[pjrule]

This is a fantastic idea, and I think I’ll definitely be applying! Have you considered offering two tiers of grants? Most AI side projects don’t necessarily require ~$30k in resources, but I’m sure many people, including myself, could benefit from ~$3k in GCE credit. Opening up the same total resources to more people would allow for a greater number of moonshots, which are what this grant seems to be about. :)

[danicgross]

Great idea, and something we’re considering for future batches. We may even do it this batch, depending on the final mix.

[maaaats]

I could use the cash and some of the credits, but most of it wouldn't be used. But maybe I'm one of few doing AI projects not involving a deep neural net? Hehe.

[freehunter]

>maybe I'm one of few doing AI projects not involving a deep neural net

You're not. While progress in machine learning is astounding and is going to do great things for humanity, it is kind of frustrating that "AI" to many people these days means neural nets, deep learning, high-end GPUs, and huge datasets. Not every problem needs machine learning as a solution, and there are still some of us solving hard problems with classical AI and old fashioned data analysis. But it's not trendy right now.

[solarkraft]

What kinds of problems are best solved by AI, but not deep learning/neural nets?

[freehunter]

Natural language processing is a big area where classical AI is heavily used and machine learning is lagging behind. ML is terrible at teaching computers to read and/or produce human writing. Compare what the company Automated Insights is doing for news articles versus neural networks writing books and movie scripts (with humorous results).

Anything with a known and well-defined set of rules can also be programmed more efficiently with decision trees/Markov chain than through machine learning. There's no need to use tons of data to train a model to do something over time when all it needs is the rule book.

Machine learning is basically advanced pattern matching, neural nets often require you to brute force the problem with tons of data. There are plenty of places where you're not matching patterns and you don't have tons of data but the computer still needs to make a decision after analyzing the data that is available.