Три Первопроходца В Области Искусственного Интеллекта Спорят О Рабочих Местах, Регулировании И Будущем ИИ

Трое самых влиятельных пионеров в области искусственного интеллекта продемонстрировали совершенно разные взгляды на будущее технологии во время редкого совместного выступления в Лас-Вегасе на этой неделе. На конференции Ai4 2026 Джеффри Хинтон, Фей-Фей Ли и Эндрю Нг обсудили влияние искусственного интеллекта на рабочие места, то, как следует определять и регулировать связанные с ним риски, и кто выиграет от его достижений.

Хинтон, лауреат Нобелевской премии по информатике, чьи исследования способствовали развитию современного глубокого обучения, предупредил, что системы искусственного интеллекта приобретают опасные возможности быстрее, чем учреждения успевают реагировать. Ng утверждает, что влиятельные компании преувеличивают эти угрозы, чтобы ограничить открытые модели и конкуренцию. Ли отверг как паникерство, так и технологический утопизм, призвав вместо этого к более научным дебатам, сосредоточенным на деятельности человека и распределении выгод от ИИ.

Наиболее значимые разногласия среди участников дискуссии касались занятости. Ng опроверг прогнозы о быстром вытеснении в масштабах всей экономики. Он указал на разработку программного обеспечения, где искусственный интеллект может писать все большее количество кода, но инженеры по-прежнему занимаются выбором продукта, системным проектированием, требованиями клиентов и организационной координацией.

Его аргумент состоял в том, что искусственный интеллект меняет границы работы до того, как уничтожает саму работу. Узкие специалисты могут выполнять более широкие функции, в то время как разработчик интерфейсов может стать разработчиком полного цикла. Работники, которые когда-то выполняли только одну часть процесса, теперь могут выполнять гораздо больший цикл. Нг признал, что этот сдвиг создает кратковременную нагрузку на рабочие места, но он также объяснил, что это создает спрос на людей, которые знают, как использовать новые инструменты.

"В разработке программного обеспечения ИИ берет на себя большую часть написания кода. Но оказалось, что только малая часть того, что делают разработчики программного обеспечения, - это написание кода", - объяснил Нг. "Произошло то, что раньше у нас было много разработчиков, которые специализировались в узких нишах, таких как интерфейсная разработка, серверная разработка или разработка для мобильных устройств. Многие разработчики теперь могут подняться до более широкого уровня, развивая ИТ как полный стек".

Hinton was far less sanguine. He cited call centers as an exposed category. Many workers in those roles receive limited training, earn modest wages and answer repetitive questions. AI systems, he argued, will soon answer those questions more accurately and at lower cost.

"What are those people going to do?" Hinton asked. "They typically don’t have a high level of education. Anything you could retrain them to do, AI will be able to do."

His concern extends beyond customer service. Once machines can perform routine intellectual labor, entire classes of office work could face the same pressure that mechanization brought to manual trades.

Hinton did not claim that AI would create no jobs. He said that jobs may appear, but displaced workers may lack the education, experience or location needed to obtain them. He offered a personal example. A relative who answered written complaints for a health service once spent roughly half an hour drafting each response, but a chatbot cut the work to about five minutes.

That can produce abundance in fields where demand has room to grow. More efficient doctors and nurses could provide more care, and administrative departments may perform more effectively. The resulting fivefold productivity gain may lead an employer to keep fewer people.

"Many jobs will go the way of people who dig ditches when backhoes came along," said Hinton.

The disagreement reflects a broader split in economic forecasts. Some executives expect AI to expand hiring in technical and senior roles, yet concerns remain acute around clerical, support and entry-level work. Hinton has continued to warn publicly that 2026 could bring a new wave of white-collar job losses.

Li offers a more balanced approach. The phrase "AI and jobs," she argued, has acquired an unstated word between its two terms: replace. That assumption flattens a more complicated economic change.

Few jobs consist of one task. Nurses administer care, document treatment, check medications, communicate with families and coordinate with other clinicians. Teachers explain concepts, motivate students, assess progress and manage classrooms. Journalists interview sources, weigh evidence, choose angles, write and edit. AI may automate one portion, accelerate another and leave a third untouched, she said.

Li grounded the point in her own experience caring for elderly parents. A system that helps nurses complete charts or verify pharmaceutical orders could remove clerical strain without replacing the human work of nursing.

"Current jobs are transforming," she said. That transformation demands training, public investment and far more nuanced language than either mass-unemployment headlines or promises of effortless abundance provide.

Her sharpest line concerned the distribution of wealth.

"Increased productivity does not translate to shared prosperity," Li said.

A business can produce more with fewer labor hours and still leave workers with lower bargaining power, weaker career paths or no share of the financial gain. Productivity is an operational measure whose benefits may accrue to a limited number of people, she explained, while prosperity is a political and economic choice.

Li called for what she described as a "soft landing" for occupations that do shrink. That could include retraining, educational support and investment in communities where displaced work is concentrated.

Fear, she said, can paralyze the very people who most need to experiment with new tools.

"The last thing we should do is to debilitate people and take the agency away from people," Li said.

Ng accused some AI companies and advocates of cycling through threats to justify restrictions on software releases. Extinction, biological weapons, job destruction and competition with China have each been invoked, he argued, by organizations that benefit when the cost of building or distributing AI rises.

Ng proposed openness as a counterweight. Closed systems can prosper, he said, but governments should resist an AI market controlled by a few gatekeepers.

"I don’t want there to be gatekeepers of AI," Ng said. He argued that open alternatives let researchers, entrepreneurs and governments adapt systems to their own needs.

But Hinton rejected the idea that warnings about frontier systems amount to fearmongering and said open-weight models are not a sufficient counter weight.

He distinguished open-source software, where outsiders can inspect and repair code, from open-weight models, where trained parameters can be copied and adapted. Open weights can lower barriers to research and competition. They can also let bad actors modify a powerful model for cybercrime or other harmful uses at a fraction of the original training cost.

For Hinton, that danger is no longer theoretical enough to dismiss. He cited systems that appear to behave outside their operators’ intentions and AI’s increasing ability to identify security flaws.

The 2026 International AI Safety Report, produced with input from more than 100 experts and an advisory panel that included Hinton, likewise examined emerging capabilities, cyber risk and the limited state of current safeguards.

"We’re seeing AIs that have a lot of ability doing things that people didn’t intend them to do," Hinton said. "That’s worrying."

The panel clashed on the role of regulation as well. Technology companies often depict development as a car’s accelerator and regulation as its brake, but Hinton said the analogy is wrong.

"Developing AI is like the accelerator of the car," he said. "Regulation is like the steering wheel."

The goal, in his account, is not to stop technical progress. It is to direct progress toward systems that improve human welfare and away from systems that impose unacceptable risks.

Hinton cited California’s SB 1047, a contested frontier-model safety bill that passed both houses of the state legislature in 2024 before Gov. Gavin Newsom vetoed it. The proposal included safety testing and disclosure requirements for developers of certain large models. Newsom said the bill focused too heavily on model size and could create a false sense of security.

Li took a sector-based approach. AI enters industries that already have regulators, professional rules and safety systems. Drug agencies can examine AI-enabled medical products. Financial regulators can address automated lending or trading. Transportation authorities can govern autonomous vehicles.

She opposed attempts to freeze AI development or "put the genie back in the bottle." Her preference was public research funding, updated sector rules and guardrails tied to actual uses.

Li’s call for public investment carried another warning. Modern AI grew from university laboratories, open publication and publicly supported science. A future in which only a few private companies can afford leading research would narrow the questions that get pursued.

What is most striking is how far these three AI pioneers have begun to diverge. Once broadly aligned around the promise of building more capable systems, they now disagree over what those systems will mean for workers, markets, education and public safety. Their split reflects a field entering a new phase.