EA Hotel Fundraiser 7: Pitch focusing on case studies with counterfactuals

We want to help promis­ing EAs max­imise their value.

There are many smart and highly mo­ti­vated peo­ple in the EA move­ment who lack visi­ble achieve­ments. As a move­ment, we are un­der­us­ing their tal­ents. By op­er­at­ing at min­i­mal costs, the EA Ho­tel is able to make high EV/​$ in­vest­ments in large num­bers of them as a form of hits-based giv­ing, al­low­ing them to move from soft­ware to AI Safety, move from part time to full time study, write origi­nal re­search, and many more pos­si­bil­ities.

Whilst the EA Ho­tel is open to peo­ple pur­su­ing work across all EA cause ar­eas, we note that ap­prox­i­mately half of our res­i­dents have been fo­cused on AI Safety/​x-risk (~45% of per­son-days spent at the EA Ho­tel to date have come from those work­ing in the AI Safety/​x-risk cause area).

The problem

At the mo­ment, it’s hard for tal­ented and mo­ti­vated EAs to leave sta­ble em­ploy­ment and work on valuable in­de­pen­dent pro­jects. Three fac­tors typ­i­cally make this difficult:

  1. Risk aver­sion: even for en­trepreneurial EAs, the fi­nan­cial bur­den of leav­ing one’s job with­out fund­ing is a fright­en­ing prospect.

  2. Lack of struc­ture: main­tain­ing mo­ti­va­tion is difficult with­out the right en­vi­ron­ment or rou­tine.

  3. Road­blocks: progress can stall with­out out­side in­put.

Our solution

  1. Re­duce the fi­nan­cial bur­den by set­ting up in Black­pool, a city with some of the low­est liv­ing costs in the UK. We can offer up to 18 as­piring EAs full ac­co­mo­da­tion and board (in­clud­ing cater­ing and house­keep­ing) for un­der £6,000 per year per per­son.

  2. Foster a com­mu­nity of like-minded peers to al­low reg­u­lar so­cial ac­tivi­ties, dis­cus­sion, phys­i­cal and men­tal health sup­port, ac­countabil­ity bud­dies, co-work­ing and friendly com­pe­ti­tion; all in an en­vi­ron­ment in which we’re iter­at­ing ways to min­imise stress and max­imise pro­duc­tivity.

  3. The same com­mu­nity of peers can provide feed­back and in­sights, fur­ther con­tacts and un­ex­pected col­lab­o­ra­tions that en­able faster progress than that achieved by the sum of its parts (this isn’t the­o­ret­i­cal—see ex­am­ples of this in the case stud­ies be­low).

In short, we are giv­ing many EAs an op­por­tu­nity to do im­por­tant work, in an en­vi­ron­ment that cataly­ses it.

It’s already working

We’ve been able to at­tract an im­pres­sive set of peo­ple, from AI Safety re­searchers, to PhDs, to highly pro­duc­tive en­trepreneurs, with an av­er­age of 4.8 years work ex­pe­rience (based on a sur­vey ear­lier this year).

We ex­pect most of the value of the EA Ho­tel will come from a hand­ful of in­di­vi­d­u­als whose ca­reer tra­jec­tory we can help change. That said, it looks like the coun­ter­fac­tual im­pact for the me­dian res­i­dent is already high, with an av­er­age of 2.2x pro­duc­tivity in­crease re­ported in a sur­vey ear­lier this year. A sum­mary of out­puts so far (see here for ex­panded list):

  • The in­cu­ba­tion of 3 EA pro­jects with po­ten­tial for scaling

  • 18 on­line course mod­ules followed

  • 2.5 on­line course mod­ules produced

  • 46 posts on LessWrong and the EA Fo­rum (with a to­tal of ~1500 karma)

  • 2 pa­pers published

  • 3 AI Safety events, 1 ra­tio­nal­ity work­shop, 2 EA re­treats hosted; and 2 EA re­treats organised

  • 2 in­tern­ships and 2 jobs earned at EA organisations

(In­puts: Money: ~£110.4k, of which ~£17.9k was con­tributed by res­i­dents
Time: ~7.6k per­son-days spent at the ho­tel.)

Based on self-re­port­ing (see square brack­ets on out­puts page), it’s es­ti­mated that over half of the value of these out­puts can be coun­ter­fac­tu­ally at­tributed to the EA Ho­tel. Other, less tan­gible value has also been re­ported by our res­i­dents. We fol­low with some case stud­ies.

Case Studies

Derek Foster

Be­fore mov­ing to Black­pool, Derek was study­ing for a Masters in Health Eco­nomics and De­ci­sion Model­ling. Out of run­way, he ini­tially came to the EA Ho­tel to work on the 2019 Global Hap­piness and Well-Be­ing Policy Re­port, which was pub­lished in Jan­uary. He sub­se­quently got a job at Re­think Pri­ori­ties.

After some time away, he has re­turned to the Ho­tel. Still with Re­think, he is cur­rently work­ing on a re­search agenda for the cre­ation of a wellbe­ing-ad­justed life-year met­ric.

How the Ho­tel has helped

It gave him some­where to stay when he had run out of money, in­creased his pro­duc­tivity and run­way, and ex­panded his net­work of po­ten­tial col­lab­o­ra­tors.

The counterfactual

Derek would have done similar work, but per­haps to a lower stan­dard. He’s not sure how he would have paid his rent for the first cou­ple months be­fore get­ting his first pay­cheque, and would not have been able to save as much. He is also con­fi­dent that he would not have at­tended EAG Lon­don 2018, where he gained a num­ber of valuable con­tacts in ad­di­tion to those he met at the Ho­tel.

Linda Linsefors

Linda is do­ing AI Safety re­search. After in­tern­ing at MIRI, she went back to her home town in Swe­den and found her­self largely un­mo­ti­vated to con­tinue her re­search alone. She went to the EA Ho­tel to con­tinue her re­search and stud­ies among like minded peo­ple.

Dur­ing her stay here Linda also ex­plored other ca­reer paths. She did some web pro­gram­ming for Matt Gold­en­berg and co-or­ganised Athena Ra­tion­al­ity Work­shop. After re­dis­cov­er­ing the joy of or­ganis­ing, Linda has started to or­ga­nize AI Safety events. So far she has done two work­shops and one un­con­fer­ence, which have all been ap­pre­ci­ated, and she will most likely do more such events in the fu­ture.

How the Ho­tel has helped

Life at the Ho­tel has made Linda much more mo­ti­vated and al­lowed her to put con­sid­er­ably more fo­cused hours into study­ing. She also man­aged to find some col­lab­o­ra­tion, which gave birth to two posts on the AI Align­ment Fo­rum.

Liv­ing in the same house as most of the RAISE team al­lowed Linda to give high qual­ity, high band­width feed­back on the AI Safety course they were de­vel­op­ing.

Linda has re­ceived a great amount of men­tal de­bug­ging help and emo­tional sup­port from other Ho­tel res­i­dents. Be­cause of this, her men­tal health has im­proved a lot which has in­creased her pro­duc­tivity and re­li­a­bil­ity. Without this up­grade she would not have had the con­fi­dence that she would be re­li­able enough to or­ganise events.

Be­ing able to use the Ho­tel as a venue for events has made Linda able to do more events for less effort. The fact that EA Ho­tel already have reg­u­lar cater­ing ser­vice (Ja­cob cooks for us) saves lots of work.

The counterfactual

She would have con­tinued her life in Swe­den, do­ing re­search and study­ing at a rate of ap­prox 30% of the cur­rent.

None of the events would have hap­pened.

Max Carpendale

Max is a re­searcher work­ing on in­ver­te­brate sen­tience and suffer­ing who was liv­ing in Van­cou­ver. He has re­cently worked with Re­think Pri­ori­ties on this sub­ject and has been suc­cess­ful in get­ting fund­ing from the EA com­mu­nity for his re­search. Max also started work­ing part-time for An­i­mal Ethics as of Septem­ber 2019. Max does re­search on wild an­i­mal suffer­ing for An­i­mal Ethics and he is cur­rently work­ing on cre­at­ing an on­line course on wild an­i­mal suffer­ing.

How the Ho­tel has helped

It ex­tended his run­way con­sid­er­ably. Max sus­pects that he would have ran out of money to pay rent if not for the Ho­tel, and re­ally isn’t sure what he would’ve done in that case. Max thinks that there is a 75% chance he wouldn’t have got his cur­rent job at An­i­mal Ethics if it wasn’t for the Ho­tel. Max does not think that he has been more pro­duc­tive at work while at the Ho­tel, but he thinks that the Ho­tel has been very helpful for per­sonal growth and his life­time im­pact as an EA.

The counterfactual

Be­ing at the Ho­tel has given him the op­por­tu­nity to travel much more eas­ily than he would be able to in Canada. He has gone to sev­eral EA con­fer­ences and re­treats that he would not oth­er­wise have been able to go to. It gave him the op­por­tu­nity to live abroad in Europe, which he thinks has benefited him greatly. He ex­pects that his net­work has ex­panded and will con­tinue to ex­pand due to the Ho­tel.

More case stud­ies can be found in Ap­pendix A.

Why we need funding

Quit­ting your job to work on a new pro­ject or study a new field pre­sents a fi­nan­cial risk—or for some a fi­nan­cial im­pos­si­bil­ity. We en­able those with the po­ten­tial to do great EA work to do it re­gard­less of cir­cum­stance.

Stay­ing at the Ho­tel is in­cred­ibly cheap as we have bought the build­ing out­right. Hospi­tal­ity costs for a res­i­dent are ~£80/​week (in­clud­ing men­tor­ing). As the cost of liv­ing in Black­pool is low, we can cover the re­main­ing costs with a stipend for grantees of £30/​week.

We offer grants of free or sub­sidised ac­com­mo­da­tion and board, and a stipend, to those ap­ply­ing to do EA work, sub­ject to pass­ing through our vet­ting pro­ce­dure (see Grant Mak­ing Policy).

Up un­til now, our de­fault has been pro­vid­ing free stays to any­one do­ing EA work that we offer a place to, and ask­ing those with salaries or sig­nifi­cant sav­ings (>24 months run­way) to vol­un­teer pay­ment of cost price for their stay. Given our ap­pli­ca­tion for char­i­ta­ble sta­tus, we have re­vised our model: ap­pli­cants re­quest the level of sub­sidy based on their needs.

Ul­ti­mately, we’d like to sus­tain the Ho­tel pri­mar­ily via pledges from suc­cess­ful alumni—it’s cheap enough that it could plau­si­bly be funded by just a hand­ful of suc­cess­ful bene­fac­tors (so far we have already had one such alum­nus who has made sig­nifi­cantly more in dona­tions to the Ho­tel than his stay cost). Un­til we reach that point, we’ll ex­per­i­ment with var­i­ous (hon­esty-based) means-tested ap­proaches, bal­anc­ing the need for at least par­tial sus­tain­abil­ity with ad­minis­tra­tive effi­ciency and more im­por­tantly a de­sire not to de­ter great can­di­dates. For ex­am­ple, we’re cur­rently tak­ing the fol­low­ing ap­proach:

  1. The first 6 weeks of stay is free for all suc­cess­ful ap­pli­cants. We have a pay-what-you-feel-it’s-worth ap­proach for this pe­riod, em­pha­sis­ing that you should pay zero if you’re un­cer­tain of the value of the Ho­tel to you, or if cost is a real con­cern to you. This is to aid in at­tract­ing tal­ent; for ex­am­ple one guest said he would have paid af­ter hav­ing ex­pe­rienced the benefits for 6 weeks but be­fore that would have un­der­es­ti­mated the value.

  2. After 6 weeks, we re­quire a manda­tory (though hon­esty-based) con­tri­bu­tion of £12/​day or more, to cover the costs of host­ing them, to be paid by any­one ei­ther earn­ing a salary, or with the means to sup­port them­selves for over 24 months (fol­low­ing 80,000 Hours’ ad­vice).

We may fur­ther up­date this model as we learn more about how it plays out in prac­tice.

This ap­proach will help cover some of our costs, but will nec­es­sar­ily still leave a short­fall as we are es­pe­cially in­ter­ested in fund­ing work that wouldn’t hap­pen oth­er­wise (i.e. work by those not in a po­si­tion to pay for their stay).

In our cur­rent situ­a­tion, fac­tor­ing in the above, we need ~£5,000/​month in ad­di­tional fund­ing to sus­tain the Ho­tel. Note that this is an ap­prox­i­mate figure given our costs are vari­able de­pen­dent on differ­ent lev­els of guest con­tri­bu­tion, oc­cu­pancy, staffing and main­te­nance costs. Our cur­rent run­way ex­tends un­til the be­gin­ning of March 2020 (see our Fundraiser page for up to date figures).

What var­i­ous lev­els of fund­ing would provide

£15,000 (3 months fund­ing) would al­low us to run a hiring round for a Com­mu­nity & Pro­jects Man­ager, which could in turn bring in money from EA Grants/​CEA, who are in­ter­ested in pro­vid­ing fund­ing for this role con­di­tional on us find­ing an ex­cel­lent can­di­date for it.

£60,000 (12 months fund­ing) would al­low us to provide 8 slots for free stays filled with great peo­ple for the next 12 months, and 18 6-weeks’ free stay grace pe­ri­ods for new ap­pli­cants.

£100,000 would give us an in­dus­try-stan­dard run­way, and flex­i­bil­ity to provide more free stays to great peo­ple over the next 18 months.

Conclusion

Low cost EA hubs have been talked about a lot but have rarely ma­te­ri­al­ised, due to high ac­ti­va­tion costs and co­or­di­na­tion difficul­ties. We have built the com­mu­nity, shown pre­limi­nary re­sults in our first 18 months and now need out­side funds to con­tinue pro­vid­ing cheap/​free stays to great peo­ple, and bridge the gap to long-term sus­tain­abil­ity.

Our cur­rent fund­ing situ­a­tion is pre­car­i­ous, with­out fur­ther funds we’ll need to start wind­ing down and giv­ing res­i­dents no­tice in un­der 3 months. Un­der rea­son­able as­sump­tions, we could reach sus­tain­abil­ity within 3-5 years, once we have suffi­cient alumni pledg­ing sup­port. But we need your help to bridge the gap and make this hap­pen. If you like this ex­per­i­ment and want it to con­tinue, please donate here.

Pre­vi­ous posts in this se­ries:

Writ­ten by oth­ers:


Ap­pendix A: Fur­ther Case Studies

We dis­cuss ad­di­tional case stud­ies, as­sess­ing the im­pact of the Ho­tel on their work from their own per­spec­tive, and then link to ad­di­tional bios for past guests.

Fredi Bechtoldt

Fredi is cur­rently writ­ing his mas­ter’s the­sis on the De­mand­ing­ness Ob­jec­tion to eth­i­cal the­o­ries and he did an in­tern­ship at An­i­mal Ethics, work­ing on wild an­i­mal welfare as­sess­ment meth­ods.

How the Ho­tel has helped

He heard about the in­tern­ship po­si­tion through one of his many new ac­quain­tances he made at the Ho­tel. The Ho­tel en­abled him fi­nan­cially to do the un­paid in­tern­ship.

The at­mo­sphere of the Ho­tel keeps him mo­ti­vated and in­creases his pro­duc­tivity.

The counterfactual

Fredi would still write his mas­ter’s the­sis, but would’ve not been able to start the in­tern­ship. He would work part time as a coach to sup­port him­self, like he did for the last ten years. Maybe, he would have drifted out of the EA com­mu­nity.

Kris Gulati

Kris has an un­der­grad­u­ate de­gree in Poli­tics and a two-year MSc in Econo­met­rics and Eco­nomics from The Univer­sity of Not­ting­ham. He is cur­rently ap­ply­ing to PhD pro­grammes in Eco­nomics, whilst learn­ing the Math­e­mat­ics needed for grad­u­ate school. His pri­mary re­search in­ter­ests are in Growth The­ory and Em­pirics with a wide range of ap­pli­ca­tions in­clud­ing: de­vel­op­ing coun­tries, poli­ti­cal econ­omy, and the reper­cus­sions of tech­nolog­i­cal in­no­va­tion.

How the Ho­tel has helped

The Ho­tel has helped sub­stan­tially. Without the Ho­tel he would have been forced to join a PhD pro­gramme at Not­ting­ham or Manch­ester (where he had ap­plied to and been ac­cepted). The Ho­tel has given him the op­por­tu­nity to ap­ply to the top PhD pro­grammes in the UK (Oxford, Cam­bridge, LSE etc.). The Ho­tel has also al­lowed him to con­tinue de­vel­op­ing the math­e­mat­i­cal/​statis­ti­cal knowl­edge needed for a top tier PhD pro­gramme. It has also re­fined his re­search in­ter­ests, with a greater fo­cus on EA-re­lated causes.

The counterfactual

Kris says: “Although the coun­ter­fac­tual will not be com­pletely known un­til I hear back from the PhD pro­grammes, I still be­lieve the coun­ter­fac­tual im­pact of the Ho­tel on me is very large. I would have not been able to take a year out (for fi­nan­cial rea­sons) to ap­ply to PhD pro­grammes with­out the Ho­tel. Firstly, academia is quite elitist and so if I get into a bet­ter school, the po­ten­tial re­wards (both in terms of re­search qual­ity and salary which would be donated) are very large. Se­condly, the knowl­edge I’ve gained will trans­late to a bet­ter re­search out­put, ir­re­spec­tive of where I ul­ti­mately end up as a re­searcher. I now aim to fo­cus 100% of my re­search on EA-re­lated causes. If I had not stayed at the Ho­tel, I may have ded­i­cated some of my re­search ca­pac­ity to less im­por­tant causes. Thirdly, I strongly feel that my com­mit­ment to EA has dra­mat­i­cally im­proved. Value drift seems to be a con­cern amongst EAs. How­ever, I truly be­lieve in­ter­act­ing with the other EAs in the Ho­tel and the net­work I’ve gained has made me into a lifelong EA. Prior to com­ing to the Ho­tel, I felt slightly alienated with the views I had held. How­ever, this has changed dra­mat­i­cally. If these effects were to be quan­tified, I be­lieve they are by far in ex­cess of the costs.”

Sa­muel Knoche

Sa­muel is study­ing Ma­chine Learn­ing. Since he ar­rived at the Ho­tel, he has worked through the In­tro­duc­tion to Statis­ti­cal Learn­ing text­book, and made great strides in his study of deep learn­ing. He has repli­cated many pa­pers (DQN, World Models, Deep Dream, Style Trans­fer, Pix2Pix…) and hopes to get an earn­ing to give po­si­tion in ML re­search, and ul­ti­mately con­tribute to AI safety.

How the Ho­tel has helped

For an au­to­di­dact, the Ho­tel is ideal. It al­lows one to have full con­trol over one’s learn­ing progress, while also pro­vid­ing the so­cial sup­port to stave off de­pres­sion, and stay pro­duc­tive. Sa­muel there­fore also thinks that his ex­pe­rience suc­cess­fully self-teach­ing him­self deep learn­ing will give him more cred­i­bil­ity in call­ing for ed­u­ca­tion re­forms. He thinks that uni­ver­si­ties and star­tups in the ed­u­ca­tion space have a lot to learn from the EA Ho­tel model.

The counterfactual

Sa­muel would have ei­ther taken a ju­nior soft­ware en­g­ineer­ing po­si­tion, started some boot-camp/​cod­ing school or tried self-teach­ing at a sig­nifi­cantly slower pace while liv­ing with his par­ents. In any way, his long term ca­reer plans would have prob­a­bly taken at least 2 years longer, and his com­mit­ment to effec­tive al­tru­ism would have been sig­nifi­cantly lower than it is now.

Chris Leong

Chris origi­nally at­tended the EA Ho­tel in or­der to work on the prob­lem of In­finite Ethics. After find­ing the prob­lem sig­nifi­cantly harder than ex­pected, he de­cided to split his time be­tween try­ing to make head­way on this prob­lem and learn­ing more about AI Safety. Chris used his time at the Ho­tel to read Su­per­in­tel­li­gence, the RAISE course­work ma­te­rial and the Align­ment Fo­rum Se­quences. He also de­vel­oped the for­get­ting ap­proach to Log­i­cal Coun­ter­fac­tu­als at the Ho­tel. He be­lieves that this is the most promis­ing area of AI safety for him to work on and that he has already made sig­nifi­cant progress, most no­tably in his Align­ment Fo­rum post De­con­fus­ing Log­i­cal Coun­ter­fac­tu­als. He also found out about his cur­rent po­si­tion at Triple­byte from an­other per­son at the EA Ho­tel. This job is al­low­ing him to sup­port him­self while work­ing rel­a­tively few hours so that he can pur­sue re­search on the side and to work re­motely so that he can at­tend pro­grams like AI Safety Camp.

How the Ho­tel has helped

The Ho­tel al­lowed him to find out about the op­por­tu­nity to work at Triple­byte. It freed up his time with­out which his in­ves­ti­ga­tions into Log­i­cal Coun­ter­fac­tu­als might not have pro­ceeded very far, in­stead end­ing up on a long list of things to pur­sue some day and per­haps even for­got­ten.

The counterfactual

Chris prob­a­bly would have taken a nor­mal soft­ware de­vel­oper po­si­tion and not have any­where near as much time to pur­sue re­search on the side or been able to join AI Safety Camp. He would likely still be in Aus­tralia and have much less op­por­tu­nity to learn from oth­ers who are in­ter­ested in AI Safety.

Denisa Pop

Denisa has a PhD in Hu­man-An­i­mal In­ter­ac­tion and was plan­ning to pur­sue a ca­reer in academia. The Ho­tel al­lowed her to fo­cus her re­search en­tirely on high im­pact top­ics. She has since de­vel­oped and de­liv­ered work­shops on her psy­cholog­i­cal frame­work, “Ra­tional Com­pas­sion”, as a tool for as­piring EAs. Separately, she is also able to bring her 5 years’ ex­pe­rience as a coun­sel­ing psy­chol­o­gist to the Ho­tel to offer men­tal health ses­sions. She has also helped EA Nether­lands with or­ganis­ing con­fer­ences and re­treats and cur­rently works as an In­terim Com­mu­nity Man­ager at the Ho­tel.

How the Ho­tel has helped

She feels the Ho­tel pro­vided her a sup­port­ive, en­courag­ing and mo­ti­vat­ing en­vi­ron­ment, with peo­ple who share similar val­ues. She thinks that liv­ing, work­ing and do­ing ac­tivi­ties to­gether with this type of peo­ple am­plifies the benefits of be­ing part of an EA com­mu­nity.

The counterfactual

Denisa would have prob­a­bly been en­rol­led in a post-doc­torate pro­gramme (most likely on a non-EA topic) or do­ing men­tal health coun­sel­ing for non-EAs (she is also a li­censed psy­chol­o­gist). By not be­ing able to com­pletely put her val­ues into ac­tion and be­ing sur­rounded by peo­ple who don’t share her life views, she be­lieves that her pro­duc­tivity and life satis­fac­tion would have been de­creased.

Markus Salmela

Markus is a de­vel­oper for the up­com­ing AI Strat­egy role-play­ing game In­tel­li­gence Ris­ing. He also works on un­der­stand­ing the effects of tech­nolog­i­cally ex­tended hu­man lifes­pans from an ex­is­ten­tial risk per­spec­tive. He pre­vi­ously finished a re­search col­lab­o­ra­tion on x-risk fore­cast­ing while at the Ho­tel. His back­ground is mostly in Philos­o­phy and So­cial Sciences.

How the Ho­tel has helped

The Ho­tel pro­vided es­sen­tial fi­nan­cial sup­port, and the rel­a­tively easy ac­cess to Lon­don made it much more likely that Markus could use his con­tacts to join the de­vel­op­ment of In­tel­li­gence Ris­ing. Other Ho­tel res­i­dents were of­ten helpful for his pro­jects, giv­ing good sug­ges­tions and be­ing gen­er­ally sup­port­ive, which led to no­tice­ably greater mo­ti­va­tion and pro­duc­tivity.

The counterfactual

In the 3-4 years be­fore com­ing to the Ho­tel, Markus did oc­ca­sional minor con­tri­bu­tions to EA pro­jects, equiv­a­lent to about two weeks full time work an­nu­ally, while not mak­ing any dona­tions. Without the Ho­tel he would likely have con­tinued at this rate, un­less he re­ceived about the same amount of fund­ing from el­se­where to pay for liv­ing ex­penses.

Davide Zagami

Davide ini­tially joined the RAISE team and de­vel­oped les­sons on In­verse Re­in­force­ment Learn­ing, Com­putabil­ity, Logic and Set The­ory. After the pro­ject got dis­con­tinued, he at­tended AISC for ten days and then co-au­thored a work­shop pa­per about wire­head­ing while at the Ho­tel. Now he is al­lo­cat­ing part of his time to think­ing and read­ing more about tech­ni­cal AI Align­ment, and part of his time to self-study­ing Data Science and Ma­chine Learn­ing.

How the Ho­tel has helped

Davide thinks that the EA Ho­tel pro­vides him time and space to think about AI Align­ment and to ed­u­cate him­self about Ma­chine Learn­ing, as well as other top­ics. His gen­eral pro­duc­tivity im­proved con­sid­er­ably, as well as his abil­ity to bounce off ideas with other peo­ple in real time.

The counterfactual

Origi­nally, Davide would have ac­cepted a job offer in Italy as a soft­ware en­g­ineer for a non EA cause (ML is also not very pop­u­lar in Italy right now). He thinks this may have caused him to grad­u­ally fall out of touch with EA ideas, or at least to de­lay his EA work by about one year. He may have stud­ied AI Safety as an au­to­di­dact from time to time, and at a much re­duced pace, as he did for about six months be­fore join­ing the Ho­tel (there is no EA/​AI com­mu­nity in Italy, whereas the Ho­tel en­abled him to at­tend events such as AISC, HAAISC, IJCAI, and TAISU). If he had to leave now, he would spend much less time think­ing about AI. He may be able to find a job as a Data Scien­tist in Lon­don, if he is lucky, or he could go back to some­thing similar to that job op­por­tu­nity in Italy, more likely.

Anony­mous 1*

Anon used to be a mu­si­cian, en­couraged to pur­sue a ca­reer in the arts, even though they pri­vately thought this was a waste of time and were more in­ter­ested in EA. They came to the Ho­tel to re­train. They are now self-study­ing full time, aiming to be­come a re­searcher or op­er­a­tions spe­cial­ist.

How the Ho­tel has helped

The Ho­tel has given them the free­dom to leave home and be­gin to put their val­ues into ac­tion. They can study what they want, with­out hav­ing to ob­serve a cur­ricu­lum in­flated with use­less trivia. They can talk to and learn from plenty of in­tel­li­gent peo­ple who they would never have ap­proached had they not been in their im­me­di­ate en­vi­ron­ment. They are a lot hap­pier at the Ho­tel than they would’ve been oth­er­wise.

The counterfactual

They would still be self-study­ing, but at a much slower rate, whilst work­ing full-time out­side of the EA sphere. It would take con­sid­er­ably longer for them to make progress un­der these al­ter­nate cir­cum­stances, and they doubt they would have been able to sus­tain the mo­ti­va­tion to per­se­vere whilst staving off de­pres­sion.

Anony­mous 2*

Anon re­cently com­pleted an MS in math­e­mat­ics, with their the­sis re­search fo­cus­ing on math­e­mat­i­cal logic. Hav­ing finished, they moved to the Ho­tel 3 months ago to fo­cus on AI safety re­search. They’ve been try­ing to find new an­gles of at­tack on the prob­lem(s), and drafted up a few ideas along the way. Their posts on the Align­ment Fo­rum have been rel­a­tively well-re­ceived, not to men­tion the 8 other ideas they dropped on the Open Thread, a few of which they are cur­rently draft­ing up.

How the Ho­tel has helped

The Ho­tel pro­vides an en­vi­ron­ment where Anon can do re­search with­out hav­ing to worry about fi­nances or even prepar­ing food. They can also be more clued-in to EA and bounce ideas off peo­ple. In ad­di­tion, there have already been 3 AI safety-re­lated work­shops in Anon’s time here (or­ga­nized by Linda) which have been helpful for Anon in var­i­ous ways: giv­ing talks, airing ideas, hav­ing con­ver­sa­tions with peo­ple, hear­ing other ideas, and, in the case of the Learn-By-Do­ing work­shops, get­ting an ex­cuse to spend an af­ter­noon re­think­ing the Align­ment Prob­lem from the ground up. Also, Anon could at­tend EAG Lon­don thanks to be­ing at the Ho­tel, which they found valuable.

The counterfactual

Anon isn’t sure but thinks they would have con­tinued on in grad school, ei­ther in math or the­o­ret­i­cal CS, and con­tinued do­ing re­search they found in­ter­est­ing but ul­ti­mately ir­rele­vant to EA.

*these res­i­dents preferred not to be iden­ti­fied by name

RAISE

RAISE (Road to AI Safety Ex­cel­lence) was a char­ity that aimed to lower the cost of study­ing and teach­ing AI Safety. They were a team of 1.5 FTEs, in differ­ent coun­tries, un­til they moved to the EA Ho­tel to­gether and be­came a team of 4.3 FTEs, greatly in­creas­ing their pro­duc­tion rate.

How the Ho­tel helped

They had some fund­ing, but only enough to sup­port one of the two peo­ple work­ing on the pro­ject. The founder of the pro­ject was run­ning out of run­way. Part of the rea­son was that, de­spite their best efforts, pro­duc­tion speed was not on a satis­fac­tory level.

The Ho­tel al­lowed RAISE to add 3 more peo­ple to their pro­duc­tion team, in­creas­ing their pro­duc­tion speed four­fold. It helped the pro­ject sur­vive—when fund­ing dried up en­tirely in Novem­ber 2018, the pro­ject was able to keep afloat at nearly full ca­pac­ity thanks to the Ho­tel. 3 weeks later, they were given an ex­per­i­men­tal grant, cov­er­ing all of their ex­penses.

The pro­ject is now de­funct, af­ter con­sult­ing with the EA Long Term Fu­ture Fund.

The counterfactual

A startup is an ex­per­i­ment. The goal is not to suc­ceed, but to test a busi­ness model in com­bi­na­tion with a few ini­tial con­di­tions (founders, their cap­i­tal, etc). In this par­tic­u­lar case, the les­son was quite valuable for the peo­ple in­volved. Without the ex­tra 8 months that the Ho­tel ap­pended to the ex­is­tence of RAISE, the ex­per­i­ment would not have been car­ried out as thor­oughly, and the take­aways would have been as mea­ger as “they didn’t want to fund it but we don’t know why”.

Other re­cent guests (Case Stud­ies writ­ten in March 2019):

Hoagy Cunningham

Hoagy is cur­rently work­ing on an in­ter­na­tional pro­ject try­ing to ad­vance in­ter­pretabil­ity in re­in­force­ment learn­ers in his spare time while work­ing for RAISE dur­ing the day. Hoagy has a de­gree in Eco­nomics and Philos­o­phy from Oxford Univer­sity.

How the Ho­tel has helped

Be­fore com­ing to the Ho­tel he had a grow­ing in­ter­est in AI safety but had no friends in the area and lacked some skills to con­tribute, es­pe­cially in pro­gram­ming. He now has a fast-grow­ing net­work in the field and the abil­ity to con­tribute to pro­jects in a va­ri­ety of ways. He is now try­ing to work on AI safety full time. He cur­rently has an offer from the Civil Ser­vice Fast Stream as an economist and plans to use this as a jump­ing point to en­ter an AI policy role in the UK gov­ern­ment if he can­not find a role as a re­searcher.

[Up­date (Oc­to­ber 2019): Was a part of MIRI sum­mer fel­lows which he would have (90% con­fi­dence) been able to at­tend with­out hav­ing gone to the Ho­tel. He is now work­ing in the Home Office and ac­tively pur­su­ing op­por­tu­ni­ties to move into AI policy in gov­ern­ment.]

The counterfactual

Hoagy would have con­tinued self-study in math­e­mat­ics and search­ing for op­por­tu­ni­ties both in and out of AI safety. He would most likely have be­gun work­ing at a non-EA cor­po­rate role in Lon­don (60%) or found some­thing more ex­cit­ing (40%).

Ed­ward Wise

Ed­ward is a re­searcher and as­piring effec­tive al­tru­ist cur­rently work­ing on re­search­ing ob­jec­tions to effec­tive al­tru­ism, par­tic­u­larly from the left-wing poli­ti­cal sphere. Par­allel to this, Ed­ward has, on the ad­vice of oth­ers at the Ho­tel, ap­plied for fur­ther ed­u­ca­tion in In­ter­na­tional Re­la­tions, and is now be­gin­ning to ap­ply for jobs that would in­crease his ca­reer cap­i­tal.

How the Ho­tel has helped

The Ho­tel has helped in two pri­mary ways. The first was to give the fund­ing nec­es­sary for Ed­ward to re­search and be­gin writ­ing his ar­ti­cle(s) on the topic of left-wing crit­i­cisms of effec­tive al­tru­ism. The sec­ond, and more sig­nifi­cant one, is the way in which the com­mu­nity of the Ho­tel has en­couraged Ed­ward to shift his fo­cus to­ward the fu­ture. In par­tic­u­lar, he has been in­spired to pur­sue a ca­reer miti­gat­ing ex­is­ten­tial risk, and taken steps to­ward achiev­ing this.

The counterfactual

Pri­mar­ily, Ed­ward would likely not have iden­ti­fied as part of the effec­tive al­tru­ist com­mu­nity, and would have drifted away from the move­ment com­pletely. More con­cretely, he would not have been in­spired to pur­sue an EA-al­igned ca­reer, or taken steps to achieve this by ap­ply­ing for rele­vant fur­ther ed­u­ca­tion and jobs, in­stead likely rel­e­gat­ing this to a vague as­pira­tion. Fi­nally, he would not have had the time to pur­sue re­search which he had long been in­ter­ested in, and sincerely hopes will be use­ful to the EA com­mu­nity for both out­reach and in­tra-EA con­ver­sa­tions when pub­lished.

See here for a longer (but still in­com­plete) list of past guests


Ap­pendix B: FAQs

How do you mea­sure your im­pact?

Guests have fort­nightly check-ins to dis­cuss their work with the Com­mu­nity & Pro­jects Man­ager, and fill out a monthly sur­vey as part of a reg­u­lar group meet­ing. We think it’s tricky to gauge the out­come of pro­jects over the rel­a­tively short time pe­riod as has so far elapsed with the EA Ho­tel, es­pe­cially con­sid­er­ing that most of our guests are at an early stage in their ca­reer, but are track­ing sto­ries (on­go­ing coun­ter­fac­tual value self-re­ports are a re­quire­ment of longer stays at the Ho­tel) and col­lect­ing out­puts. In the fu­ture we in­tend to look at met­rics such as im­pact-ad­justed sig­nifi­cant plan changes (fol­low­ing 80,000 Hours) and coun­ter­fac­tual ad­di­tional money donated to EA char­i­ties.

We are mak­ing an in-depth at­tempt to es­ti­mate the (rel­a­tive) EV of the Ho­tel.

Why not have the res­i­dents pay by ap­ply­ing for EA grants?

Grants op­er­ate at a differ­ent level of gran­u­lar­ity—it takes a cer­tain amount of time to ap­ply for and pro­cess them, and this will in­evitably be longer for higher value grants. Since grants range in value from the equiv­a­lent of sev­eral months to sev­eral years of a one-per­son stay at the Ho­tel, they’ll in­evitably be sought by peo­ple who need more re­sources and are typ­i­cally fur­ther down the path. In other words, far fewer peo­ple will plau­si­bly re­ceive them than will stay at the Ho­tel.

One might con­sider EA Grants/​EA Funds as EA equiv­a­lents to ven­ture cap­i­tal funds, whereas the Ho­tel is more like an in­cu­ba­tor. Just as in the for-profit world, we think both roles are dis­tinct but com­ple­men­tary in get­ting the most from the EA ecol­ogy.

That said, some res­i­dents prob­a­bly will end up pay­ing with an EA grant, since grant and Ho­tel aren’t ex­clu­sive—and liv­ing in a low-cost hub would act as a mul­ti­plier on their grant’s value.

What are the risks?

Brand as­so­ci­a­tion: other EA or­gani­sa­tions have found that the ideas of effec­tive al­tru­ism are hard to ex­plain in low-fidelity me­dia. As long as we don’t have con­sid­er­able PR ex­per­tise on hand, we will re­ject all me­dia re­quests.

In­ter­per­sonal con­flict: given many peo­ple in close prox­im­ity, in­ci­dents of in­ter­per­sonal con­flict will some­times arise. We take the risk posed by such con­flicts se­ri­ously, and have sys­tems in place to both pre­vent con­flicts emerg­ing in the first place, and to me­di­ate and pre­vent es­ca­la­tion if they do.

Sup­port­ing pro­jects with nega­tive value: This is miti­gated by hav­ing a com­mu­nity on hand that is proac­tive in giv­ing cor­rect­ing feed­back. We are also have some re­spected com­mu­nity mem­bers act­ing as ad­vi­sors par­tic­u­larly on this is­sue.

What are com­pa­rable fund­ing op­por­tu­ni­ties?

EA Grants, EA Long-Term Fund and EA Meta Fund

Many of the pros and cons here are es­sen­tially two sides of a coin:

Com­par­a­tive ad­van­tages of EA Grants/​Funds:

    • They can sup­port pro­jects on a larger scale than the Ho­tel will (like a ven­ture cap­i­tal fund)

    • The av­er­age grantee might be higher value than the av­er­age Ho­tel guest, since they’ll have done more to prove themselves

    • Vir­tu­ally un­limited room for more funding

Com­par­a­tive ad­van­tages of the Ho­tel:

    • It can sup­port pro­jects on a smaller scale than EA funds will (like an in­cu­ba­tor)

    • The to­tal value might be higher per dol­lar, since a) benefi­cia­ries are (by defi­ni­tion) liv­ing at ex­tremely low cost for the de­vel­oped world and there­fore b) there are many more of them!

    • Limited room for more fund­ing, but acute short term need for it

In ad­di­tion, we see three non-analo­gous benefits to the Ho­tel:

    • The peer sup­port net­work dis­cussed above

    • By helping a wider spread of peo­ple rather than just those per­ceived as the elite, it (and hope­fully fu­ture pro­jects like it) can help build a sense of com­mu­nity sup­port sorely lack­ing in the EA world

    • Since funds are a one-off trans­ac­tion, they need to front-load their eval­u­a­tion, whereas the Ho­tel can (and does) have fort­nightly check-ins with its guests to en­sure they’re mak­ing rea­son­able progress to­wards a valuable goal. This gives the Ho­tel a unique ad­van­tage in eval­u­at­ing e.g. pro­jects with high var­i­ance in pos­si­ble out­comes—we can give sup­port con­di­tional on the pro­ject re­al­is­ing its po­ten­tial.

Char­ity En­trepreneur­ship—pro­jects run by Char­ity En­trepreneur­ship are well run and have been suc­cess­ful. How­ever, Char­ity En­trepreneur­ship seem to have plenty of run­way at the mo­ment and (as far as we are aware) com­par­a­tively lit­tle room for more fund­ing.

What will hap­pen to the EA Ho­tel if it isn’t funded?

Ini­tially, it will stay open and charge peo­ple cost price to stay, prob­a­bly re­mov­ing man­age­ment and some ameni­ties. This would re­move many of the benefits of the Ho­tel, and it would be go­ing against the main ini­tial idea (of pro­vid­ing a kind of “UBI” for liv­ing costs for those do­ing use­ful EA work). The size and value of the com­mu­nity at the Ho­tel would very likely be sub­stan­tially re­duced. It would in effect just be­come a large EA group house.

If this proves fi­nan­cially un­vi­able, the Ho­tel would be rented to a third party, or sold. A knock-on effect of the pro­ject ceas­ing due to lack of fund­ing would be a diminished like­li­hood of similar pro­jects be­ing at­tempted, de­spite their grass-roots pop­u­lar­ity among the EA move­ment.