I think that arti­fi­cial intel­li­gence absolute­ly pos­es a chal­lenge for soci­ety and human­i­ty. I think there are var­i­ous things when we say arti­fi­cial intel­li­gence. There’s arti­fi­cial gen­er­al intel­li­gence, which is the idea that there’s a sin­gu­lar­i­ty com­ing and that some­thing will become so smart that we won’t be able to con­trol it, and it might even decide that human beings are kind of a bad idea and get rid of us. And I think that’s a real threat. I think that it’s not…in my view immi­nent. I think that we have a lit­tle bit of time. And I think actu­al­ly what I’m more con­cerned about per­son­al­ly is the machine learning.

Machine learn­ing sys­tems that we have today have become so pow­er­ful and are being intro­duced into every­thing from self-driving cars, to pre­dic­tive polic­ing, to assist­ing judges, to pro­duc­ing your news feed on Facebook on what you ought to see. And they have a lot of soci­etal impacts. But they’re very dif­fi­cult to audit. They’re not like nor­mal soft­ware pro­grams where you can just read the code and under­stand what they do. In fact even the devel­op­ers unless you test them don’t under­stand exactly—they can’t pre­dict exact­ly what the out­come is going to be.

So there are esti­mates that self-driving cars may reduce traf­fic acci­dents by 90%. There are types of diag­nos­tics where machines seem to be far­ing much bet­ter than human beings at diag­nos­ing dis­eases. There is a pos­si­bil­i­ty that things like parole or bail may be very quick­ly shown to be bet­ter judged by machines. These raise some inter­est­ing ques­tions because these are lives at risk, and lives that could be saved.

As we start to intro­duce these things, our reg­u­la­to­ry frame­works, the way we think about how soci­ety will work under these new sys­tems whether we’re talk­ing about jobs or whether we’re talk­ing about the law or whether we’re talk­ing about tech­ni­cal archi­tec­ture, all of these things are going to change.

At the Media Lab, we use the word anti-dis­ci­pli­nary because we find that the tra­di­tion­al dis­ci­plines, both in busi­ness and acad­e­mia, tend to rein­force a spe­cial­iza­tion which sort of the cliché is you learn more and more about less and less. And that’s impor­tant when you’re going deep. But when you have a tech­nol­o­gy like AI that cuts across all of these dis­ci­plines in terms of their impact, you need to cre­ate this tis­sue in between. And I wor­ry a lit­tle bit that the peo­ple who are design­ing and deploy­ing these sys­tems are com­put­er sci­en­tists who are try­ing to solve the world’s prob­lems through com­put­er sci­ence, and that the con­nec­tive tis­sue between the dis­ci­pline of machine learn­ing and com­put­er sci­ence, and the oth­er dis­ci­plines like social sci­ences or law, or even phi­los­o­phy, that those com­mu­ni­ties aren’t real­ly able to talk to each oth­er because the lan­guage is so dif­fer­ent and there isn’t a lot of cul­ture of inter­ac­tion between those com­mu­ni­ties, and I think the way we address this is to start cre­at­ing much more inter­dis­ci­pli­nary work.

As we were think­ing about how we might tack­le some of the miss­ing pieces in think­ing about where AI should go, I thought about it with var­i­ous hats. I thought about it with my MacArthur Foundation hat, with my Knight Foundation hat, the Media Lab hat, and just sort of a cit­i­zen of the world hat. And I real­ized that all of the pieces that need­ed to be at the table weren’t in a sin­gle insti­tu­tion. You could­n’t give all the mon­ey to Media Lab. You could­n’t give all the mon­ey to Berkman Center. You could­n’t give all the mon­ey to any­body and get all of the dif­fer­ent voic­es that we need­ed. And not just voic­es. Everybody has a dif­fer­ent frame­work. So the way that the Harvard Law School thinks about the the­o­ry of change in think­ing through prob­lems is very dif­fer­ent than the way that the Media Lab would do it.

So I think the key thing, and you can see through the diver­si­ty of the dif­fer­ent peo­ple fund­ing this ini­tia­tive as well as the peo­ple who are involved in coor­di­nat­ing it, we’re hop­ing to bring both a diver­si­ty of geog­ra­phy, a diver­si­ty of tech­nol­o­gy ver­sus also field diver­si­ty. But also just a fun­da­men­tal the­o­ry of change and at what lay­er we should inter­vene. And so I think the first cou­ple of years we’re going to be doing a lot of real­ly inter­est­ing exper­i­ments, and hope­ful­ly by the end of this process we’ll have a pret­ty good idea of sev­er­al dif­fer­ent things that we should set up either as insti­tu­tions or as fund­ing opportunities. 

And I think it’s impor­tant to start hav­ing the conversation—not just a con­ver­sa­tion but doing the work around the pol­i­cy, think­ing about how soci­ety should be inte­grat­ed and respond, before it’s too late. Because I think one of the prob­lems is that once you move past cer­tain points it’s going to be dif­fi­cult to roll back. And so I think timing-wise, sort of begin­ning last year and this year, is real­ly the key point in bring­ing oth­ers into this process. Because up till now the com­put­er sci­ence was just get­ting to the point where it was ready to be deployed. Right now it’s sort of just right or almost a lit­tle bit too late to get start­ed. So I think the tim­ing is super important.