Re:What happened when MIT students used A.I. to make pizza
by olooney on Thursday September 13, 2018 @05:34PM
(#17982841)
Google did something similar with cookies[1], basically treated it like a non-gradient optimization problem[2], using survey results from human participants to estimate the empirical distribution function to be optimized.
From the article it makes it seems the MIT students used some kind of generative model to generate new, random recipes and didn't incorporate any human feedback, making it less interesting. After all, we know that RNNs can learn and model the much more complex syntactical structure of code[3], so generating recipes isn't exactly cutting edge.
[1]: https://www.blog.google/technology/research/makings-smart-co...
[2]: https://en.wikipedia.org/wiki/Derivative-free_optimization
From the article it makes it seems the MIT students used some kind of generative model to generate new, random recipes and didn't incorporate any human feedback, making it less interesting. After all, we know that RNNs can learn and model the much more complex syntactical structure of code[3], so generating recipes isn't exactly cutting edge.
[3]: http://karpathy.github.io/2015/05/21/rnn-effectiveness/