Finally, a paper on the philosophy of error analysis in ML

Scorzato’s Reliability and Interpretability in Science and Deep Learning is one of the few epistemology of ML papers which has a strong grasp of recent technical results in ML. It succeeds in connecting these results to a wider philosophical discussion without inventing artificial philosophical problems in ML.Scorzato starts by discussing recent technical approaches to error estimation in ML. The epistemological problem… Continue reading Finally, a paper on the philosophy of error analysis in ML

LLMs and meaning (again)

Jumbly Grindrod has a recent paper arguing that LLMs produce meaningful output. He makes an argument fielding teleosemantic theories very similar to the argument advanced by Lyre. He differs from Lyre by thinking that the success of LLMs is also evidence for the distributional hypothesis, something that is not discussed by Lyre at all. I… Continue reading LLMs and meaning (again)

Transformers as Markov processes

If one can view transformers as Markov chains, then limiting results for Markov chains apply to transformers. Cosma noted for example that “[t]here are finite-state probabilistic languages which cannot be exactly represented by finite-order Markov chains.“. There are also arguments that finite-state, finite-order Markov processes cannot model human language in principle (Debowski). For Cosma and… Continue reading Transformers as Markov processes

Rota’s rant about mathematical philosophy

Gian-Carlo Rota who, if you didn’t know it, held a professorship for mathematics and philosophy took the opportunity of a Synthese special edition to condemn mathematical philosophy before it was invented. His basic argument is that mathematics is essentially clear and philosophy is not. Therefore any aspiration for clarity is damaging to philosophy. He also… Continue reading Rota’s rant about mathematical philosophy

Do LMMs really train themselves?

Recently Holger Lyre presented his paper “Understanding AI”: Semantic Grounding in Large Language Models in our group seminar. And while I generally remain skeptical about his claims of semantic grounding (maybe the occasion for a separate post) here I want to address a misunderstanding in his paper about what he calls “self-learning”, “self-supervised learning” or… Continue reading Do LMMs really train themselves?

The illusion of generalization

Contrary to optimistic claims in ML literature, I often cannot help but think that deep neural nets are indeed overfit and do not generalize well. But of course that claim hinges on what one means by generalizing well. About this there has been considerable confusion in the more practical engineering oriented ML literature, which at… Continue reading The illusion of generalization

A study in LIME

LIME for images explanations are extremely dependent on the choice of tiling. Tiling is one inductive assumption of LIME but it is hidden from the end user. I know I’m late to the party, but here we go. I was recently asked to say something about philosophical problems in XAI and LIME is an obvious… Continue reading A study in LIME

IACAP23 recap

Before my memory fails me – I didn’t take notes – I wanted to write down whatever I remember about the IACAP conference that took place in Prague in the beginning of July. There was a certain old vs. new guard feeling pervading the whole conference which played out mainly between traditional philosophy of computation… Continue reading IACAP23 recap