More Writing: Using Artificial Intelligence: Trying and Backup Plans, "AI Power," and Capability Exchanges/Purchases and Zero-Knowledge Proofs

 

AI is really big these days, that's for sure.  I wrote in another post about how AI could be seen as "using a natural-language-like programming language," with queries, commands, and axioms.


There are more insights about how to use AI too.  If you want to be strategic and powerful--whether you are being sure to be safe and ethical or not--you might try to do something I would never do, which is, to try to use AI to manipulate the world.  For example, you might issue an AI command like this:  "Try to manipulate France into allowing my side business selling doughnuts to become a big hit in France."  If the AI, perhaps after computing for 2-3 months, fails at this task, you could have a "back up plan," where you either try something else that might be hard, or, you could do something more punitive and sure-fire, as a sort of "tantrum" or "retaliation" for your first-choice plan not working.  That way, if you have the ability to "hint" to a target what you want to do, if you are then "disobeyed," you would use your power to do something that is more within your capability for sure, such as "short lots of shares of French companies in a way that damages the French economy" (note, that might be illegal in the USA due to market manipulation laws...I'm not necessarily assuming that the AI user is ethical or law-abiding, I'm talking about what's possible.)


The other big question I wanted to mention today is, "What happens when two AI-enabled powers go head to head?  Who wins?"  I would argue that it has a lot to do with provably effective capabilities.  In particular, there are probably relatively easy ways, hopefully also safe ways, to *test* a particular "AI routine" and see that it works.  The computer science idea of zero-knowledge proofs could be used to prove that one party has a particular capability that can be effective and powerful, without revealing at all how the capability works.  Such interactive proofs could be used to vet ideas for exchanges and purchases of AI capabilities.  In theory, if this caught on and people were made aware of it, this could help to "save" programming jobs and the value of computer science education; instead of coding websites or whatever for humans, individual programmers could work to develop "AI capabilities," and then use those capabilities as sold source code to sell capabilities to different automated AI algorithms.  In particular, someone could build a capability, and sell it literally to an automatic computer program...no human interaction or deal-making needed.  (One challenge would be to maintain exclusivity of the capability by having only one AI power receive it...but that could be handled either by not requiring exclusivity, or, using some sort of AI binding "hex" that punishes the AI developer for revealing the capability to anyone else.)


So that's a possibility for the future of AI.

Comments

Post a Comment

Popular posts from this blog

Math Update: My P != NP Proof Might Be Correct!!!!! -- (update, no it's not, see the note at the beginning and the updates near the bottom)

EDIT:  No, this proof is not correct.  See the updates at the bottom for details on why (based on Dr. Zelinsky's comment).  Fortunately, my GLL proof and MSR proofs still appear to be correct. It could be that Joshua Zelinsky's comment actually helps me to complete the proof. The GLL proof appears to be correct, AFAIK. The thing is, the fact that we can prove "PA is consistent --> P != NP" within PA and the fact that we can prove "PA is consistent" as a theorem of ZFC just shows that P != NP is a theorem of PA.  That isn't a contradiction, it's just another step in the proof. I think the striking idea is, P != NP is a theorem of PA, but not of ZFC.  That is what I worked to prove.  AFAIK it is possible for a proposition to be a theorem of PA but not of ZFC.  In particular, PA can be proved consistent from ZFC, so we would think it s possible and even likely that there are some non-theorems of ZFC that are theorems of PA. So this is great!  It l...
Read more

Hello, Texas - A Quick Monopoly Lesson

  So today, I thought I saw some people running around blurting code gestures at me.  One of the codes seemed to say, "Riot for a[buse] v[ictim] time -M".  Oh, M?  M as in Monolith, or M as in Monopoly fan who doesn't get it? Here, Houstonian code-talkers...let me teach you how to play Monopoly, since you are so hungry for free knowledge:  - Make deals with other players that benefit you.  - Offer deals to difficult players that are somewhat beneficial to them, knowing that they might reject.  - Watch which deals other players reject, and identify what *deal types* they are not likely to accept from you or anyone else.  - Be very very careful about your own deals--and behavior--lest you get figured out by your opponents and backed into a corner, with good, beneficial deals being made without you by cunning adversaries who know your every move. There's a poker lesson in there, too, but I don't want to get too educational right now.
Read more

Flash Non-Fiction: James Bond and Masculinity

  This was going to be a longer essay, but I decided to just blurt out my point rather than trying to write well. Basically, I think James Bond movies are good, and the songs are high-quality too, but it sends semi-dangerous messages to men in particular about what it means to be a man.  The Bond character is intransigent, aggressive, violent, and absolutely sure of himself in all cases.  He is portrayed as a big winner--able to date and have sex with beautiful women, almost impossible to kill, and lethal and elite at both armed and unarmed combat. I'm concerned that some men, including my estranged father ASW, have had their sense of morality warped by Bond movies.  They think that this dramatically presented way to be masculine is the only way.  I think that's unhealthy, and when people get the not-so-original idea to "try to be more Bond-like" in their personal lives, that leads to catastrophic conflicts with and major potential for deep pain for people that ...
Read more