>> Even though it's not practical, it's an approach that previously wasn't possible so at least some cheering for it is valid.

The approach taken by DeepCode -- generate-and-test -- was always possible and is a staple in inductive program synthesis. DeepCode's version is a particularly crude and backwards-looking attempt at implementing that approach. It's no advance of any kind and it only looks like it is because DeepMind's preprint reporting on it carefully ignores most of what has gone before in inductive program synthesis, the better to show off its new system. It's a scandalous tactic that is a staple of DeepMind's publications and I'm very sorry that I can't find anything positive to say about it.

Also, I'm not talking about AGI. At this point, AGI is a fantasy. What we can hope for is to create systems that we can control well, for example to make them learn what we want them to learn and not to overfit to their dataset. In this, also, there seems to be a regression, recently, as the tendency is to train systems end-to-end while trying to remove "the human in the loop" (which is, of course, impossible).

> At this point, AGI is a fantasy.

I thought you did, because fully automated problem solving is AGI, but I don't claim that we are anywhere near to reach it.

I agree it's probably far away, but it's a fantasy I like to imagine and I don't think it's an impossible goal. Don't be sorry, I appreciate a fair criticism.

> What we can hope for is to create systems that we can control well, for example to make them learn what we want them to learn and not to overfit to their dataset.

I wish you strong nerves, because it will have AGI written all over it :) Again, this is not a claim, just a prediction.

I don't think that having control over machine learning systems must necessarily mean AGI. For example, suppose I train an image classifier to learn to identify dogs in images. And suppose I find that my classifier has learned to label all images that have a dog collar as "dog" and no image that doesn't have a dog collar as "dog". That means my classifier has learned an over-specialised definition of "dog" that doesn't generalise to dogs without dog collars, and at the same time an over-general definition that may accept, say, a fetish model with a dog collar as a "dog". That is not a system that performs as I want it to perform: I wanted a dog classifier but I got a dog collar classifier. What I want is to be able to control my system to learn to identify a dog as a "dog" rather than a dog collar as a "dog". That's "control" and when I achieve it my system is still not an AGI, just an image classifier that generalises well.