Offer of collaboration and/​or mentorship

I have two mo­ti­va­tions for mak­ing this offer. First, there have been dis­cus­sions re­gard­ing the lack of men­tor­ship in the AI al­ign­ment com­mu­nity, and that be­gin­ners find it difficult to en­ter the field since the ex­pe­rienced re­searchers are too busy work­ing on their re­search to provide guidance. Se­cond, I have my own re­search pro­gramme which has a sig­nifi­cant num­ber of shovel ready open prob­lems and only one per­son work­ing on it (me). The way I see it, my re­search pro­gramme is a very promis­ing ap­proach that at­tacks the very core of the AI al­ign­ment prob­lem.

There­fore, I am look­ing for peo­ple who would like to ei­ther re­ceive men­tor­ship in AI al­ign­ment rele­vant top­ics from me, or col­lab­o­rate with me on my re­search pro­gramme, or both.


I am plan­ning to al­lo­cate about 4 hours /​ week to men­tor­ship, which can be done over Skype, Dis­cord, email or any other means of re­mote com­mu­ni­ca­tion. For peo­ple who hap­pen to be lo­cated in Is­rael, we can do in per­son ses­sions. The math­e­mat­i­cal top­ics in which I feel qual­ified to provide guidance in­clude: lin­ear alge­bra, calcu­lus, func­tional anal­y­sis, prob­a­bil­ity the­ory, game the­ory, com­putabil­ity the­ory, com­pu­ta­tional com­plex­ity the­ory, statis­ti­cal/​com­pu­ta­tional learn­ing the­ory. I am also more or less fa­mil­iar with the state of the art in the var­i­ous ap­proaches other peo­ple pur­sue to AI al­ign­ment.

Nat­u­rally, peo­ple who are in­ter­ested in work­ing on my own re­search pro­gramme are those who would benefit the most from my guidance. Peo­ple who want to work on em­piri­cal ML ap­proaches (which seem to be dom­i­nant in OpenAI, Deep­Mind and CHAI) would benefit some­what from my guidance, since many the­o­ret­i­cal in­sights from com­pu­ta­tional learn­ing the­ory in gen­eral and my own re­search in par­tic­u­lar, are to some ex­tent ap­pli­ca­ble even to deep learn­ing al­gorithms whose the­o­ret­i­cal un­der­stand­ing is far from com­plete. Peo­ple who want to work on MIRI’s core re­search agenda would also benefit some­what from my guidance but I am less knowl­edge­able or in­ter­ested in for­mal logic and ap­proaches based on for­mal logic.


Peo­ple who want to col­lab­o­rate on prob­lems within the learn­ing-the­o­retic re­search pro­gramme might re­ceive a sig­nifi­cantly larger frac­tion of my time, de­pend­ing on de­tails. The com­mu­ni­ca­tion would still be mostly re­mote (un­less the col­lab­o­ra­tor is in Is­rael), but phys­i­cal meet­ings in­volv­ing flights are also an op­tion.

The origi­nal es­say about the learn­ing-the­o­retic pro­gramme does men­tion a num­ber of more or less con­crete re­search di­rec­tions, but since then more shovel ready prob­lems joined the list (and also, there are a cou­ple of new re­sults). In­ter­ested peo­ple are ad­vised to con­tact me to hear about those prob­lems and dis­cuss the de­tails.


Any­one who wants to con­tact me re­gard­ing the above should email me at vanessa.kosoy@in­tel­li­, and give me a brief in­tro about em­self, in­clud­ing knowl­edge in math /​ the­o­ret­i­cal comp­sci and pre­vi­ous re­search if rele­vant. Con­versely, you are wel­come to browse my writ­ing on this fo­rum to form an im­pres­sion of my abil­ities. If we find each other mu­tu­ally com­pat­i­ble, we will dis­cuss fur­ther de­tails.