provability

3 年前議論

I like this question. I've always wanted to be able to compute, somehow, the "partial derivative of x with respect to y," where x is a proposition and y is an argument or a piece of evidence. It seems to me as though we're closer than ever before. The models are still fairly opaque, but I'm hoping the interpretability research will succeed and allow such introspection!

provability

4 年前議論

I understand your objections and in fact I mostly agree. So I should qualify some of my prior statements. They should be understood in the context (my context when I began this line of thought) of anarcho-capitalist values as expressed in David D. Friedman's book /The Machinery of Freedom/. (By the way, to anyone who read this 1973 book, Uber and Lyft were inevitable; the book was prescient enough that it may still contain gems worth mining.) My goal was to make professional law enforcement less important.

I am not quite still an anarcho-capitalist, as after 2004 I began to see a lot of truth in Marx and other anti-capitalist writers. So I'm in a healthy tension that enables me to embrace paradoxes and even hypocrisy. (Anyone not a hypocrite, I suspect either has less than an admirable set of values or is in prison or dead or swiftly heading toward one of those states.)

Bottom line, my hunch and hope is that some of us will see that if our goal is safety, we need not always give up freedom for it. There are other ways to achieve some of the goals we've entrusted to the State.

Yes, I am aware that with alarmingly increasing frequency the State can compel us to divulge our encrypted evidence and the keys to it. Perhaps there would be more pushback against this if such evidence were to play a greater role. Right now, people tend to say "I have nothing to hide." If everyone had a lot to hide, already recorded and encrypted by default, then I think most of them would push back against 4th amendment violations (in the US; and equivalent protections in other jurisdictions).

I am aware that the stupid and the evil will mess things up. I don't know what to do about that. I suspect they'll always be with us. The latter we might reduce in number through greater accountability, through more evidence. The former, I have no idea; our society doesn't really quite have natural selection.

I should walk back "24/7" or whatever I said. I was a bit hyperbolic. I think certain types of situations known for being contentious should be recorded routinely by both parties, by default, maybe. That should be normalized, maybe. For instance, in any situation wherein at least one party is required to carry liability insurance, the insurers could compel recording -- maybe.

Also, and I am VERY FAR FROM CERTAIN about this, but POSSIBLY even in some instances of sexual encounters, at least where there is not a fully executed (signed, maybe notarized) written agreement, i.e. enthusiastic consent -- possibly, if the tech were sound enough (which should be our goal) these too should be recorded, as a matter of course -- MAYBE.

provability

4 年前議論

I feel like your statement reduces to "people are not affected by evidence."

There is a category of people for whom that is true, but is there not also a category of people whose beliefs and actions are affected by evidence?

That may be the case now, or maybe it only was in the past? What will we do when ALL true evidence is indistinguishable from fake evidence?

(By the way, I think you mean "provenance," and as I've edited my post and bio blurb to explain, I too am against methods that disclose the identity of the witness. Rather, my methods would help multiple witnesses to an atrocity [such as police firing live ammo at nonviolent protestors] to corroborate one another's accounts before identifying themselves. They could possibly even effectively discredit an official, faked account without identifying themselves.)

I mean, at least in the US, trial lawyers bill by the hour, which, as I see it, is already a conflict of interest on a meta level. This system is pretty inefficient. What if we could prevent most "he said, she said" situations (and, probably, billable hours) by making recording of all one's experiences the norm, while also safeguarding the privacy of those recordings until such time as a dispute (such as a court case) arises? (This latter, especially in the context of doctor-patient relationships, was the original root, circa 2004, of my line of thought. The idea then was "universal surveillance by private individuals of their own lives, encrypted by default, with proof of existence, optional decryption, and sharing.")

provability

4 年前議論

I know, right?

provability

4 年前議論

I agree! I've put some thought into how to keep the power decentralized -- and ideally to make it more so than the status quo. I think the techniques are compatible with that. I don't like the emphasis of Truepic et all on identifying which individual or which device recorded the evidence. I mean, yes, many in the West would like to know that a video clip is just as a specific, named, respected photojournalist filmed it, but imagine what would happen under a despot who doesn't respect press freedom.

provability

4 年前議論

Great! Please do email me.

provability

4 年前議論

To reduce the adverse effects of fake evidence (eventually called deepfake), off and on since 2004, and especially since ten years ago, I've been thinking up, NOT DETECTION methods, but prospective, pro-active methods to instrument a data stream, to increase the odds that video, audio, or other time-series data (including e.g. scientific experimental data) eventually can be proven to have been created from real life (and not faked) at the time & place claimed.

This is not just "proof of existence" by inserting hashes of content into a Merkle tree, like Haber & Stornetta implemented in 1990 and like we talked about on the Cypherpunks email list in the early 1990s, although of course that would be part of a complete system.

Nor is it relying on just the metadata that e.g. TruePic uses. I have been developing additional methods to asymptotically approach provability that the content depicted actually happened in real life. Think undetectable watermarks by multiple observers (courts want to see multiple witnesses) but it goes beyond that.

Relevant to this are efforts by CAI, C2PA, and companies like TruePic. I am hoping my methods are different enough, with little enough overlap, that we can realize them without licensing their IP. My goal is a more decentralized system that protects the privacy of witnesses.

My contact info is in my profile. I'd love to discuss this with anyone. I'd even be relieved if someone would explain that my ideas are not useful, or already have been done, or even would have unintended negative consequences for society. (It's hard to stay current in the relevant literature on signal processing, steganography, cryptography, and zero-knowledge proof, plus politics and game theory.)

I would even considering simply donating my trade secrets (such as they are) to some person or organization who impresses me as sincere and likely to move them forward, to avoid having to sit on them hands-off (again) during my tenure at (yet another) company that is about to offer me a full-time job today to work on something unrelated. While some of my ideas "toward authenticatability" may have been ahead of their time when I first wrote them up ten years ago, they probably aren't anymore, so the window is (at best) closing now.

コメント