I think one of the things I want to say from the start is it’s not like AI is going to appear. It’s actu­al­ly out there, in some instances in ways that we nev­er even notice. So for exam­ple check­ing cred­it card usage, pre­dict­ing patients who are like­ly to come back into the emer­gency room and there­fore keep­ing them from going home and then hav­ing to come back. There are some very clever uses of arti­fi­cial intel­li­gence in edu­ca­tion. But increas­ing­ly in ways in which we do notice it, for exam­ple the var­i­ous per­son­al assis­tants on our phones. So it’s out there mak­ing a dif­fer­ence, in most cas­es in sit­u­a­tions where it’s not replac­ing peo­ple but real­ly work­ing with people.

So I stress that dis­tinc­tion between replac­ing peo­ple and com­ple­ment­ing peo­ple because so much of the sci­ence fic­tion that’s out there and so much that’s in the press pre­sumes that the goal would be to replace peo­ple. But there’s a per­fect­ly won­der­ful way to replace human intel­li­gence, you know. It takes a man, a woman, cer­tain acts and you’re done. And human intel­li­gence is lim­it­ed in cer­tain ways, so why make that the aim? I mean, it has fas­ci­nat­ed peo­ple for cen­turies, prob­a­bly tied back to reli­gion and peo­ple won­der­ing or being con­cerned that peo­ple would try to imi­tate God, as it were. This is the sto­ry of the golem, it’s the sto­ry of Frankenstein, it’s the sto­ry of Ex Machina.

But that’s not the best way to think about devel­op­ing arti­fi­cial intel­li­gence meth­ods nor embody­ing them in com­put­er sys­tems. Rather, it would be bet­ter to com­ple­ment peo­ple, as many com­put­er sys­tems do now. So that’s the rea­son I make that dis­tinc­tion and urge it, is that regard­less of which two aims you pick, the sys­tems are going to exist— Unless we just send them to Mars by them­selves, they’re going to exist in a world that’s pop­u­lat­ed with human beings.

You can see this play­ing out, actu­al­ly, in some­thing that’s been in the press a lot recent­ly, which is autonomous and semi-autonomous vehi­cles. So for exam­ple autonomous vehi­cles, the idea is they just dri­ve; no per­son­’s involved in the dri­ving at all. Semi-autonomous vehi­cles do some dri­ving but then shift off with peo­ple. In both cas­es they’re inter­act­ing with peo­ple, so until we build roads on which the only vehi­cles are ful­ly autonomous, the vehi­cles are going to have to inter­act with peo­ple. And even if all the vehi­cles are ful­ly autonomous, we have to get rid of all of the pedes­tri­ans and all of the bicy­cles and everything.

That’s the issue with ful­ly autonomous, they will still have to inter­act with peo­ple. Semi-autonomous vehi­cles have to take into account peo­ple’s cog­ni­tive capac­i­ties in order to han­dle the so-called hand­off between peo­ple and com­put­er sys­tems appropriately. 

So, except in a few instances there’s no tak­ing peo­ple out of the pic­ture. I think it’s much more valu­able and soci­etal­ly use­ful to think from the very begin­ning of design­ing in ways to inter­act appro­pri­ate­ly with peo­ple, rather than build­ing some­thing sep­a­rate from peo­ple and then pre­sum­ing peo­ple will adjust to it.

What’s cru­cial at this point is to bring togeth­er exper­tise from these dif­fer­ent fields, and that that exper­tise has to be brought in before the sys­tems are designed and released to the world. And now is the time to think about this, to bring togeth­er peo­ple who are experts in arti­fi­cial intel­li­gence with peo­ple who under­stand ethics deeply, with psy­chol­o­gists who under­stand human cog­ni­tion, with social sci­en­tists who under­stand social orga­ni­za­tions, so that we can, as the rubric now is, make AI for social good.” And that rubric actu­al­ly cov­ers also build­ing sys­tems that help low-resource com­mu­ni­ties, build­ing sys­tems that pro­tect the envi­ron­ment, build­ing sys­tems that con­tribute to edu­ca­tion and healthcare.

I think both that we need to train and teach peo­ple about ethics— And here I want to say I’m not talk­ing about pro­fes­sion­al ethics. I’m talk­ing about real­ly under­stand­ing the trade­offs between con­se­quen­tial­ist ideas and deon­to­log­i­cal ideas, grap­pling with virtue ethics, think­ing about jus­tice, think­ing about who you’re serving—really a deep sense of ethics and about these sys­tems, and then make it part of the process of design of the sys­tems. It’s a years-long process of hav­ing peo­ple from these dif­fer­ent fields come togeth­er, explain their work, explain their per­spec­tives to each oth­er in ways that are acces­si­ble, treat those dif­fer­ent per­spec­tives with respect, and devel­op a com­mon vocab­u­lary and a way of approach­ing things togeth­er. That can’t be short-circuited. It’s real­ly a years-long process.