AI ALIGNMENT FORUM
AF

Logical InductionLogical Uncertainty
Personal Blog

3

The set of Logical Inductors is not Convex

by Scott Garrabrant
27th Sep 2016
1 min read
4

3

Logical InductionLogical Uncertainty
Personal Blog
The set of Logical Inductors is not Convex
1Sam Eisenstat
0Abram Demski
0Vanessa Kosoy
New Comment
3 comments, sorted by
top scoring
Click to highlight new comments since: Today at 8:41 PM
[-]Sam Eisenstat9y10

Nicely done. I should have spent more time thinking about liar sentences; you really can do a lot with them.

Follow-up question - is the set of limits of logical inductors convex? (Your proof also makes me curious as to whether the set of "expanded limits" {liminfn→∞Pn(ϕn)}¯¯¯ϕ efficiently computable is convex, though that question is a bit less nice than the other).

Reply
[-]Abram Demski8y00

This uses logical inductors of distinctly different strengths. I wonder if there's some kind of convexity result for logical inductors which can see each other? Suppose traders in Pn have access to P′n and vice versa. Or perhaps just assume that the markets cannot be arbitrarily exploited by such traders. Then, are linear combinations also logical inductors?

Reply
[-]Vanessa Kosoy8y00

This is somewhat related to what I wrote about here. If you consider only what I call convex gamblers/traders and fix some weighting ("prior") over the gamblers then there is a natural convex set of dominant forecasters (for each history, it is the set of minima of some convex function on ΔOω.)

Reply
Moderation Log
Curated and popular this week
3Comments

Sam Eisenstat asked the following interesting question: Given two logical inductors over the same deductive process, is every (rational) convex combination of them also a logical inductor? Surprisingly, the answer is no! Here is my counterexample.

We construct two logical inductors over PA, {Pn}, and {P′n}.

Let {Pn} be any logical inductor over PA.

We consider an infinite sequence of sentences ϕn:⇔"Pn(ϕn)<1/2".

Let Pn(ϕn) be computable in f(n) time.

We construct {P′n} as in the paper, but instead of using all traders computable in time polynomial in n, we use all traders computable in time polynomial in f(n) time. Since this also includes all polynomial time traders, {P′n} is a logical inductor.

However, since the truth value of ϕn is computable in f(n) time, if the difference between P′n(ϕn) and the indicator of ϕn did not converge to 0, a trader running in time polynomial in f(n) can easily exploit {P′n}. Thus,

Pn(ϕn)≂n1/2,

and

P′n(ϕn)≂nThmPA(ϕn).

Now, consider the market

{P′′n=Pn+P′n2}.

Observe that

P′′n(ϕn)≂n1+2⋅ThmPA(ϕn)4,

so

|P′′n(ϕn)−1/2|≂n1/4.

Now, consider the trader who exploits {P′′n} by repeatedly either buying a share of ϕn when the price is near 3/4, or selling a share when the price is near 1/4, waiting for that sentence to be resolved, and then repeating. Eventually, in each cycle, this trader will make roughly 1/4 of a share, because eventually the price will always be close enough to either 1/4 or 3/4, and all shares that this trader buys will be true, and all shares that this trader sells will be false.

Thus {P′′n} is not a logical inductor.