Why not string theory? Because enough is enough

>I've got a theory that string theory is more of a sociological phenomena than real science.

Yes. I have a friend with a PhD in physics, and he says that a generation of string theory people basically just shouted everyone else down in a rather obnoxious way. Other critics (e.g. Lee Smolin) have said similar things, albeit slightly more politely.

At some point it stopped being science and became more about careers, reputations, funding, and paper mills. Which, given the way that academia works at the moment, became self-sustaining.

Some interesting math has fallen out of string theory, but it should never have been allowed to hold back other approaches to quantum gravity to the extent that it has.

Considering the amount of time and paper involved, there's a huge wall of maybe-perhaps-if but very little solid physics to show for the effort.

Incidentally, that blog is a real find - some of the clearest explanations of hard concepts I've seen anywhere.

Pretty sure what you will end up with in the best case is a numerical integrator of the solar system as represented by a neural net. I also expect to have pretty crappy stability, but its an interesting question what sort of time step method it would approximate.

> Every mathematical system has things which are true but can't be discovered from first principals.

This is not true, see Gödel's completeness theorem [1], which states "... that a deductive system of first-order predicate calculus is 'complete' in the sense that no additional inference rules are required to prove all the logically valid formulas." It basically means that every "true" statements are provable with a good enough deductive system.

What you possibly think about is Gödel's first incompleteness theorem [2] which is quite a different statement. It states that in "most" formal systems (that are complex enough to be able to describe natural numbers) there is always a statement A that both A and "not A" are not provable. It basically means that these statements can't be "logically valid", which means that there exist models of the formal system where A is true and models where "not A" is true.

[1] https://en.wikipedia.org/wiki/G%C3%B6del%27s_completeness_th... [2] https://en.wikipedia.org/wiki/G%C3%B6del%27s_incompleteness_...

You misunderstand, first: it is not the case that new observations about the world forced the changes. Instead, it is mostly observations about the internal properties of string theory which has forced modifications.

After it was proposed as a self-consistent theory of quantum gravity it was realized it wasn't self-consisten because it had ghosts, which was then fixed by adding 22 dimensions. But this is a problem because that's not how many dimensions the world has so they solved it by compactification that folds all the extra dimensions tiny so they don't bother us. But then it turns out that there is no particular reason to fold just so so that we get the world we live in. And now they are trying to find a way to make it so its natural that the folding happens just so that we get the world we live in.

Secondly, and most importantly, while it is true that some models get revised whenever there is more data, those are poor models. Good models agree with all the data and don't need modification. (The best models agree with most of the data, but sometimes don't so that the scientist gets a mystery to solve!) Realistically this stage takes a while to happen, and proponents of string theory argue that they right now have a models that is going to be great one day, but right now needs a bit of tinkering. What critics say is that string theory has been in the "tinkering" stage for around 50 years now and maybe we should try other ideas a bit more.

Finally, I don't think Sabine likes being refereed to as a "he" and rather have that pronoun used to describe the father of her children rather than herself.

So we have instance based learning which basically is about approximating this, and support vector machines which abstract the idea into representing the space of decisions represented by all the data points via a kernel which is a mapping function. The problem I see which deep learning seems to be "winning on" is that the space of instances that you have does not either abstract a meaningful theory of the distribution of these points (therefor predicting outside of this space) or describe things exterior to that space - but that exist, or might exist. A scientific theory is considered valid if it predicts things that have not been seen yet (and you can then test that, hence the post today about "string theory : enough"), I think IBL and SVMs can't represent the unseen, I used to work using something called inductive logic programming (I used a tool called Progol that Steven Muggleton made, it was good!) and I thought that that did produce such insights, sometimes, but it turned out that often these were actually me due to fiddling. Some play projects I did recently made me think that conv. nets were doing the same sort of things, but it was /is harder for me to catch and show this (and as I say, I ended up thinking that I'd often fooled myself with ILP, even though it was very useful).

The old skool difficulty I have with deep nets is that I used to think in terms of structural risk minimization vs empirical risk minimisation when dealing with overfitting. The idea was that if you had the right size of information store in your learned system and it was experimentally producing results that showed it was an effective predictor you could say that it was generalizing properly. Deep nets seem to me to have all the information storing capability of the domains they address and I worry....

But I am shocked, shocked, by how well they seem to work.