David Cox: Okay so I’m here today to tell you about an excit­ing fron­tier at the inter­sec­tion of neu­ro­science and com­put­er sci­ence. Now, I sus­pect that when many of you think about neu­ro­science, the first things that come to mind are med­ical appli­ca­tions: men­tal dis­or­ders, phar­ma­ceu­ti­cals. But what I’m going to try and argue today is that the stakes are much greater in the year 2015. 

So neu­ro­science is the study of the brain as mech­a­nism. And through­out his­to­ry we’ve sought to ratio­nal­ize and under­stand the brain and the mind in terms of our cur­rent tech­nol­o­gy. So, in the era of Decartes, we used hydraulic analo­gies: ani­mat­ing flu­ids that would move the body. In the era of Freud, steam: pres­sure build­ing up in the psy­che. In the era of radio, we start­ed to talk about chan­nels and communication. 

Now, in the era of the com­put­er we’re sur­round­ed by com­put­ers, so it’s nat­ur­al for us to use the metaphor of the com­put­er. So we talk about net­works in the brain. We talk about com­pu­ta­tions per­formed by neu­ronal cir­cuits. But the dif­fer­ence is that this isn’t just a metaphor. And that’s because com­put­er sci­ence gives us for­mal math­e­mat­i­cal tools for rea­son­ing about infor­ma­tion pro­cess­ing systems. 

So we have in com­put­er sci­ence a notion of the sep­a­ra­tion of an algo­rithm, which is what you’re com­put­ing, from the imple­men­ta­tion, which is how you’re com­put­ing it. So the same equiv­a­lence that lets us write code for a super­com­put­er and for a cell phone also lets us con­tem­plate writ­ing soft­ware for very dif­fer­ent kinds of hard­ware, in par­tic­u­lar bio­log­i­cal hard­ware. And seen through this lens, the brain is an amaz­ing­ly fas­ci­nat­ing and com­plex com­put­er. So there’s on the order of 100 bil­lion neu­rons in the brain, and on the order of 100 tril­lion con­nec­tions in between them called synaps­es. And the amaz­ing thing about the brain is it can do all kinds of things that com­put­ers for all of their sophis­ti­ca­tion are bad at. 

But the prob­lem isn’t that we lack the com­pu­ta­tion­al pow­er. Increasingly, we have enor­mous com­pu­ta­tion­al resources avail­able to us on this plan­et. The NSA has a zettabyte of stor­age in Utah. Google has count­less data cen­ters that crunch mil­lions of hours of video every day. Sheer com­pu­ta­tion­al pow­er isn’t the prob­lem, the prob­lem is that we don’t know the algo­rithms. We don’t know the soft­ware of the brain. 

Google Acquires Artificial Intelligence Startup DeepMind For More Than $500M;
NYU ​“Deep Learning” Professor LeCun Will Head Facebook’s New Artificial Intelligence Lab;
Twitter Acquires Image Search Startup Madbits;
Ex-Google Brain head Andrew Ng to lead Baidu’s arti­fi­cial intel­li­gence drive

But it turns out that already, labs around the world includ­ing my own lab have start­ed build­ing algo­rithms that are inspired by the brain and it turns out these are tremen­dous­ly eco­nom­i­cal­ly valu­able, even though they’re very prim­i­tive. So, pri­vate indus­try in the last two years alone has poured bil­lions of dol­lars into study­ing biologically-inspired computing.

But we’re very far from actu­al­ly still match­ing the pow­er of the brain. And to go that next step, what I argue is that we need to actu­al­ly go back to the brain, and reverse engi­neer the brain, to fig­ure out those pieces of those algo­rithms, of that soft­ware and that hard­ware that we’re miss­ing. So we have to study the nat­ur­al sys­tem and then build arti­fi­cial sys­tems that work the same way.

And it turns out that in 2015 there’s nev­er been a bet­ter time to study the brain. So we have advanced phys­i­o­log­i­cal tools. So this is basi­cal­ly wire­tap­ping the brain. So, dri­ven by sil­i­con micro­ma­chin­ing and nanofab­ri­ca­tion tech­nolo­gies, we can now put hun­dreds to thou­sands of elec­trodes in the brain, lit­er­al­ly wire­tap­ping the activ­i­ty of cells. 

And then there’s also been a renais­sance in imag­ing. So we can insert genes for flu­o­res­cent pro­teins that make neu­rons glow when they’re active. And this lets us lit­er­al­ly watch a thought in a liv­ing brain. So right now you’re watch­ing over on the right, a rat expe­ri­enc­ing a three-dimensional object, and you can see the cells lin­ing up in real time. 

Likewise, we also now have genet­ic tools for manip­u­lat­ing, not just record­ing but manip­u­lat­ing the activ­i­ty of cir­cuits in the brain. So we can per­turb them and see how they work. So this is a pic­ture that was tak­en by Feng Zhang, a col­league at MIT. And basi­cal­ly what this lets you do is you shine light on the cells and they either turn on or turn off. So lit­er­al­ly we con­trol neu­ronal cir­cuits at the flip of a light switch. 

And then more recent­ly and sort of on the more spec­u­la­tive cut­ting edge, we also have this new tech­nol­o­gy, or new fron­tier called con­nec­tomics, where we can thin­ly slice the brain and then actu­al­ly trace out the wiring dia­gram across many many slices. So this was an image cre­at­ed by a col­league down the hall from me, Jeff Lichtman at Harvard. And what this lets us do, this is just one small piece of the brain that’s been recon­struct­ed. But you could imag­ine that the entire wiring dia­gram is an enor­mous cor­pus of data. 

So, we can now begin to con­tem­plate hav­ing the func­tion of the brain, the behav­ior, being able to per­turb its activ­i­ty, and then also hav­ing the wiring diagram. 

Now, the stakes of this can’t be under­es­ti­mat­ed. So one obvi­ous thing is that there are many dis­or­ders, dis­eases, that we can con­tem­plate approach­ing once we under­stand the soft­ware of the brain. So, many dis­eases don’t have a genet­ic or anatom­i­cal smok­ing gun; they seem to be errors in the soft­ware of the brain. And if we under­stand the soft­ware of the brain, we can start to think about how we might start to cor­rect some of these problems. 

But even beyond that, we can imag­ine a world where if we can recre­ate the abil­i­ties of the brain, we can start hav­ing machines do things for us. So dri­ve our cars. We can have machines that can diag­nose our med­ical images. We can have machines that free us from labor. If you think about many of the jobs in the world that basi­cal­ly require a work­ing brain, these are things we can start to replace and aug­ment with machines. 

Now this is going to be an enor­mous, mul­ti­dis­ci­pli­nary effort. It’s going to take biol­o­gists, com­put­er sci­en­tists, neu­ro­sci­en­tists, chemists. It will also take ethi­cists, it will take pol­i­cy­mak­ers, for us to real­ly nav­i­gate this new fron­tier of pos­si­bil­i­ties that neu­ro­science and com­put­er sci­ence make pos­si­ble. Thank you.