[This pre­sen­ta­tion was in response to Tarleton Gillespie’s ​“The Relevance of Algorithms”]

I’m real­ly excit­ed to be here. I don’t often meet peo­ple who think about algorithms.

Just by way of pref­ac­ing, I don’t actu­al­ly study algo­rithms. What I study is a com­pa­ny that has been in the busi­ness of mak­ing algo­rithms for finan­cial insti­tu­tions for about 50 years. And my top­ic of research is a com­pa­ny called Fair Isaac, which most Americans have prob­a­bly heard of. They are the com­pa­ny behind the American cred­it score that’s known by the acronym FICO.

Now, because of my inter­est in the busi­ness of mak­ing algo­rithms and the his­to­ry of cred­it mar­kets, and of infor­ma­tion that is sold to run the cred­it mar­kets, I have cer­tain pre­oc­cu­pa­tions which don’t often appear in con­fer­ences about media stud­ies and top­ics of media. The his­to­ry of algo­rithms that I’m inter­est­ed is not the math­e­mat­i­cal his­to­ry, it’s the busi­ness his­to­ry. And it’s the busi­ness his­to­ry with­in the era of com­put­ing. And I would just note (I’m not sure this fits yet) to Kate that ago­nis­tic plu­ral­ism is of course the basis of the free mar­ket, and of the way that things are sup­posed to be exchanged in free mar­kets run by infor­ma­tion. I think there’s a link between the way I’m think­ing about algo­rithms and the way that Kate is. I just give that back­ground for you to keep in mind because I’m not going to step into the lan­guage of media stud­ies. I’m going to stick to the lan­guage that I know, and I hope it does­n’t feel too for­eign to you.

My pre­sen­ta­tion’s orga­nized in three parts.

In the first part I’d like to talk about the ecol­o­gy of com­mer­cial algo­rithms. What’s strik­ing to me as some­one who stud­ies algo­rithms in a com­plete­ly dif­fer­ent indus­try from Tarleton is that his state­ment of rel­e­vance applies first and fore­most to algo­rithms that play a role in the dis­tri­b­u­tion of media to a broad pub­lic. And Tarleton’s analy­sis applies to those algo­rithms whose com­mer­cial pur­pose (kind of like a light switch) is to bring a flow of infor­ma­tion, like the flow of elec­tric­i­ty, into the room to many peo­ple in many places. And those peo­ple are con­ceived of as con­sumers and as the public.

But not all algo­rithms, and cer­tain­ly not all com­mer­cial algo­rithms, actu­al­ly inter­face with pub­lic life. So to help refine the con­tours of what Tarleton is up to, I thought I’d try to sketch out a very pro­vi­sion­al and incom­plete an ecol­o­gy of the oth­er kinds of algo­rithms that cir­cu­late, so that we can sit­u­ate exact­ly what he’s talk­ing about in a more spe­cif­ic way. To me what Tarleton’s top­ic is will be the third and youngest cat­e­go­ry of algorithm. 

The first cat­e­go­ry is maybe the grand­fa­ther of the com­mer­cial algo­rithms. Those are the ones that were orig­i­nal­ly built and sold to busi­ness in the post-war peri­od, that had noth­ing to do with pub­lic com­mu­ni­ca­tion. The first com­mer­cial algo­rithms had as their pur­pose to help a very small num­ber of peo­ple, a set of peo­ple in exec­u­tive posi­tions, to wrest con­trol over large orga­ni­za­tions and to make bet­ter deci­sions in their posi­tion as exec­u­tives. So the first cat­e­go­ry of algo­rithms I would point to were man­age­r­i­al aids made by com­pu­ta­tion­al experts and offered to firms and gov­ern­ments to improve per­for­mance in the three-dimensional world.

And our men­tor, Chandra Mukerji, who is a men­tor both to me and to Tarleton, calls this pre­oc­cu­pa­tion with the move­ment of peo­ple and paper and things out­side in the three-dimensional world ​“logis­ti­cal pow­er.” She says this pow­er orig­i­nates with the state. And the very help­ful dis­tinc­tion she draws is that logis­ti­cal pow­er is not the pow­er of knowl­edge; it is the pow­er of engi­neer­ing. So state pow­er, she’s argu­ing, is not nec­es­sar­i­ly found­ed in knowl­edge. It may well be found­ed in the pow­er to engi­neer. So hold that thought because I’d like come back to it at the end.

Somewhere along the line, com­put­er sci­en­tists will engi­neer algo­rithms into the machine. So the algo­rithm will become part of the inside of the dig­i­tal infra­struc­ture. Now, inside the machine, the pur­pose of the algo­rithm will change. Its pur­pose will no longer be to pro­vide infor­ma­tion to an inde­pen­dent human deci­sion mak­er, but its pur­pose appears to me (as a non-technical per­son) to be to move infor­ma­tion itself inside of dig­i­tal infrastructure.

So, inside infor­ma­tion sys­tems, algo­rithms seem to play the role that mechan­ics play in indus­tri­al pro­duc­tion. The algo­rithm is an instru­ment of con­sis­tent repli­ca­tion of move­ment that brings the spir­it of indus­tri­al con­sis­ten­cy to bureau­cra­cy and infor­ma­tion man­age­ment. But of course it does this with one very impor­tant dif­fer­ence. Unlike its indus­tri­al pre­de­ces­sors, the algo­rithm as a machine does some­thing dif­fer­ent than a phys­i­cal mechan­i­cal sys­tem which sim­ply repeats the same action over and over and over and over. The algo­rithm has a kind of flex­i­bil­i­ty in it in its struc­ture, through math, that allows it to exe­cute action with a degree of respon­sive­ness. And that inter­nal math­e­mat­i­cal struc­ture allows it to adjust out­put depend­ing on chang­ing input conditions. 

The third cat­e­go­ry of algo­rithm, then, mov­ing on from algo­rithms that help exec­u­tives make deci­sions, to algo­rithms that move things inside machines… I think that third kind of algo­rithm is the one that belongs to Tarleton. His algo­rithms are inside machines, but they are medi­at­ing the move­ment of infor­ma­tion not to a small num­ber of peo­ple, and not to the machine itself, but they are medi­at­ing the trans­fer of infor­ma­tion to a broad­er spec­trum of users. And this cat­e­go­ry of com­mer­cial algo­rithms obvi­ous­ly does not exist until you have the wide­spread use of per­son­al com­put­ing. And that’s why I call it the youngest algo­rithm; that’s why I place it last in time.

So of course by now my ecol­o­gy isn’t real­ly just an ecol­o­gy, it’s also a chronol­o­gy. It’s about the trans­for­ma­tion of the use of com­mer­cial algo­rithms in dif­fer­ent ways. So as the sec­ond part of my pre­sen­ta­tion, I’d like to raise the ques­tion of how algo­rithms have changed in time.

Since I’m not used treat­ing algo­rithms as an inde­pen­dent top­ic, I hopped over to the Department of Management as LSE to look up my new friend Keith, who is a retired oper­a­tions researcher for British Airways. And of course the use of algo­rithms in con­trol­ling busi­ness was pio­neered in the air­line indus­try because get­ting peo­ple and planes togeth­er to move between geo­graph­i­cal loca­tions on time is a prob­lem that has large­ly been man­aged by algo­rithms. So Keith, who’s worked for British Airways for his entire career, it seemed to me he was the per­fect per­son to post the ques­tion, ​“What is an algo­rithm, and what is the scope of things I can expect to encounter at this con­fer­ence I’m going to in New York?” So this is what he said.

“That’s a very good ques­tion. When I start­ed,” he said, ​“there was a fair­ly pre­cise mean­ing of the term. An algo­rithm was a set of rules, which would gen­er­ate an opti­mum answer to the prob­lem that you’d posed it. It was a state­ment of rules that gave you the best pos­si­ble answer in a finite amount of time.” And he empha­sizes ​“finite amount of time” because he’s talk­ing about the days when he was still run­ning punch cards to do the com­pu­ta­tion. ​“As time has gone on”, he con­tin­ued, ​“the def­i­n­i­tion of algo­rithm has got­ten weak­er and weak­er. The strong def­i­n­i­tion still applies, but it’s not what most peo­ple mean when they use the term.”

“So what about Google?” I asked him, sort of press­ing him on.

And he says, ​“That’s not an algo­rithm. At least not the way that I mean it.”

So what can we make of this lit­tle frag­ment of empir­i­cal data? It seems to me that the two key com­po­nents in Keith’s response, the two key things that define an algo­rithm for him, are that they pro­vide not just any answer but an opti­mum answer, and it does so in a rea­son­ably finite peri­od of time.

Here’s my take on what has hap­pened: I think Keith is giv­ing me a clas­sic def­i­n­i­tion. And over the past fifty years there has been a per­mu­ta­tion in how algo­rithms are made, what they do, and what they’re sold for. And if my intu­ition is cor­rect, then we need to be very care­ful about how we frame the ques­tion raised by con­tent man­age­ment and finan­cial infor­ma­tion. Because it seems to me that to con­front algo­rithms on their own terms, we may have to mod­i­fy our pre­oc­cu­pa­tion with the pol­i­tics of knowl­edge and take up an inter­est in the pol­i­tics of logis­ti­cal engineering.

So this is my way of sort of rais­ing a ques­tion of geneal­o­gy. What is the geneal­o­gy of these algo­rithms? From a busi­ness per­spec­tive, if you trace the mak­ing of algo­rithms for sale as com­mer­cial objects, then Google looks a lot less like a pub­lic library and it looks a lot more like UPS. 

Part Three, very briefly.

How might think­ing in terms of con­trol over logis­tics help us to fig­ure out what it might mean to gov­ern algo­rithms? I hope you don’t feel like I’ve changed the top­ic too abrupt­ly. But in case you have, let me just tie quick­ly back to Tarleton’s work.

Tarleton has made a very point­ed obser­va­tion about val­ues of knowl­edge and objec­tiv­i­ty and how these become
resources for con­tent medi­a­tion com­pa­nies, even though engi­neer­ing inter­ven­tions are made on these com­mer­cial algo­rithms on a rou­tine basis, in a dis­cre­tionary fash­ion, by the cor­po­ra­tions that con­trol them, all the time. This is true of cred­it scor­ing as well. The empir­i­cal ques­tion then is, what does it mean to tam­per with an algo­rithm? And does tam­per­ing with an algo­rithm, does chang­ing the algo­rithm, change the way that the pub­lic is being constituted?

My response to Tarleton is to say well, it seems to me that that tam­per­ing means some­thing very dif­fer­ent with­in the log­ic of com­mer­cial engi­neer­ing than it does with­in the epoch of knowl­edge and epis­te­mol­o­gy. More specif­i­cal­ly, from an engi­neer­ing stand­point, opti­miza­tion is what anchors the con­cept of objec­tiv­i­ty. Optimization is the prin­ci­ple that allows you to test a sys­tem and then make a cri­tique that it is less than opti­mal or it is biased in some way. 

But what will it mean, and what can it mean to do a crit­i­cal analy­sis of algo­rithms that are commercially-engineered sys­tems in the absence of an opti­miza­tion imper­a­tive? So what Keith is sug­gest­ing [to] me is that today’s algo­rithms, the things that we call algo­rithms, don’t look any­thing like the ones that he calls algo­rithms because they don’t face an opti­miza­tion imperative.

So I don’t real­ly know the answer to this ques­tion. But I think the ambi­gu­i­ty here, the ten­sion between a prag­mat­ics of pro­pri­etary engi­neer­ing that per­mits con­stant adjust­ment of the tech­nol­o­gy on the one hand, and on the oth­er hand claims that what these tech­nolo­gies do is man­age the lega­cy of human knowl­edge, might help to explain what kind of pol­i­tics and what kind of gov­er­nance are at stake in the objects that Tarleton is studying.

Thank you very much.