abramdemski

Sub­sys­tem Alignment

Ro­bust Delegation

Embed­ded World-Models

De­cision Theory

Embed­ded Agents

A Ra­tion­al­ity Condi­tion for CDT Is That It Equal EDT (Part 2)

• abramdemski 4 Oct 2018 14:38 UTC

  3 points

  in reply to: jessica.liu.taylor’s comment on: A Ra­tion­al­ity Condi­tion for CDT Is That It Equal EDT (Part 1)

  I maybe should have cla­ri­fied that when I say CDT I’m re­fer­ring to a steel-man CDT which would use some no­tion of lo­gical caus­al­ity. I don’t think the phys­ical coun­ter­fac­tu­als are a live hy­po­thesis in our circles, but sev­eral people ad­voc­ate reas­on­ing which looks like lo­gical caus­al­ity.

  Im­ple­ment­ab­il­ity as­serts that you should think of your­self as lo­gico-caus­ally con­trolling your clone when it is a per­fect copy.

A Ra­tion­al­ity Condi­tion for CDT Is That It Equal EDT (Part 1)

• abramdemski 26 Sep 2018 7:52 UTC

  2 points

  in reply to: dranorter’s comment on: In Lo­gical Time, All Games are Iter­ated Games

  I have been think­ing a bit about evol­u­tion­ar­ily stable equi­lib­ria, now. Two things seem in­ter­est­ing (per­haps only as ana­lo­gies, not lit­eral ap­plic­a­tions of the evol­u­tion­ar­ily stable equi­lib­ria concept):

  • The mo­tiv­a­tion for evol­u­tion­ary equi­lib­ria in­volves dumb se­lec­tion, rather than ra­tional reas­on­ing. This cuts the tricky knots of re­cur­sion. It also makes the my­opic learn­ing, which only pays at­ten­tion to how well things per­form in of round, seem more reas­on­able. Per­haps there’s some­thing to be said about ra­tional learn­ing al­gorithms need­ing to cut the knots of re­cur­sion some­how, such that the evol­u­tion­ary equi­lib­rium concept holds a les­son for more re­flect­ive agents.

  • The idea of evol­u­tion­ary sta­bil­ity is in­ter­est­ing be­cause it mixes the game and the metagame to­gether a little bit: the play­ers should do what is good for them, but the res­ult­ing solu­tion should also be self-en­for­cing, which means con­sid­er­a­tion is given to how the solu­tion shapes the fu­ture dy­nam­ics of learn­ing. This seems like a ne­ces­sary fea­ture of a solu­tion.

• abramdemski 25 Sep 2018 20:10 UTC

  3 points

  in reply to: justinpombrio’s comment on: Track-Back Meditation

  Meta com­ment: are your up­votes worth 7 points?

  Seems that way. I don’t know what the ex­act for­mula is, but it is based on karma.

• abramdemski 25 Sep 2018 6:54 UTC

  3 points

  in reply to: justinpombrio’s comment on: Track-Back Meditation

  Ah, I wouldn’t have ex­pec­ted that. Good to know!

  Think­ing about your fram­ing from TMI, per­haps I’m sup­posed to put my aware­ness but not my at­ten­tion on the dis­tract­ing thoughts. The reason it is tempt­ing to do more than this is to “fully in­teg­rate the part of me that wants to not be do­ing this”—IE, put full aware­ness onto it to dia­logue with it, and de­cide what I really want to do with the full­ness of what I want right now.

  In your ex­per­i­ence, is it enough to have aware­ness on the stubbed toe, or is it ne­ces­sary to put at­ten­tion on it? You de­scribe your at­ten­tion go­ing to the toe.

• abramdemski 24 Sep 2018 18:50 UTC

  3 points

  in reply to: justinpombrio’s comment on: Track-Back Meditation

  From my past ex­per­i­ence, when I had more aware­ness I felt like I was way worse at things. I no­ticed every time I for­got some­thing, and saw lots of dumb thoughts. But it seems likely to me now that that’s just how things have al­ways been, and that was one of the few times I was aware of it. Do you say your pro­ductiv­ity was worse be­cause you ac­tu­ally got fewer things done, or just be­cause you no­ticed lots of awk­ward gaps and tan­gents that might’ve al­ways been there, pre­vi­ously un­noticed?

  I got less done.

  The Power of Now is not very gears-y. It does get a litte gears-y when it talks about con­crete prac­tices which aim to help you live in the now, but only a little. You have to fill in gears. I gen­er­ated very char­it­able in­ter­pret­a­tions as I was read­ing. it. It’s def­in­itely a soft skills book.

  It sounds like in your com­ment, you’re say­ing one of these two things:

  The com­ment you’re re­fer­ring to was mak­ing a more philo­soph­ical point that mo­ments are only mean­ing­ful in their con­nec­tion to other mo­ments. I could have made the same point by point­ing out that there is al­ways a delay from when one neuron fires to when an­other fires in re­sponse. I was push­ing against the strong Now-ish on­to­logy in the power of now. Non­ethe­less, you re­sponse is not ir­rel­ev­ant, be­cause part of what I was claim­ing was that one should do some­thing much like “check­ing in” as you de­scribe it, in or­der to see whether they were really con­scious/​aware/​at­tent­ive of what was hap­pen­ing in the pre­vi­ous mo­ment.

  I would ex­pect it to de­crease your at­ten­tion and in­crease your peri­pheral aware­ness. And more spe­cific­ally to de­crease your at­ten­tion on thoughts, and in­crease your peri­pheral aware­ness of ex­ternal senses. Does that sound right?

  I think this is part of it, but an­other part of it is that The Power of Now re­com­mends some­thing like dir­ectly fa­cing your suf­fer­ing in or­der to trans­mute it into con­scious­ness. This led me to non­lin­guist­ic­ally fo­cus on any dis­tract­ing feel­ings such as not want­ing to be do­ing what I was do­ing and want­ing to pro­cras­tin­ate in­stead. This was some­what in­ter­est­ing, but dis­rupt­ive of pro­ductiv­ity.

• abramdemski 24 Sep 2018 18:38 UTC

  33 points

  on: Real­ism about rationality

  Ra­tion­al­ity real­ism seems like a good thing to point out which might be a crux for a lot of people, but it doesn’t seem to be a crux for me.

  I don’t think there’s a true ra­tion­al­ity out there in the world, or a true de­cision the­ory out there in the world, or even a true no­tion of in­tel­li­gence out there in the world. I work on agent found­a­tions be­cause there’s still some­thing I’m con­fused about even after that, and fur­ther­more, AI safety work seems fairly hope­less while still so rad­ic­ally con­fused about the-phe­nom­ena-which-we-use-in­tel­li­gence-and-ra­tion­al­ity-and-agency-and-de­cision-the­ory-to-de­scribe. And, as you say, “from a his­tor­ical point of view I’m quite op­tim­istic about us­ing maths to de­scribe things in gen­eral”.

• abramdemski 22 Sep 2018 1:09 UTC

  7 points

  in reply to: gworley’s comment on: In Lo­gical Time, All Games are Iter­ated Games

  I agree that “func­tional time” makes sense, but some­how, I like “lo­gical time” bet­ter. It brings out the para­dox: lo­gical truth is time­less, but any lo­gical sys­tem must have a proof or­der­ing, which brings out a no­tion of time based on what fol­lows from what.

• abramdemski 21 Sep 2018 23:05 UTC

  4 points

  in reply to: justinpombrio’s comment on: Track-Back Meditation

  Since re­cently read­ing The Power of Now, which thor­oughly and vis­cer­ally de­scribes the per­spect­ive in which track-back would be harm­ful (be­cause it dis­tracts from the now), I want to elab­or­ate on this some more.

  Tho Power of Now did some­thing in­ter­est­ing to my mo­ment-to-mo­ment aware­ness, but at least in the short term, it seemed to wreck my pro­ductiv­ity. Return­ing to the track-back move­ment rather than the “now” type move­ment seems to help bring me back quite a bit.

  No mo­ment ex­ists in isol­a­tion; any­thing which you can men­tally la­bel as a mo­ment is an ex­ten­ded mo­ment. Fur­ther­more, al­though you can cul­tiv­ate what feels like heightened states of aware­ness (as in The Power of Now), the only way to verify aware­ness is by check­ing whether you per­ceive more de­tail, and re­mem­ber more ac­cur­ately. To sim­ul­tan­eously be the ob­server and the ob­served is an il­lu­sion; you are al­ways only ob­serving a pre­vi­ous self, even if the delay is very slight.

  So, check­ing whether you can re­mem­ber what happened in your head over the past few seconds is a check of aware­ness. Fur­ther­more, Sh­in­zen Young sug­gests that aware­ness after-the-fact is in some sense as good for your prac­tice, and more com­pat­ible with in­tel­lec­tual work. Fur­ther­more, I find that pay­ing at­ten­tion to the train of thought/​feel­ing which led to dis­trac­tion is really help­ful for main­tain­ing fo­cus and mo­tiv­a­tion.

  I’m phras­ing this as in op­pos­i­tion to “be­ing in the now”, but, I’m not sure to what ex­tent I really mean that. I do think I learned things from The Power of Now; and, I’m not deeply ex­per­i­enced in either style.

• abramdemski 21 Sep 2018 20:24 UTC

  2 points

  in reply to: migueltorrescosta’s comment on: In Lo­gical Time, All Games are Iter­ated Games

  If you have no memory, how can you learn? I re­cog­nize that you can draw a formal dis­tinc­tion, al­low­ing learn­ing without al­low­ing the strategies be­ing learned to de­pend on the pre­vi­ous games. But, you are still al­low­ing the agent it­self to de­pend on the pre­vi­ous games, which means that “learn­ing” meth­ods wich bake in more strategy will per­form bet­ter. For ex­ample, a learn­ing method could learn to al­ways go straight in a game of chicken by check­ing to see whether go­ing straight causes the other player to learn to swerve. IE, it doesn’t seem like a prin­cipled dis­tinc­tion.

  Fur­ther­more, I don’t see the mo­tiv­a­tion for try­ing to do well in a single-shot game via it­er­ated play. What kind of situ­ation is it try­ing to model? This is dis­cussed ex­tens­ively in the pa­per I men­tioned in the post, “If multi-agent learn­ing is the an­swer, what is the ques­tion?”

In Lo­gical Time, All Games are Iter­ated Games

• abramdemski 19 Sep 2018 2:48 UTC

  2 points

  in reply to: justinpombrio’s comment on: Track-Back Meditation

  Yeah, one might think I’m go­ing against the grain, re­com­mend­ing some­thing that more ex­per­i­enced med­it­at­ors warn against. On the other hand (and imho), we could take it as a warn­ing against or­din­ary dis­trac­ted­ness and or­din­ary un­mind­ful in­volve­ment in thoughts. Fo­cus­ing in­ten­tion­ally on one spe­cific men­tal mo­tion is very dif­fer­ent.

  Of course, that’s for the goal of the book, which is about mind­ful­ness med­it­a­tion, which in­volves sta­bil­iz­ing your at­ten­tion and strength­en­ing your peri­pheral aware­ness.

  The goal of mind­ful­ness might be in­ter­preted in dif­fer­ent ways (and I haven’t read that book yet), but en­der the in­ter­pret­a­tion of de­fu­sion, I think there’s noth­ing par­tic­u­larly harm­ful about the track-back ex­er­cise. It is pos­sible that it can get you caught up in his­tory and there­fore fused with the thoughts, but it is also pos­sible that look­ing at the his­tory helps put thoughts at a little dis­tance.

  For ex­ample, someone us­ing mind­ful­ness to deal with ci­gar­ette crav­ings (try­ing to quit) is sup­posed to pay mind­ful at­ten­tion to the crav­ing, and “ride the wave” un­til the crav­ing is over. It is pos­sible that track­ing back to what gave rise to the crav­ing helps con­tex­tu­al­ize it and thus put it at a re­move (“I was stressed just now, and then I star­ted hav­ing the crav­ing”). It is also pos­sible that it takes you away from mo­ment-to-mo­ment pres­ence, and the next thing you know, you find your­self reach­ing for a ci­gar­ette. I don’t know for sure.

  If your goal is re­lated to debi­as­ing, though, I think it’s a pretty good form of mind­ful­ness: the ques­tion “why did I have that thought?” is closely re­lated to epi­stemic hy­giene. “Why am I think­ing this plan is bad? Ah, I star­ted out be­ing an­noyed at El­len for her bad driv­ing, and then she men­tioned this plan. But, her driv­ing is un­re­lated to this plan...”

• abramdemski 12 Sep 2018 21:13 UTC

  5 points

  on: Good Cit­izen­ship Is Out Of Date

  This is sim­ilar to, but slightly dif­fer­ent from, the story in Bowl­ing Alone. (Dis­claimer: I haven’t read Bowl­ing Alone, only had sev­eral dis­cus­sions about it with someone who has.)

  One very in­ter­est­ing ques­tion is: why were good cit­izen­ship norms at their peak in the 1930s-1950s?

  Ac­cord­ing to Bowl­ing Alone the an­swer is that there was a massive club-form­a­tion burst from the late 1800s to the early 1900s. These clubs cre­ated the strong so­cial fab­ric which al­lowed trust in the so­ci­ety over­all to be high.

  Why was there a burst of club cre­ation? I don’t know.

Track-Back Meditation