Here’s a bar­gain of the most hor­ri­fy­ing kind: For less than $100,000, it may now be pos­si­ble to use ar­ti­fi­cial in­tel­li­gence to de­velop a vi­rus that could kill mil­lions of peo­ple. That’s the con­clu­sion of Ja­son Ma­theny, pres­i­dent of the Rand Corp., a think tank that stud­ies se­cu­rity mat­ters and other is­sues.

“It wouldn’t cost more to cre­ate a patho­gen that’s ca­pa­ble of kill­ing hun­dreds of mil­lions of peo­ple ver­sus a patho­gen that’s only ca­pa­ble of kill­ing hun­dreds of thou­sands of peo­ple,” Ma­theny said. It could cost bil­lions of dol­lars to pro­duce a new vac­cine or an­ti­vi­ral in re­sponse.

Tar­geted bioweap­ons

Some years ago, a re­search team cre­ated a cousin of the small­pox vi­rus, horse pox, in six months for $100,000, and with AI it could be eas­ier and cheaper to re­fine the vi­rus.

A re­tired Chi­nese gen­eral has raised the pos­si­bil­ity of bi­o­log­i­cal war­fare that tar­gets par­tic­u­lar races or eth­nic­i­ties (prob­a­bly im­per­fectly). This may be why the U.S. De­fense Depart­ment has said that the most im­por­tant long-term threat of bio­war­fare comes from China.

Advertise with us

Scientists for years have explored how AI may dominate warfare, with autonomous drones or robots programmed to find and eliminate targets instantaneously. Warfare may come to involve robots fighting robots.

One great uncertainty is the extent and timing of job losses that could amplify social unrest and lead to an explosion of deaths of despair and to the rise of right-wing populism.

AI has a more hope­ful side as well, of course. It holds the prom­ise of im­prov­ing ed­u­ca­tion, re­duc­ing auto ac­ci­dents, cur­ing can­cers and de­vel­op­ing mi­rac­u­lous new phar­ma­ceu­ti­cals.

One of the best-known ben­e­fits is in pro­tein fold­ing, which can lead to rev­o­lu­tion­ary ad­vances in med­i­cal care. Sci­en­tists used to spend years or de­cades fig­ur­ing out the shapes of in­di­vid­ual pro­teins, and then a Google ini­tia­tive called Al­phaFold was in­tro­duced that could pre­dict the shapes within min­utes.

Sci­en­tists have since used up­dated ver­sions of Al­phaFold to work on phar­ma­ceu­ti­cals in­clud­ing a vac­cine against ma­laria, one of the great­est kill­ers of hu­mans through­out his­tory.

Advertise with us

So it’s un­clear whether AI will save us or kill us first.

Tech­nol­ogy can op­press

One rea­son for my war­i­ness of AI is that while I see the prom­ise of it, the past 20 years have been a re­minder of tech­nol­ogy’s ca­pac­ity to op­press.

Smart­phones were daz­zling, but there’s ev­i­dence ty­ing them to de­te­ri­o­rat­ing men­tal health in young peo­ple. A ran­dom­ized con­trolled trial pub­lished just this month found that chil­dren who gave up their smart­phones en­joyed im­proved well-be­ing.

Dic­ta­tors have ben­e­fited from new tech­nol­o­gies. Liu Xiaobo, the Chi­nese dis­si­dent who re­ceived a Nobel Peace Prize, thought that “the in­ter­net is God’s gift to the Chi­nese peo­ple.” It did not work out that way: Liu died in Chi­nese cus­tody, and China has used AI to ramp up sur­veil­lance and tighten the screws on cit­i­zens.

AI may also make it eas­ier to ma­nip­u­late peo­ple. A study re­leased this year found that when Chat GPT-4 had ac­cess to ba­sic in­for­ma­tion about peo­ple it en­gaged with, it was about 80% more likely to per­suade some­one than a hu­man was with the same data. Con­gress was right to worry about ma­nip­u­la­tion of pub­lic opin­ion by the TikTok al­go­rithm.

Advertise with us

All this un­der­scores why it is es­sen­tial that the United States main­tain its lead in ar­ti­fi­cial in­tel­li­gence. As much as we may be leery of put­ting our foot on the gas, this is not a com­pe­ti­tion in which it is OK to be the run­ner-up to China.

The lim­its Joe Biden placed on China’s ac­cess to the most ad­vanced com­puter chips will help pre­serve our lead. The Biden ad­min­is­tra­tion has re­cruited first-rate peo­ple from the pri­vate sec­tor to think through these mat­ters and is­sued an im­por­tant ex­ec­u­tive or­der last year on AI safety, but we will also need to de­velop new sys­tems in the com­ing years for im­proved gov­er­nance.

I’ve writ­ten about AI-gen­er­ated deep­fake nude im­ages and vid­eos, and the ir­re­spon­si­bil­ity of both the deep­fake com­pa­nies and ma­jor search en­gines that drive traf­fic to deep­fake sites.

And tech com­pa­nies have pe­ri­od­i­cally used im­mu­ni­ties to avoid ac­count­abil­ity for pro­mot­ing the sex­ual ex­ploitation of chil­dren. None of that in­spires con­fi­dence in these com­pa­nies’ abil­i­ties to self-gov­ern re­spon­si­bly.

Sili­con Val­ley de­cides

“We’ve never had a cir­cum­stance in which the most dan­ger­ous, and most im­pact­ful, tech­nol­ogy re­sides en­tirely in the pri­vate sec­tor,” said Su­san Rice, who was Pres­i­dent Barack Obama’s na­tional se­cu­rity ad­viser. “It can’t be that tech­nol­ogy com­pa­nies in Sili­con Val­ley de­cide the fate of our na­tional se­cu­rity and maybe the fate of the world with­out con­straint.”

Advertise with us

I think that’s right. Man­ag­ing AI with­out sti­fling it will be one of our great chal­lenges as we adopt per­haps the most rev­o­lu­tion­ary tech­nol­ogy since Pro­metheus brought us fire.

Nicholas Kristof is a columnist for The New York Times, where this article first appeared. His previous article was “The genocide the world doesn't care about.”