Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Hey, welcome fellows. Um, we're in the second discussion on chapter 9 for cohort 7. So, we could go to any part of chapter 9 or jump in wherever you prefer. What about the provocative first sentence? Ultimately, the models described in this book are only useful if they can answer scientific questions. Uh yeah, I liked uh this is uh old chapter because of its uh uh well because of the way that it um inverts everything for the sake of uh uh applying this to science. I think that that's it's interesting the cognizance of it all to realize like uh maybe what's been done so far is helpful as a model but that doesn't really um uh that's not the same as let's say a scientific um I guess result and so but to turn it around and say it is quite impressive if you can uh do it backwards so that if you can if I understood this correctly If you can suppose that there's this model in a subject or maybe a group of subjects and then you can um given that supposition then maybe figure out what the parameters should be um or what the what the results should be and then work back to say okay well if the if these are the results then what are the initial beliefs you know between different um participants in a population and then if you can do that then it's interesting oh that you have a a you know potentially a scientific explanation of how these different uh individuals in a population differ and why and to be able to show the helpfulness of the model and as explanatory and then you can correlate that with other information. So that's how what I took from this. Yeah, great points. It has a lot to do with what makes a scientific account like any let alone a satisfying or useful one. And what is the relationship between empirical observables like the brute fact of the data points and inferred latent states like causal attributions. Mhm. or unobserved features which are derived from empirical data but derived from empirical data does a lot of work because you could derive in in so many different directions from empirical data. You could say that this art piece was derived from the weather data and there's so many degrees of freedom with what weather data you took and and what art piece you resulted in that somebody else might not be able to look at it and know that you use weather data at all. Right now, I'm interpreting this statement to mean uh I'm interpreting answer scientific questions to mean make testable predictions. That may not be everyone's interpretation, but I always thought models are only useful if they can make testable predictions. Yeah, that's a great point that's related to um like falsificationism and the idea that we create secondary hypotheses that could then be either assessed to be true or false under the logic like if what you learned from the answer didn't give any derivative product that could be deemed to be more or less valid. It was kind of a dead end answer because it doesn't it doesn't give any uh consequences. And then testable predictions is like the empiricist way of framing it and that can even be framed and staying within the actontology like answering scientific questions means reducing uncertainty about the scientific question. Oh, I like that. Now, alternatively, um, pointing at your the first part of what you just said, it's still useful if we're moved forward in our investigations, even if it doesn't have explicitly testable predictions. If we're still developing the field of inquiry, that's still valuable. And and uh when you say testable predictions, I think that that comes at it from both ends that are necessary in science because you want to have things related to data that's observed, you know. So if it's if it if there's no data involved and there's no observations involved, then that's a funky science. But then also you want to have a theory. Uh so you want to have a a model that is um meaningfully breaking down or decomposing or structuring um these variables or questions or ideas and so that you…