G Gordon Worley III(G Gordon Worley III)

• G Gordon Worley III 26 May 2020 3:04 UTC

  LW: 2 AF: 1

  AF

  on: Think­ing About Filtered Ev­i­dence Is (Very!) Hard

  As­sump­tion 3. A listener is said to have min­i­mally con­sis­tent be­liefs if each propo­si­tion X has a nega­tion X*, and P(X)+P(X*)≤1.

  One thing that’s in­ter­est­ing to me is that this is as­sump­tion is fre­quently not satis­fied in real life due to un­der­speci­fi­ca­tion, e.g. P(I’m happy) + P(I’m not happy) ≥ 1 be­cause “happy” may be un­der­speci­fied. I can’t think of a re­ally strong min­i­mal ex­am­ple, but I feel like this pops up a lot of dis­cus­sions on com­plex is­sues where a di­alec­tic de­vel­ops be­cause nei­ther the­sis nor an­tithe­sis cap­ture ev­ery­thing and so both are un­der­speci­fied in ways that make their naive union ex­ceed the available prob­a­bil­ity mass.

• G Gordon Worley III 26 May 2020 1:14 UTC

  8 points

  in reply to: G Gordon Worley III's comment on: Bak­ing is Not a Ritual

  To say a lit­tle more, pro­gram­ming is the thing that res­onates for me most strongly with the above de­scrip­tion of bak­ing as not rit­ual. I no­tice it be­cause there are lots of peo­ple who learn to pro­gram by rit­ual. They do well enough to get jobs and be­come my cowork­ers, but then they also write code full of cruft be­cause they are pro­gram­ming by rit­ual rather than through deep un­der­stand­ing.

  The clas­sic ex­am­ples of this that come to my mind are things like peo­ple think­ing the iter­a­tor vari­able on a for loop must be named i or else it won’t work, or strug­gling with sed be­cause they don’t re­al­ize they can pick the sep­a­ra­tor char­ac­ter on each in­vo­ca­tion, or think­ing they need [tech­nol­ogy X] be­cause they heard suc­cess­ful busi­nesses use [tech­nol­ogy X].

• G Gordon Worley III 26 May 2020 1:03 UTC

  9 points

  on: Bak­ing is Not a Ritual

  I like it. FWIW, I find this is ex­actly the same line that must be crossed to be­come an ex­pert at many things. Some ex­am­ples that come to mind, there are prob­a­bly more I just don’t have enough ex­pe­rience with to be able to be sure they are ex­am­ples:

  • programming

  • mathematics

  • writing

  Things that seem like anti-ex­am­ples, i.e. things that are not in this cat­e­gory with bak­ing:

  • rock climb­ing (al­though there is a lot of skill in­volved, you don’t have to un­der­stand the gears to suc­ceed, you can learn to in­tu­itively make effi­cient mo­tions that move you up with­out un­der­stand­ing too much about why or how that hap­pens)

  • med­i­tat­ing (al­though lots of peo­ple get into the gears, plenty of peo­ple suc­cess­fully med­i­tate ei­ther with­out a deep un­der­stand­ing of what they’re do­ing or pos­si­bly ac­tively mis­taken the­o­ries that work any­way)

  Again, I’m sure there are more, and I think for all of these there’s plenty of room to ar­gue on the mar­gins.

  If I had to guess at what the di­vider is it’s some­thing like how ro­bust the on­tolog­i­cal ab­strac­tions around the topic are at lev­els that mat­ters for the task. The more ro­bust, the less like bak­ing, the more frag­ile and leaky, the more like bak­ing.

• G Gordon Worley III 25 May 2020 3:42 UTC

  4 points

  in reply to: Chris_Leong's comment on: “It’s Okay”, In­struc­tions, Fo­cus­ing, Ex­pe­rienc­ing and Frames

  Seems like a type er­ror to me, like say­ing cry­ing is an emo­tional state rather than a thing that hap­pens in re­sponse to emo­tions or other things, or like say­ing find­ing out my friend lied to me is an emo­tion rather than a thing that causes emo­tions. Okay­ness is an un­der­stand­ing of just that which is, and emo­tions are an ex­pres­sion of that but are also not okay­ness it­self.

• G Gordon Worley III 22 May 2020 16:58 UTC

  17 points

  on: Get It Done Now

  Keep a record of tasks you’ve de­cided to do.

  So I re­ally liked GTD when I read it back in the mid 00s, and it’s stuck with me in deep ways. In par­tic­u­lar, this in­sight, but I wouldn’t de­scribe it this way. I’d say some­thing like

  use a sys­tem.

  I’ve phrased it this way be­cause there’s more to it than re­mem­ber­ing tasks to do. To me the whole core of the book is that build­ing sys­tems that sup­port us set us free. When you have a sys­tem you can trust to help you do what­ever it is that’s im­port to you, you can re­lax and get bet­ter perfor­mance at the same time be­cause you have a sys­tem you trust and it does a bet­ter job than your brain could on its own.

  Fif­teen plus years of sys­tem­atic use of sys­tems changes a per­son, so my “sys­tems” don’t look like they did when I first read GTD. In the be­gin­ning I had too much sys­tem be­cause I needed ex­tra sys­tem in or­der to be able to trust it. I didn’t know quite where the line was be­tween me and the sys­tem and which could han­dle what best. But over time it’s evolved into some­thing that works quite well. I don’t for­get to do any­thing im­por­tant, and my mind is not of­ten busy wor­ry­ing about the un­finished stuff.

  So I’ve gone from spe­cial­ized tools that were highly cus­tomiz­able to ba­si­cally just a combo of Gmail, Google Keep, Google Cal­en­dar, and Pocket. I dump stuff in, it gets done, it goes out, and I reg­u­larly re­pro­cess and re­view stuff (af­ter do­ing it enough times, I mor­phed from sched­uled re­views to con­tin­u­ous re­view). It works for me and the work I have, but some­thing else might work for some­one else. There’s lots of lit­tle de­tails to get­ting it right, and I think each per­son has to dis­cover a lot of what works for them­selves be­cause we’re all slightly differ­ently shaped. Ex­am­ples of other peo­ple’s sys­tems are great for in­spira­tion, though!

  And in­ter­est­ingly, I find it re­lates a lot to my sched­ule. My sched­ule is just one more sys­tem de­signed to sup­port my life and help make sure I do the things I in­tended with­out hav­ing to make choices or re­mem­ber ev­ery­thing all the time. I long re­sented hav­ing a sched­ule, right up un­til I re­al­ized my sched­ule could sup­port me rather than force me to do things, and that shift in mind­set to­wards a sup­port sys­tem to help rather than a co­er­cive out­side im­po­si­tion made all the differ­ence to in­creas­ing my hap­piness and suc­cess at shap­ing the world.

• G Gordon Worley III 22 May 2020 16:38 UTC

  9 points

  on: Reflec­tive Complaints

  Noth­ing is quite so an­noy­ing as see­ing an­other per­son do the thing you wish you your­self would not do.

• G Gordon Worley III 19 May 2020 15:45 UTC

  4 points

  in reply to: romeostevensit's comment on: The Mechanis­tic and Nor­ma­tive Struc­ture of Agency

  I usu­ally just call these “free vari­ables” by anal­ogy to that term’s us­age in math­e­mat­ics/​statis­tics, but “aether” seems good too since, if noth­ing else, when I say “free vari­able” to peo­ple who don’t know the jar­gon they just sort of roll over it whereas they’d have a much harder time do­ing that if I said “aether”.

• G Gordon Worley III 19 May 2020 15:42 UTC

  2 points

  in reply to: Charlie Steiner's comment on: The Mechanis­tic and Nor­ma­tive Struc­ture of Agency

  I might give this a read, but based on the ab­stract I am con­cerned that “has a per­spec­tive” is go­ing to be one of those prop­er­ties that’s so ob­vi­ous that its pres­ence can be left to hu­man judg­ment, but that nonethe­less con­tains all the com­plex­ity of the the­ory.

  I read it more as point­ing to­wards some­thing like em­bed­ded agency.

• G Gordon Worley III 18 May 2020 16:51 UTC

  LW: 4 AF: 2

  AF

  on: Con­jec­ture Workshop

  Last sum­mer when I was at the EA Ho­tel for TAISU I got the most value out of do­ing some­thing similar. I’d host a ses­sion to “work­shop” an idea I had, and it was roughly 20 min­utes of set­ting it up and 40 min­utes of back and forth with peo­ple point­ing things out, stat­ing ob­jec­tions, ask­ing for clar­ifi­ca­tions, etc.. It was less struc­tured than your ap­proach, and I quite like this idea be­cause it cre­ates a level of safety my ap­proach did not be­cause it effec­tively bans the kinds of crit­i­cism (at least for the course of the con­jec­ture work­shop) that peo­ple some­times jump to that also shut down fruit­ful idea de­vel­op­ment.

  Thanks for shar­ing!

• G Gordon Worley III 18 May 2020 16:42 UTC

  2 points

  on: “It’s Okay”, In­struc­tions, Fo­cus­ing, Ex­pe­rienc­ing and Frames

  I have some ex­pe­rience with “okay­ness”, so maybe I can say a few use­ful things.

  It’s hard to ex­plain what it’s like to be okay­ness, which is I think a bit more ac­cu­rate way of ex­press­ing the idea of “it’s okay” or “ev­ery­thing is okay”. I think it’s so hard be­cause when we try we run up against all the way words cre­ate sep­a­ra­tion be­tween this and not that, and sep­a­ra­tion, cat­e­gories, and on­tol­ogy fun­da­men­tally sep­a­rate you from okay­ness be­cause you have to make a judge­ment about what is this and what is that and mak­ing that judge­ment re­quires some fun­da­men­tal men­tal mo­tion of mea­sur­ing how much some­thing is some­thing, but that’s an­ti­thet­i­cal to what okay­ness is like.

  Okay­ness (I of­ten use the word “perfec­tion” as my poin­ter, al­though some­times it ap­pears to me as “enough­ness”, and oth­ers have used “nat­u­ral­ness” or “Bud­dha na­ture” or some other term) ex­ists prior to that mea­sure­ment, prior to feed­back, prior to our self-aware­ness.

  You’re right that okay­ness is not a thing, and its thing­ness seems to be an af­ter-the-fact at­tempt by our minds to la­bel this non-dual ex­pe­rience. It’s not an emo­tional state, al­though there are of­ten emo­tions felt while be­ing it, nor is it re­ally mak­ing any episte­mol­ogy claims in it­self be­cause while be­ing it the mechanisms of episte­mol­ogy can’t func­tion, yet strangely this doesn’t mat­ter be­cause you can only be what is when there is no map, so you have no need of episte­mol­ogy be­cause there is no map to be kept cor­re­lated with the ter­ri­tory, there is just ter­ri­tory.

The Mechanis­tic and Nor­ma­tive Struc­ture of Agency

• G Gordon Worley III 13 May 2020 3:22 UTC

  4 points

  in reply to: johnswentworth's comment on: Pro­ject Pro­posal: Gears of Aging

  I was go­ing to come here to say to look at SENS. I feel like they do have an idea of a “blueprint” for solv­ing ag­ing and are work­ing on ex­e­cut­ing on it, and they are from my out­side per­spec­tive work­ing in a way similar to MIRI: they have a prob­lem to solve, some ideas about how to do it, and some re­search agen­das they are ex­e­cut­ing on to move them to­wards the an­swer.

  Ra­di­a­tion out from SENS is a con­stel­la­tion of folks work­ing on this stuff. I’m not sure, though, that these folks also lack a blueprint, only that they work in field where they can’t eas­ily pub­li­cly claim to have one.

• G Gordon Worley III 12 May 2020 2:44 UTC

  LW: 2 AF: 1

  AF

  in reply to: Donald Hobson's comment on: Com­pet­i­tive safety via gra­dated curricula

  I agree. I’m gen­er­ally okay with the or­der (or­a­cles do seem marginally safer than agents, for ex­am­ple, and more re­stric­tions should gen­er­ally be safer than less), but also think the marginal amount of ad­di­tional safety doesn’t mat­ter much when you con­sider the to­tal ab­solute risk. Just to make up some num­bers, I think of it like choos­ing be­tween op­tions that are 99.6%, 99.7%, 99.8%, and 99.9% likely to re­sult in dis­aster. I mean of course I’ll pick the one with a 0.4% chance of suc­cess, but I’d much rather do some­thing rad­i­cally differ­ent that is or­ders of mag­ni­tude safer.

• G Gordon Worley III 12 May 2020 2:03 UTC

  4 points

  in reply to: Ben Pace's comment on: Stan­ford En­cy­clo­pe­dia of Philos­o­phy on AI ethics and superintelligence

  I also think it’s fair to have low ex­pec­ta­tions here. Although I gen­er­ally like SEP, I also have ex­pe­rienced Gell-Mann mo­ments with it enough times that now I think of it differ­ently.

  It’s not re­ally like a reg­u­lar en­cy­clo­pe­dia writ­ten by anony­mous au­thors with a “view from nowhere”. Each ar­ti­cle has an au­thor, those au­thors are al­lowed to have clear au­tho­rial bias, the whole thing has lots of se­lec­tion bias (ar­ti­cles only ex­ist be­cause some­one was will­ing to write them and they do count as pub­li­ca­tions though they are strangely in­fluen­tial and not at the same time be­cause they are read by lots of peo­ple but most of those peo­ple avoid cit­ing them and in­stead cite the work refer­enced by the SEP ar­ti­cle based on my anec­do­tal data), and as a re­sult you get things like, for ex­am­ple, ar­ti­cles writ­ten by peo­ple only be­cause they are against some po­si­tion not for it but no pro­po­nent of the po­si­tion had writ­ten the ar­ti­cle and those ar­ti­cles of­ten fail to pass the ITT.

  Com­bine this with the already low ex­pec­ta­tions around writ­ing about AI safety top­ics in gen­eral and it makes it nicely sur­pris­ing that this one turned out pretty good.

• G Gordon Worley III 10 May 2020 0:46 UTC

  5 points

  on: Is there any sci­en­tific ev­i­dence for benefits of med­i­ta­tion?

  There’s noth­ing nec­es­sar­ily wrong with what you’re look­ing for here, but I think it’s worth point­ing out that benefits or gains from med­i­ta­tion are an­ti­thet­i­cal to how many ded­i­cated med­i­ta­tors think of their prac­tice, and in fact progress is even­tu­ally ham­pered by the very idea of progress (or so many tra­di­tions claim).

• G Gordon Worley III 8 May 2020 1:42 UTC

  2 points

  on: Med­i­ta­tion: the screen-and-watcher model of the hu­man mind, and how to use it

  A book I like to recom­mend to peo­ple in­ter­ested in get­ting started in med­i­ta­tion is “A Path With Heart” by Jack Korn­field. It’s writ­ten by an au­thor who is very de­cided not a ra­tio­nal­ist and it’s filled with lots of refer­ences to su­per­nat­u­ral things, but it’s also a very kind and gen­tle in­tro­duc­tion to med­i­ta­tion and wider prac­tice of the way. If you think of the su­per­nat­u­ral stuff as metaphors rather than claims about phys­i­cal re­al­ity, I think it can be quite helpful and teaches a lot of tech­niques and gives some good mo­ti­va­tions for why and how they are use­ful.

• G Gordon Worley III 8 May 2020 1:22 UTC

  2 points

  on: Topolog­i­cal meta­physics: re­lat­ing point-set topol­ogy and lo­cale theory

  Since you are aiming to­wards philos­o­phy with this one, I’ll share some­thing about my in­tu­itions around empti­ness (as op­posed to form, in Bud­dhist Mad­hya­maka philos­o­phy) as they re­late to open sets in topol­ogy.

  In my mind it has been fruit­ful to think of empti­ness like open­ness and re­late the two, speci­fi­cally think­ing of empti­ness as de­scribing the same as­pect of re­al­ity that make “open” a good in­tu­itive la­bel for “open sets”. This has helped me un­der­stand what is pointed at by “empti­ness” by un­der­stand­ing it as “open­ness” and reusing my topol­ogy in­tu­itions as a ground­ing point.

  I’m not sure I can be more pre­cise, so I’ll have to leave it at that and hope it’s helpful.

• G Gordon Worley III 8 May 2020 1:12 UTC

  2 points

  on: Stop say­ing wrong things

  Lots has been said within these servers on find­ing ways to say new ex­cit­ing cor­rect things. Or cor­rect­ing things many think are right, but didn’t know were ac­tu­ally wrong. This is a re­lentlessly op­ti­mistic per­spec­tive. I have got­ten grand mileage by merely try­ing to hold back from say­ing or act­ing on in­for­ma­tion or ideas I know are wrong, as op­posed to wor­ry­ing about think­ing bet­ter, or not mak­ing mis­takes, or be­ing more ra­tio­nal when I am con­cerned with the truth. This other part is the meaty part, the in­tense part—the strug­gle to stop mouthing the words, to stop my­self from will­ingly burn­ing more money or time.

  I think this per­spec­tive is ap­peal­ing and use­ful ad­vice, but only up to a point. At some point you get diminish­ing re­turns from ceas­ing to punch your­self in the face, as it were, and you find you can’t do the things that mat­ter to you with­out stick­ing your neck out and risk­ing be­ing wrong by try­ing to say some­thing new that might be cor­rect.

  I find noth­ing dis­agree­able in your post other than what I read as a sen­ti­ment that maybe just stop say­ing wrong things and you can im­prove. Some­times you’ve already stopped do­ing that so much that you have to start say­ing some wrong things.

  (For what it’s worth, my com­ment ba­si­cally comes down to say­ing don’t for­get about the law of equal and op­po­site ad­vice.)

• G Gordon Worley III 6 May 2020 18:09 UTC

  LW: 4 AF: 2

  AF

  on: How should po­ten­tial AI al­ign­ment re­searchers gauge whether the field is right for them?

  In my mind it’s some­thing like you need:

  • strong in­ter­est in solv­ing AI safety

  • be­ing okay with break­ing new ground and hav­ing to figure out what “right” means

  • strong math­e­mat­i­cal rea­son­ing skills

  • de­cent com­mu­ni­ca­tion skills (you can less rely on strong ex­ist­ing pub­li­ca­tion norms and may have to get more cre­ative to con­vey your ideas than in other fields)

  • the courage and care to work on some­thing where the stakes are high and if you get it wrong things could go very badly

  I think peo­ple tend to em­pha­size the tech­ni­cal skills the most, and I’m sure other an­swers will offer more spe­cific sug­ges­tions there, but I also think there’s an im­port as­pect of hav­ing the right mind­set for this kind of work such that a per­son with the right tech­ni­cal skills might not make much progress on AI safety with­out these other “soft” skills.

• G Gordon Worley III 6 May 2020 18:02 UTC

  LW: 7 AF: 3

  AF

  in reply to: adamShimi's comment on: How should po­ten­tial AI al­ign­ment re­searchers gauge whether the field is right for them?

  Am I the right “kind” of re­searcher for work­ing in AI Safety? Here, my main in­tu­ition is that the field needs more “the­ory-builders” than “prob­lem-solvers”, to take the archetypes of Gower’s Two Cul­tures of Math­e­mat­ics. By that I mean that AI Safety has not yet cristal­lize into a field where the main ap­proaches and ques­tions are well un­der­stood and known. Al­most ev­ery re­searcher has a differ­ent per­spec­tive on what is fun­da­men­tal in the field. There­fore, the most use­ful works will be the ones that clar­ify, de­con­fuse and char­ac­ter­ize the fun­da­men­tal ques­tions and prob­lems in the field.

  To add on to this, it also means it’s go­ing to be some­what hard to know if you’re right kind of re­searcher or not be­cause the feed­back cy­cle is long and you may be do­ing good work but it’s work that will take months or years to come to­gether in a way that can be eas­ily eval­u­ated by oth­ers.

  This doesn’t mean it all looks max­i­mally like this. This is less of an is­sue with, say, safety re­search fo­cused on ma­chine learn­ing than safety re­search fo­cused on the­o­ret­i­cal AI sys­tems we don’t know how to build yet or safety re­search fo­cused on turn­ing ideas about what safety looks like into some­thing math­e­mat­i­cal pre­cise enough to build.

  Thus a corol­lary of this an­swer might be some­thing like “you might be the right kind of re­searcher only if you’re okay with long (multi-year) feed­back cy­cles”.