Sil­i­con Val­ley fears los­ing its edge in AI

The Idaho Statesman (Sunday) - - NEWS - BY CADE METZ New York Times SAN FRAN­CISCO

A com­mon be­lief among tech in­dus­try in­sid­ers is that Sil­i­con Val­ley has dom­i­nated the in­ter­net be­cause much of the world­wide net­work was de­signed and built by Amer­i­cans.

Now a grow­ing num­ber of those in­sid­ers are wor­ried that pro­posed ex­port re­stric­tions could short­cir­cuit the pre-em­i­nence of U.S. companies in the next big thing to hit their in­dus­try: ar­ti­fi­cial intelligence.

In Novem­ber, the Com­merce Depart­ment re­leased a list of tech­nolo­gies, in­clud­ing ar­ti­fi­cial intelligence, that are un­der con­sid­er­a­tion for new ex­port rules be­cause of their im­por­tance to na­tional se­cu­rity.

Tech­nol­ogy ex­perts worry that block­ing the ex­port of AI to other coun­tries, or ty­ing it up in red tape, will help AI in­dus­tries flour­ish in those na­tions – China, in par­tic­u­lar – and com­pete with U.S. companies.

“The num­ber of cases where ex­ports can be suf­fi­ciently con­trolled are very, very, very small, and the chance of mak­ing an er­ror is quite large,” said Jack Clark, head of pol­icy at Ope­nai, an ar­ti­fi­cial intelligence lab in San Fran­cisco. “If this goes wrong, it could do real dam­age to the AI com­mu­nity.”

The ex­port con­trols are be­ing con­sid­ered as the United States and China en­gage in a trade war.

The Trump ad­min­is­tra­tion has been crit­i­cal of the way China deals with U.S. companies, of­ten re­quir­ing the trans­fer of tech­nol­ogy to Chi­nese part­ners as the cost of do­ing busi­ness in the coun­try. And fed­eral of­fi­cials are mak­ing an ag­gres­sive ar­gu­ment that China has stolen U.S. tech­nol­ogy through hack­ing and in­dus­trial es­pi­onage.

Tech companies, aca­demics and pol­i­cy­mak­ers are call­ing on the Com­merce Depart­ment to take a light hand with AI ex­port rules be­fore a Jan. 10 dead­line for pub­lic com­ment. Their ar­gu­ment has three main points: Re­stric­tions could harm companies in the United States and help in­ter­na­tional com­peti­tors. They could sti­fle tech­nol­ogy im­prove­ments. And they may not make much of a dif­fer­ence.

In Au­gust, Congress passed the Ex­port Con­trols Act of 2018, which added ex­port re­stric­tions to “emerg­ing and foun­da­tional tech­nolo­gies.” In mid-novem­ber, the Com­merce Depart­ment, tasked with over­see­ing the re­stric­tions, pub­lished a list of tech­nolo­gies for con­sid­er­a­tion, in­clud­ing sev­eral cat­e­gories of AI like com­puter vi­sion, speech recog­ni­tion and nat­u­ral lan­guage un­der­stand­ing.

The re­stric­tions would af­fect the ex­port of tech­nol­ogy to cer­tain coun­tries. Though it does not spec­ify which ones, the Com­merce Depart­ment pro­posal points to coun­tries that have faced trade and arms em­bar­goes in the past. Those in­clude China, Rus­sia and Iran.

The Com­merce Depart­ment de­clined to com­ment on the pro­posed re­stric­tions. Af­ter tak­ing pub­lic com­ments, the depart­ment will draft a for­mal plan. It could re­sult in any­thing from new li­cens­ing rules for AI ex­ports to out­right bans.

Ar­ti­fi­cial intelligence could be as vi­tal to the tech in­dus­try as the World Wide Web was 25 years ago.

It is easy to for­get that be­fore the mid-1990s, the tech in­dus­try in the United States was in a slump and the consumer elec­tron­ics gi­ants of Ja­pan looked un­beat­able. That changed as Sil­i­con Val­ley companies pop­u­lar­ized in­no­va­tions like the web browser and e-com­merce. To­day, the likes of Google, Ap­ple and Face­book dom­i­nate tech all over the world.

But U.S. do­min­ion in ar­ti­fi­cial intelligence is far from guar­an­teed. China is in­vest­ing bil­lions in AI for both com­mer­cial and mil­i­tary use. And the ex­per­tise that has flooded to Sil­i­con Val­ley over the years is start­ing to stay home, partly be­cause of im­mi­gra­tion re­stric­tions. Cen­ters for AI re­search are pop­ping up all over the world, from Toronto to Cam­bridge, Eng­land, and Bei­jing.

En­forc­ing ex­port con­trols on ar­ti­fi­cial intelligence would present an un­usual chal­lenge for reg­u­la­tors.

AI is some­thing pol­i­cy­mak­ers call a dual-use tech­nol­ogy. It has in­nocu­ous com­mer­cial ap­pli­ca­tions, like help­ing to steer your car or ask your phone a ques­tion. It also has im­por­tant mil­i­tary uses, like help­ing a weaponized drone find its tar­gets.

De­fense plan­ners believe AI rep­re­sents the next no­table change in mil­i­tary weapons, akin to the tac­ti­cal tar­get­ing of smart bombs a gen­er­a­tion ago or nu­clear weapons be­fore that. AI can also aid sur­veil­lance sys­tems and even dis­in­for­ma­tion cam­paigns through soft­ware that can pro­duce fake pho­tos and videos.

But “try­ing to draw a line be­tween what is mil­i­tary and what is com­mer­cial is ex­ceed­ingly dif­fi­cult,” said R. David Edel­man, a tech­nol­ogy pol­icy re­searcher at the Mas­sachusetts In­sti­tute of Tech­nol­ogy. “It may be im­pos­si­ble.”

It is dif­fi­cult to put a “made in Amer­ica” la­bel on ar­ti­fi­cial intelligence. Re­search on the tech­nol­ogy is of­ten done col­lab­o­ra­tively by sci­en­tists and en­gi­neers all over the world.

Companies rarely hold on to the details of their AI work, as if it were a se­cret recipe. In­stead, they share what they learn, in hopes that other re­searchers can build on it. One com­pany’s “break­through” is of­ten the lat­est it­er­a­tion of what many re­searchers at pri­vate companies and uni­ver­si­ties have been work­ing on.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.