the relative importance of existing insights extracted from neuroscience has been consistently decaying over time
I don’t think this is a very good indicator. The resolution of an fMRI could easily increase by two orders of magnitude in 10 years given enough federal funding (currently 100 μm but nowhere near the whole brain), bringing us much closer to modelling all neural activity in an entire human brain and using it as a layer for a foundation model. Humanity’s fMRI technology would be totally incapable of doing that, for decades, until the moment that those numbers (magnetic imaging resolution) are pumped high enough to record brain activity at the fundamental level.
While on the topic of funding, I also think that the AI industry is large enough that the expected return of latching neuroscience to it is quite high. So we should also anticipate more galaxy-brain persuasion attempts coming from neuroscience.
I agree that more detailed measurements of brains could be technologically feasible over the coming decades and could give rise to a different kind of insight that is more directly useful for AI (I don’t normally imagine this coming from fMRI progress, but I don’t know much about the area).
That said, I think “what is the current trend” is still an important indicator.
And the paper talks explicitly about the influence of past progress in neuroscience, both its past influence on AI and the possible future influence. So I think it’s particularly relevant to the argument they are making.
I don’t think this is a very good indicator. The resolution of an fMRI could easily increase by two orders of magnitude in 10 years given enough federal funding (currently 100 μm but nowhere near the whole brain), bringing us much closer to modelling all neural activity in an entire human brain and using it as a layer for a foundation model. Humanity’s fMRI technology would be totally incapable of doing that, for decades, until the moment that those numbers (magnetic imaging resolution) are pumped high enough to record brain activity at the fundamental level.
While on the topic of funding, I also think that the AI industry is large enough that the expected return of latching neuroscience to it is quite high. So we should also anticipate more galaxy-brain persuasion attempts coming from neuroscience.
I agree that more detailed measurements of brains could be technologically feasible over the coming decades and could give rise to a different kind of insight that is more directly useful for AI (I don’t normally imagine this coming from fMRI progress, but I don’t know much about the area).
That said, I think “what is the current trend” is still an important indicator.
And the paper talks explicitly about the influence of past progress in neuroscience, both its past influence on AI and the possible future influence. So I think it’s particularly relevant to the argument they are making.