Hello, can you please give me tips/pointers? How do I separate commands for each beakers? I've been using very simple words (less pink triangle, etc.) but it doesn't seem like the slime really retains the learned vocabularies. Please help and thank you.
A downloadable game
Lab Assistant is a puzzle game where you solve puzzles with your chemical slime by teaching it text commands. It runs on machine learning, so it learns language from you as you play!
Billy Peake - sound & data
Brian Handy - programming
Gwen Bassett - art
James Fiacco - AI programming
Jessica Kernan - design
Liz Fiacco - animation/scripting
Based on this paper: Learning Language Games Through Interaction
Click download now to get access to the following files:
- Update!Jan 31, 2018
Log in with itch.io to leave a comment.
Since it is machine learning, there is always a chance that it could learn something weird under the hood that can impact it's efficiency. It also won't learn it perfectly the first time it tries something (you may have to correct it on what pink means over several rounds.
It can identify beakers by left-most and right-most (it doesn't have vocab for it, but that is the mechanism it selects by), but maybe that helps you describe what you want it to do.
Also, hypothetically, you can have it practice commands. It only moves you ahead when the puzzle is complete, but you can give it any arbitrary command, and tell it good job for getting it right! (ex. the puzzle suggests it turn the beaker pink, but you tell it to turn yellow, then praise it when it is correct)
Beyond that, fewer characters helps improve loading times, and consistency is PARAMOUNT. Even capitalization/not looks different to it.
OK, but let's say that there are 2 beakers and I want to practice pink. Do I just need to type "pink" or "pink pink"? Will it make a difference? Thanks for replying.
I'm sorry if I don't have a cut and dry answer for you! It learns the best it can from what you type, so it really depends on being consistent with whatever diction and syntax you end up using.
My guess is you probably won't get great results if you repeat words in that way (because then it has to infer left or right based only on the "space" character between the words, which probably is not the only place you use spaces).
For practice, as far as I know it won't matter -what- you practice, as long as you know the result you want to practice and it 100% succeeds at it. But you can certainly experiment with it, this is new technology, and a new kind of gameplay! We aren't certain of the optimal strategies either! XD
OK, thank you! I'll do my best to come up with new strategies! So far I've gotten to three beakers and I use (amount)(shape)(color) and repeat that 3 times for each beaker. Most of the time it works, but sometimes the slime makes weird guesses, so I might have to tweak it. Thanks for making this game, I am very interested in "teaching" AIs!
is there a Linux version planned?
Please publish a macOS version
Sorry Matthew, we don't have a computer to compile that on! :c
( Unity does cross-compiling though? You're using some funky libraries, or your own compiled code? )
Oh dang! I don't think I've tried cross compiling before. I am used to the ye olde days. XD
But yeah, the machine learning stuff was written separate from Unity, I doubt it would translate simply to Mac. x___x It would be great to continue to develop this tech beyond this super short project and make it it more accessible one day.