Google did something similar with cookies[1], basically treated it like a non-gradient optimization problem[2], using survey results from human participants to

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. [1]: https://www.blog.google/technology/research/makings-smart-cookie/ https://www.blog.google/technology/research/makings-smart-co... [2]: https://en.wikipedia.org/wiki/Derivative-free_optimization 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/ http://karpathy.github.io/2015/05/21/rnn-effectiveness/