lie" to teach the immune system how to ward off a real virus. If that's not like cave paintings warning children about the dangers outside, I don't know what is.

As an artist, the influence of AI like VQGAN, artbreeder or the more recent MidJourney are having a big impact in the professional concept art and character art community, expect to see AI inspired concept art in Hollywood movies in a year or 2

> example of AlphaGo discovering new moves

Is AI, ML, or just 8 jillion Monte Carlo simulations in a fixed search, outcome, and rule space?

This is q learning based so ML.

Relevant quote:

Mr Smith: I say your society, after which we started to do your thinking for you and it became our society

It's grey goo for human culture.

Reply on news.ycombinator.com