Why is Physics So Difficult?

Subscribe to Max Stirner.

Benji Metha on Jaynes and Probability as Logical Inference

So if there's another scientist in his audience, he is going to have to consider the probability of the results of the successive coin flips as being dependent on some hypotheses. For example, the scientist flipping the coins may not be using a fair coin, he may be basing the results on a distribution he observes from the predictions of the audience, or he may be using some mixture of these based on another random process he has invented, say by using one method if he throws four, five or six on a die, and the fair coin if he throws a one, a two or a three. Or he may be adjusting those mixtures according to how many heads or tails he has thrown so far, or, ... the possibilities are endless. So the Bayesian has to make some assumptions. The difference between the Bayesian and the frequentist is that the Bayesian will always have some idea of what their priors are conditioned on. In this case, the set of all the possible algorithms that the coin flipper may be using are a set of measure zero in the sample space, so they're effectively unknowable and E.T. Jaynes ignores them. The frequentist has only one supposition, which is that the distribution is going to be that of the sample they have seen so far.

Subscribe to Machine Learning and AI Meetup.

George Ellis and Ard Louis on top-down causation (November 2015): 

Part II:

 Part III:

 

Part IV:

 

My thoughts. Ellis' example of a digital computer is not top-down causation. A lot of people make this mistake. The reason is just that an awful lot of work is done by human engineers to make computers act like Lego sets. This is so that they are deterministic and always do the same thing, given the same instructions. We do it that way because people like to know whether the computer is working or not, and if they weren't deterministic we could never be sure there was actually anything wrong with them. This may change in the future once people start using AI and become used to getting different results every time they do anything with their computer, but for now we make them so that they behave deterministically. And that is the only reason why we can reliably run the same program on different computers and get the same results from that program and the same input. So in fact this is only top-down causality because the engineers made certain that it was bottom-up causality! 

This mistake leads to the notion that there is some kind of determinism operating from the top down, so the subsequent discussion about possibility space and how vast it is is also based on combinatorial arguments which are ultimately from the deterministic assumption of material reductionism. The reality I live in (and I think others too, even if they say otherwise) is one in which the laws of physics are part of a possibility space we define in language, and which have an empirical basis in our ability to translate descriptions of physical processes --- which include chemical, biological, sociological, psychological and theological processes --- and to empirically decide whether or not our immediate experience is adequately described by those laws. So the possibility spaces are not predetermined, they are the intellectual landscape we sculpt together through our choices about what we believe to be true descriptions of our own direct experience, and perhaps even more importantly, how we describe those direct experience to others and how we understand the descriptions others make of their own direct experience. 

Therefore life does not have to explore a vast possibility space, rather it creates that space just by what it does and how we describe that process to each other. The same is true of the possibility space that physics explores: once we became skilled enough at making experimental apparatus we created spaces such as those where completely new phenomena like superconductors and Bose-Einstein condensates could exist. The types of minerals that exist on earth and the chemistry that produces them are not chosen from an abstract space, they are created by the geological, biological and even sociological processes that go on. So top-down causation is non-deterministic too. An example of this is when someone like Michael Levin goes into a laboratory and hooks a machine learning model to some electrodes in a plate of tissue cells; he is not part of some "Platonic dynamics" which is a blueprint for reality, he's just a random dude who needed to get some money to live and hit on a way to do it by applying for trendy funding: AI and Biotechnology and Pharmacology; and the same goes for every other scientist who has a proper job, and probably a few who don't and are busy trying to design the next generation of AI chips, quantum computers or space ship propulsion systems or something.

There may be another element of non-linguistic experience involved, but I don't know how to talk about it. I subsume all non-linguistic "mental processes" into the term "direct experience" which does a lot of work here. It occurred to me today that the widespread use of AI may force people to start taking this seriously because otherwise human thought will stagnate, but perhaps that started happening with the advent of the printing press?

See David Jaz Maiers - Compositionality via 2-algebra for some ideas about how to use computers to realize genuine top-down causality.

Subscribe to Fine Tuning.

George Ellis did an interview with Curt Jaimungal recently. He is still under the illusion of top-down determinism because he claims computers have agency and cites air-traffic control systems of all things. 

 

Subscribe to Curt Jaimungal.