[Nasrudith]

Why should facts exert a slow pull? In addition to any unknown constraints with neural mechanisms an overly high learning rate has been noted with neural networks as resembling schizophrenia in some ways. Plus whatvever physiological constraints limit how fast minds could change.

I suspect it effectively functions as minor anti-mindhacking measure as bizzare and dumb as it sounds. It would protect against adversarial input.

If there was no mental inertia then "false facts" which check out to all verification measures could prove quite dangerous as input which outweighs all past known could be easily exploited by bad actors to change what they "know" and exploit it from there.

[wizzwizz4]

An Untrollable Mathematician, Illustrated: https://www.lesswrong.com/posts/CvKnhXTu9BPcdKE4W/an-untroll...

To summarise: for reasons, it's possible for somebody to endlessly drive up and down a computationally-limited agent's credence estimate for mathematical conjecture X (so long as the agent hasn't found a proof yet) just by stating implications of the conjecture, if the agent is trying to use ideal Bayesian reasoning. It's possible to make this "trolling" impossible, by basically turning your "mental inertia" up to 11 – but that proves it's possible to be untrollable.