3 ms·

Most of the neuroscience researchers I know aren't chuckling about it, aside from in a "This is incredible" way. Frankly I find academics working in neuroscienc

Most of the neuroscience researchers I know aren't chuckling about it, aside from in a "This is incredible" way. Frankly I find academics working in neuroscience tend to be way, way less dismissive of LLMs. Here's an interesting and related paper from neuroscientists and ML researchers: https://www.nature.com/articles/s42003-022-03036-1 https://www.nature.com/articles/s42003-022-03036-1 It's worth reading through and it's also worth reading some of the papers it cites and the papers that cite it.

mjburgess 3y ago
Yes, there's an absolutely alarming (perhaps desperate) theoretical grasp towards AI as being able to provide an explanatory paradigm for neuroscience. As it is trivial to show, this fails -- and produces egregious pseudoscience such as *that article under a Nature brand!* Correlating blood flow in the brain to matrix-multiplication coefficients in digitised text-embedding is some extreme nadir of AI pseudoscience. They should next correlate it with patterns in sprays of mist in the air; with enough sprays, they'll find a whole new theoretical paradigm for neuroscience.
- lucubratory 3y ago
  Person 1: "All the scientists who actually work in the field must be chuckling at this foolishness!" Person 2: "Here is evidence that scientists in the field do not even remotely share that viewpoint and in fact view current work as quite profound and in line with their own work." Person 3: "The scientists working in the field are clearly engaged in pseudoscience in a conspiracy going up to Nature - it is trivial to show this, with the proof being that an anonymous comment on an internet forum doesn't agree with them. QED."
  - mjburgess 3y ago
    I have delivered conference talks on the methodological errors of this area -- I also never made the claim of Person-1; nor are Person-1 and Person-2 engaged in the same dialectic. It is not a "conspiracy", as you'd know, if you had any familiarity with the methodological literature on these areas. The consensus view of methodological critics of these areas is (1) fMRI analysis is profoundly unreliable as a guide to relevant features of the brain; and (2) a significant majority of research in this area is unreproducible. Both of these have been demonstrated multiple times. Psychologists are extremely poorly trained in statistics and how to apply statistics to scientific enquiries; let alone on the mathematical modelling which goes into phrasing and building neural networks. That someone has written a paper correlating coefficients of a model they've no training to understand against fMRI results -- is par for the course in this papermill. The level of absurdity here is off the charts; but the epistemic culture around these areas of speculative science is obscene --- this is why a vast majority of their papers are unreporducibel.

mjburgess 3y ago

Yes, there's an absolutely alarming (perhaps desperate) theoretical grasp towards AI as being able to provide an explanatory paradigm for neuroscience. As it is trivial to show, this fails -- and produces egregious pseudoscience such as *that article under a Nature brand!* Correlating blood flow in the brain to matrix-multiplication coefficients in digitised text-embedding is some extreme nadir of AI pseudoscience. They should next correlate it with patterns in sprays of mist in the air; with enough sprays, they'll find a whole new theoretical paradigm for neuroscience.

lucubratory 3y ago
Person 1: "All the scientists who actually work in the field must be chuckling at this foolishness!" Person 2: "Here is evidence that scientists in the field do not even remotely share that viewpoint and in fact view current work as quite profound and in line with their own work." Person 3: "The scientists working in the field are clearly engaged in pseudoscience in a conspiracy going up to Nature - it is trivial to show this, with the proof being that an anonymous comment on an internet forum doesn't agree with them. QED."
- mjburgess 3y ago
  I have delivered conference talks on the methodological errors of this area -- I also never made the claim of Person-1; nor are Person-1 and Person-2 engaged in the same dialectic. It is not a "conspiracy", as you'd know, if you had any familiarity with the methodological literature on these areas. The consensus view of methodological critics of these areas is (1) fMRI analysis is profoundly unreliable as a guide to relevant features of the brain; and (2) a significant majority of research in this area is unreproducible. Both of these have been demonstrated multiple times. Psychologists are extremely poorly trained in statistics and how to apply statistics to scientific enquiries; let alone on the mathematical modelling which goes into phrasing and building neural networks. That someone has written a paper correlating coefficients of a model they've no training to understand against fMRI results -- is par for the course in this papermill. The level of absurdity here is off the charts; but the epistemic culture around these areas of speculative science is obscene --- this is why a vast majority of their papers are unreporducibel.