Ajay Agrawal

Rotman Management Magazine - - FROM THE EDITOR -

Founder, Cre­ative De­struc­tion Lab and Ma­chine Learn­ing and the Mar­ket for In­tel­li­gence Con­fer­ence; Ge­of­frey Taber Chair in En­trepreneur­ship and In­no­va­tion, Rot­man School of Man­age­ment

“YOU CAN SEE THE COM­PUTER AGE ev­ery­where but in the pro­duc­tiv­ity sta­tis­tics.” So stated No­bel Lau­re­ate and MIT eco­nomics pro­fes­sor Robert Solow in 1987. Even­tu­ally, econ­o­mists found where the pro­duc­tiv­ity gains from the com­puter age were hid­ing: in the fu­ture. While they even­tu­ally showed up, they took longer than ex­pected be­cause they were tied to in­vest­ments in ‘com­ple­ments’ — all of the things other than al­go­rithms/mod­els that are nec­es­sary to make com­mer­cial-grade AI work (data, re­designed work­flows, train­ing, reg­u­la­tion, hu­man judg­ment, in­fra­struc­ture, etc.).

As in the com­puter age, the wide­spread pro­duc­tiv­ity gains as­so­ci­ated with ma­chine in­tel­li­gence will de­pend on in­vest­ments in com­ple­ments. As we shift from tech­ni­cal achieve­ments in AI (‘Look ev­ery­one! The AI can read a hand­writ­ten ad­dress on an en­ve­lope!’ ‘The AI can drive a car!’ ‘The AI can clas­sify a med­i­cal im­age!’) to large-scale com­mer­cial de­ploy­ment, the de­sign and im­ple­men­ta­tion of com­ple­ments will be para­mount.

The com­puter sci­en­tists de­sign­ing AIS are far ahead of those build­ing the com­ple­ments — in­dus­try prac­ti­tion­ers, so­cial sci­en­tists, reg­u­la­tors and the like. Now that ev­ery­one has re­al­ized the sweep­ing po­ten­tial of AI, com­pa­nies and coun­tries are racing to cre­ate and con­trol the com­ple­ments. While the al­go­rithms are soft­ware and thus have low bar­ri­ers to en­try (not­with­stand­ing scale ad­van­tages with re­spect to train­ing data), many com­ple­ments re­quire sig­nif­i­cant cap­i­tal ex­pen­di­ture and thus have higher en­try bar­ri­ers. There­fore, com­pe­ti­tion pol­icy and mar­ket dy­nam­ics will move even fur­ther onto cen­tre stage.

In other words, we are en­ter­ing the next phase of the AI rev­o­lu­tion: com­pe­ti­tion in the mar­ket for AI com­ple­ments. This will feel dif­fer­ent from what we’ve ex­pe­ri­enced so far. The gen­teel com­pe­ti­tion among com­puter sci­en­tists on dis­play at

con­fer­ences like NIPS that is based on the per­for­mance of new AI al­go­rithms against well-spec­i­fied tech­ni­cal bench­marks like Ima­genet will give way to com­pe­ti­tion among firms over the own­er­ship and con­trol of scarce com­ple­ments such as data, in­fra­struc­ture, tal­ent and re­la­tion­ships.

For en­ter­prises, com­pe­ti­tion in the semi-sci­en­tific cul­ture of al­go­rith­mic per­for­mance against bench­marks was cu­ri­ous and novel. How­ever, com­pe­ti­tion over com­ple­ments is fa­mil­iar ter­ri­tory. And given the size of the prize, this com­pe­ti­tion is likely to get rough and tum­ble, as cor­po­rate AI strate­gies de­pend at least as much on com­ple­ments as al­go­rithms. In­ten­si­fied com­pe­ti­tion will in­crease the pres­sure on com­pa­nies to de­liver re­sults. In­ter­nal de­bates like the one at Google re­gard­ing whether to aban­don Project Maven — a col­lab­o­ra­tion with the U.S. Depart­ment of De­fence to uti­lize AI for im­age anal­y­sis that could po­ten­tially be used to im­prove drone strikes — will seem quaint. Fur­ther­more, com­pe­ti­tion will not only in­ten­sify at the com­pany level. In re­cent months, one coun­try af­ter an­other has an­nounced its na­tional AI Strat­egy — and most of them read more like in­dus­trial than science pol­icy. Com­pe­ti­tion over com­ple­ments is about to be­come fierce.

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.