Video gamers solve a biological puzzle [video] | Hacker News Reader

Video gamers solve a biological puzzle [video]

aeon.co

35 pointsOsiris305 comments

A much harder (for researchers to devise) but a lot more beneficial form of gamification is 'implicit gamification'.

Why do you have to move gamers away from their favorite Battlefield, DoTA, CSGO, LoL, Starcraft, or whatever games, in order extract value out of the huge amount of decision-making going on?

- Gamers play their favorite games.

- The decisions they make during the gameplay get translated into decisions about non-trivial aspects of scientific work. (This is the hard part).

- Those decisions lead to scientific progress.

I'd love to find out if anyone is working on this, or what is the state of the art in this area.

>The decisions they make during the gameplay get translated into decisions about non-trivial aspects of scientific work. (This is the hard part).

It's a nice dream but I suspect figuring out the mapping from game to scientific problem would take exponential time (ie be totally impractical) if you're talking about unmodified games.

There was once a movement to try to do this gasification for many scientific problems. The problem is that turning arbitrary problems into fun games is really hard, and only works in some domains.

However just now AI is getting good enough to beat human players at most games. Perhaps most of this work could soon be automated. And even the games that aren't very fun for humans could be automated. The computer doesn't care about fun.

Deep Learning systems are great at learning one thing really well, given enough data. Humans are able to learn novel things with limited data because we can apply all the other data and learning we've previously done. And that's the barrier to this sort of thing, imho.

There's only a (relatively) small number of training examples in FoldIt, but a Deep Learning system would needs thousands or millions before it would generalize.

Or maybe there's a great new research paper that will break through that limit. It's such an exciting time for the field!

Yes but the great things about games is you can generate tons of data automatically by letting the computer play the game by itself. Deepmind's Atari thing was able to master Breakout within 2 hours of self play only, so it's not implausible.

Reply on news.ycombinator.com