togelius

I had to make a few simplifications to spell out the differences clearly and avoid making the text infinitely long. It's true that most current gradient descent algorithms are stochastic because they are computed in batch mode, and that sophisticated evolution strategies approximate the gradient. I still think the differences are significant, in that evolution updates less often and the direction of the update is much less (if at all) dependent on the feedback.

Now, your point about to what extent this is really about neural networks is a good one. Could a network learn F=ma, even if we could not interpret it? Maybe. With the right data, represented the right way.

Use Google Scholar. Most technical researchers these days put all their papers on their webpage for free, and these days also on ArXiv.

Good point, I could have used "hacker" instead, though that's more specific to writing software. I didn't perceive "tinkerer" as a word with negative connotations when I wrote it.

Come to think of it, many outside the hacker community would perceive "hacker" more negatively than "tinkerer", because many people still think of hackers as criminal. At least that's my perception. I guess words have different values in different contexts.

Oh yes, I'm a terrible hacker, I rarely write code anymore. An unfortunate byproduct of academia is that you get "promoted out of the job", and all the hands-on work is done by your lab members.

I didn't mean to imply that tinkering was inferior to research - the whole premise was just to tease out how they're different, with different audiences, as you say. Interestingly, the discussion here has been dominated by people who think that I look down on them. People who've discussed it in other fora have not read the post that way.

"a systematic enterprise that builds and organizes knowledge" That is exactly my point. Doing your scholarship is doing your part to organize knowledge. Systematically building knowledge means building on others, and knowing what you build on.

So the quote (which you don't attribute to anyone other than "wiki" - a citation here would be useful) really proves my point. If you're doing not doing your scholarship, you are not doing science, or research, you are tinkering.

Again, there's nothing wrong with tinkering.

I'm the author, and I'm somewhat amused by the negative tone of many of the comments. Seems there are plenty of people with resentments towards research here ;)

This is all true. However, the large publishers (IEEE, ACM, Springer) these days all allow you to self-archive on your own webpage and typically also on ArXiv. So you can (and should) make your papers freely available. I have done that since the start of my career. It's just a matter of looking.

He mentions NEAT, which is good, but not any of the work on playing Mario, much less any work on evolving neural nets to play Mario. There's quite a few papers on the topic, not just mine.

Note that I don't say that SethBling is wrong. Seen as tinkering, this is perfectly fine, and he did a great job with the video. No hard feelings. I just use it as an example of how it is not research, because of the lack of scholarship.

Like basically all researchers in CS these days, I keep all my papers freely downloadable from my webpage. Easily findable via Google Scholar, or just a regular Google search, for anyone who cares to look.

I am an AI researcher and faculty member at a large and famous university. I probably know less math than that Reddit poster. Math is important if you are specifically interested in the math of AI. If you are interested in inventing algorithms and solutions you mostly don't need the math.

And general video game playing :)

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

Meta-heuristics as conceptualized by many (most?) does not include selection among algorithms. Hyper-heuristics does. It's fine if some of the algorithms selected among are meta-heuristics. Yes, this is very confusing, and yes, better terminology is needed. In particular, we should probably stop using the term meta-heuristics, because it has too many incompatible meanings.

Glad you liked it! We have done some work on trying to generate complete games - an overview of it can be found in this book chapter: http://pcgbook.com/wp-content/uploads/chapter06.pdf The most similar to what you describe is probably the ASP approach to game generation.

Thanks! And wish I could do something about the blogspot UX.

Thanks, appreciated!

The difference between meta-heuristics and hyper-heuristics is that the former is a much broader concept, including such things as genetic algorithms. Hyper-heuristics is specifically about selecting among heuristics, which excludes e.g. evolutionary algorithms. Graham Kendall, one of the inventors of the concept, explains that they would have called it meta-heuristics if that name was not already taken.

The source of data is a lot of simulations we ran ourselves. All of the games can be won or lost - there's also a score measure as well, but that was not used in producing that graphic.

Oops, thanks for spotting the broken paper link - fixed now!

Correct - we are not using MCTS and the approach has nothing to do with AlphaGo.

The fitness function is pretty simple, essentially a weighted piece counter. It's true that it does not look ahead and take future states into account so might fall into a trap, but empirically it's doing very well!

My intro to AI course at NYU uses the Ms. Pac-Man vs Ghost Teams framework for all of the assignments. It is indeed a very good starter problem.

These two are both interesting and important research directions. They are indeed different but have significant overlap. That "game AI can already be easily written to win 100% of games" is plain wrong. It's true for some games, especially if you cheat by giving the AI access to information the human player does not have. But even in those cases it is often impossible. We are very far from playing at high human level in for example Go or StarCraft.

So how exactly would we go about to make sure that no-one does research on AI? Which particular kinds of research would we ban? And how would we enforce it?

It's hard to create believable AI in a first-person shooter. See: http://nn.cs.utexas.edu/?botprize2012

It's actually really hard to create well-performing AI in a first person shooter, unless you give it explicit access to the internal state of the game. See: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=612949...

I would say that progress on general AI can solve a lot of the problems we are facing. To take an obvious and down-to-earth example, let's say we "solve" logistics in the sense that we can transport things wherever we want, timely and cheaply. That would help us with a lot of the problems we are facing, such as hunger. And for solving logistics we essentially need the same skills we need as for solving game-playing.

The new Mario AI contest is at http://www.marioai.org