Currently focused on AI policy and governance. Sophomore.
Please reach out at julius.olavarria@gmail.com.
On substack @juliusolavarria
Julius Olavarria đ¸
Karma: 24
TLDR: Everyone is clueless, even experts, and p1 relies on the counterfactual that we would chose actions /â general EV in an ideally sorted causal chain.
Would like some clarification, since for me preference in one case before making a decision does not move the consequences or expected value of the decision. If you were to use expert advice to c-prefer A over B, both agents are still trapped in an infinitely causal universe. C-prefer = sorting out causality, which is impossible if we admit p3 (âcosmos-wide consequencesâ). The premises rely on each other to grant the conclusion.
âEvidence need not contain the reasoning that produced it,â or âtestimony can transmit information without transmitting the underlying reasoning.â You would need to believe a doctor who is telling you treatment A is better than treatment B based on evidence. Your framing is that, if the doctor read 1000 pages on each treatment, the deciding agent would inherit the doctorâs experience as evidence.
I am wondering what work this is doing because I cannot quite formulate why I think this is more important than meets the eye. I think this example is unique. Perhaps the whole thing is that we are expounding moral cluelessness, and for this case, existing (being healthy, using doctorâs advice) is not in itself good (asymmetry argument). I say this just to propose assumptions that may be hidden.
Trusting a financial advisor to do A vs. B given you do not know the market is not moral cluelessness, and even if trusting the advisor led to a successful returns, we cannot turn back time to see what would have happened if you went with B (given things can only happen once, turning back time maybe may generate a new reality /â causal path).
I guess, if a (more) idealized agent (e.g., an expert) says A is better than B and you have reasons to believe this then we are closing in perhaps on better EV decisions. But I think EV in terms of cluelessness directly equals a fully correct model where one would know exactly what would happen from an action until the âend of time.â
Whatâs most interesting is that waiting for more information generates better decisions. Again, what I am most concerned about here is that no decision would be ideal in a messy future. So trusting testimony or waiting are not converging on a âbetterâ finite ideal, but rather remain blanket unawareness. So normatively we would prefer a decision process where this is not the case (p1).
I may not have a good understanding of this, but premise 1 is an EV claim, which rests on measuring outcomes, since perhaps EV calculations lead to better decisions.
Second, even if the A vs. B preference is justified, unawareness means there are countless other Xâs you havenât found yet.
If we challenge a premise I assume this must lead to the conclusion not obtaining. Your argument if successful here no longer forces the conclusion, it does not mean that the conclusion is false (a successful premise attack would make the conclusion false?).
Finally, it seems your claim also depends on a prediction model somehow breaking a messy, indeterminate universe (p3). Since this is impossible this is a heuristic. P1 references the idealized self; unawareness wins if we cannot make EV calculations /â know everything (relevant).
Would love clarification.
Hey Charlie, youâre a legend for this. One thing I will add is for easy video editing is to use CapCut and then auto-generate subtitles. You can export to 1080p and that gives the quality people are used to seeing for your types of videos.
Not going to lie, a few vids were grainy and almost all of them didnât fit the frame. This is an easy fix if you can spare $10 a month to edit on CapCut.
On your part about ingenuity, I immediately thought about AIs making up new languages to communicateâpossibly to hide their work from human supervisors, but possibly to invent more efficient ways to reference recognized patterns. Are they adding anything new when they do this? And a similar question: Does Newton add anything when he recognizes and proves gravity via math?
If AIs mirror human reasoning, and intelligence is nothing but recognizing causal structures and predicting the next best result, adding things over and above causality is only connecting to what actually exists. New experiences add to causal web, so experiencing novelty will be a very important part to creating superintelligence.
Letâs do it.
To this I would say adding more information to the model doesnât = better decision. P1 means an idealized agent would chose A over B given they know everything (e.g., full causal universe). Normatively we would choose this over flawed/âlimited decision making.
What makes a model âopen to correctionâ if all models are flawed?
I think a strong challenge to a critique would show that the conclusion is not forced based on a flaw with the premise as presented by DiGiovanni.
On 5) this seems pragmatic, not a critique to cluelessness.
Correct me if you think I am misled on any of these.