2 ms·
> My initial response to you came from you picking an extremely controversial field of research rather than examples of widely-held or supported beliefs counter
by gwern 11y ago
> My initial response to you came from you picking an extremely controversial field of research rather than examples of widely-held or supported beliefs countered with newer, better evidence that wouldn't be dismissed immediately by readers as psi topic would.
And why is Psi so immediately dismissed? It's not because people have knockdown arguments about how sealed radioactive number generators can be easily biased to produce long runs or anything like that. It's dismissed based on priors: psi is just too absurd, contradicts just too much successful physics. And I agree and don't believe in psi in the slightest. Extraordinary claims really do require extraordinary evidence.
- nickpsecurity 11y agoThe priors are important. They can result in a dismissal but I'd put priors in context. Was the source unreliable? Was there a specific pattern in the research that led to unreliability before and exists now? Dismissing something because others were screw-ups or unethical would've been a good way to miss out on all kinds of scientific advances that amounted to rather simple, but new, experiments or data. I'm all about priors being used to prioritize what verifiers focus on or to justify strong controls in event malice or error rates are a concern. Psi is a good example of both. Nonetheless, it's not "extraordinary evidence" I'm looking for as that's almost always an ill-defined term used for dismissal by groups being challenged. I'm just looking for reliable, repeatable evidence that fits the data better. Let me give you an example that may or may not fit but feels like it does. A common attack in computers was stack smashing where incoming data flows toward stack pointer that controls execution, overwrites it, and causes rogue execution. Discovered and rediscovered in several ways over decades with 90's forgetting what was learned. Let's pretend that didn't exist & focus on mainstream (90's onward) approach. Their theory was that it was inherent and to be countered with either cheap tactics or expensive checking. Cat and mouse game for defence and attack followed with all believing that was inevitable and different people pushing different sides. That's a widely held belief, model, supporting claims, tests, techniques, and so on over thousands of people for decades. Now, here's how you show they're all fools without extraordinary evidence or fringe science. :) You dust off a copy of the MULTICS security evaluations from 1970's to illustrate how it reduced risk of stack overflows by making data flow away from stack pointer and into new memory where available. Much better design decision & solves root issue everyone ignores. Other systems, esp hardware, tagged memory where they recognized what a stack was and had checks that turned attacks into mere error conditions. The root cause is simple to explain, the fixes are ordinary checks, the tests that they work are ordinary... everything here is simple, ordinary, and totally defies a whole industry worth of systems programming theory and practice. The "priors"... common wisdom & sociological factors here... prevent acceptance or adoption of this with many other tactics remaining. Although a select few radicals in academia and industry did implement it with effectiveness prior data showed. That's what I'm talking about. We just do science as usual but with a different perspective, test, model, or whatever. Occasionally, one of these contradicts a lot of accepted beliefs or tests because they inherently biased themselves with narrow view or incorrect assumptions the new experiment doesn't have. So, people might re-run the data, check it, attempt to reject it, confirm it, whatever. And I still haven't found the need to define the concept of extraordinary evidence in any of this: just basic science and evidence.