This A.I. Company Wants to Take Your Job

Years ago, when I started writing about Silicon Valley’s efforts to replace workers with artificial intelligence, most technology executives had at least decency for lies.

“We do not perform the automation of the workers, we more They are, “Executive Managers Tell me. “Artificial intelligence tools will not destroy jobs. They will be useful assistants that will free workers from worldly toil.”

Of course, lines like those – which were often aimed at reassuring nervous workers and giving coverage of corporate automation plans – said more than technology restrictions more than the motives of executives. At that time, artificial intelligence was not simply good enough to automate most of the jobs, and certainly was not able to replace learned workers in universities in white collar industries such as technology, consulting and financing.

This started to change. Today, some artificial intelligence systems can write programs, produce detailed research reports, and solve math and complex science problems. The latest “agents” of artificial intelligence are able to carry out long sequences of tasks and verify their work, in the way a person desires. Although these systems are still less than humans in many areas, some experts are concerned that the last height of unemployment for university graduates is a sign of this Companies already use artificial intelligence As an alternative to some novice workers.

On Thursday, she got a glimpse of a post-working future at a contract in San Francisco by a mechanic, a new company of artificial intelligence with a bold goal of automating all jobs-costumes, mining our doctors and lawyers, and people who write our programs, design our buildings and sponsor our children.

“Our goal is to fully automate,” said Tamay Besiroglu, 29, one of the founders of Mechanism. “We want to reach the entire automatic economy, and make it happen as soon as possible.”

A full automation dream is not new. John Mainard Keynes, in the 1930s, predicted that the machines would automate almost all jobs, creating the abundance of materials and leaving people free to pursue their emotions.

This never happened, of course. But recent developments in artificial intelligence raised the belief that technology capable of automating teamwork is close. Dario Amoudi, CEO of Anthropor, Recently Artificial intelligence can resolve up to half of all the functions of white collars in the next five years.

Mechanical is one of the number of startups that make it possible. The company was established this year by Mr. Besiroglu, EGE Erdil and Matthew Barnett, who worked together in EPOCH AI, a research company that studies the capabilities of artificial intelligence systems.

She has Investments attracted Among the well -known technology leaders including Patrick Coleson, founder of Stripe, Jeff Dean, the chief artificial intelligence scientist in Google. She now has five employees, and works with the leading AI companies. (I refused to say any of them, referring to secrecy agreements.)

A mechanical approach focuses on automating jobs using artificial intelligence on a technique known as reinforcement learning – the same method that has been used to train a computer to play the game of the plate at a super level about a decade ago.

Today, the leading AI companies use learning to enhance to improve their language output outputs, by making an additional account before creating an answer. These models, often called “thinking” or “thinking” models, have become impressive in some narrow tasks, such as writing a code or solving mathematics problems.

But most jobs involve doing more than one task. The best AI models today are still unreliable enough to deal with the most complex work burdens, or to move in complex companies systems.

To fix this, mechanics create new training environments for these models – mainly, detailed tests that can be used to teach models what to do in a specific scenario, and judge whether or not they have succeeded.

To automate software engineering, for example, the mechanics builds a computer -like training environment that the software engineer – a virtual device equipped with an email, reckless account, some coding and web browser tools. The artificial intelligence system is required to achieve an important use of these tools. If you succeed, it gets a reward. If you fail, it gets a penalty. Then he tries again. With enough experience and error, if the simulation is well designed, you should learn artificial intelligence in the end what a human engineer does.

“It is similar to the effectiveness of creating a very boring video game,” said Mr. Pesiroglu.

Mechanics begins with computer programming, a profession in which he showed learning to reinforce some promise already. But he hopes to use the same strategy to automate jobs in many other white collars.

“We will really know that we have succeeded as soon as we created artificial intelligence systems that are able to assume almost every responsibility that a person can implement on a computer.” books In a modern blog post.

I have some doubts about whether a mechanical approach will work, especially for non -technical jobs where success and failure are not easily measured. (What does it mean, for example, that Amnesty International “succeeds” in being a high school teacher? Hack reward By feeding the right answers, in the hope of improving their test scores?)

The founders of mechanics are not naive about the difficulty of automating jobs in this way. Mr. Parnet told me that his best appreciation is that full automation will take 10 to 20 years. (Mr. Ardel and Mr. Pesiroglu expect that it takes 20 to 30 years.)

These are conservative timelines, according to Silicon Valley standards. And I can, unlike many artificial intelligence companies that work on the technique of the back of the back behind the closed doors, Mechandize is explicit about what you are trying to do.

But I also found that the stadium is strangely devoid of sympathy for people who are trying to replace them, and are not interested in whether society is ready for such a deep change.

Mr. Pesiroglu said he believes that artificial intelligence will eventually create a “radical abundance” and the wealth that can be redistributed to the exploited workers, in the form of a global basic income that allows them to maintain a high standard of living.

But like many artificial intelligence companies that work on a technique of maintaining employment, there is no new policy proposals in a mechanic to help softening the transition to the economy that is driven by artificial intelligence, and there are no great ideas about expanding the social safety network or re-training workers to obtain new jobs-just a goal to make current jobs as old as possible.

At one point during a question and answer, I made the pipes to ask: Is it morally automated all the labor?

Mr. Barnett, who described himself as liberal, answered that he was. He believes that artificial intelligence will accelerate economic growth and stimulate life -saving breakthroughs in medicine and science, and that the prosperous society with full automation will be better for a low -growth economy as humans still have functions.

“If society becomes richer, then I think this surpasses the aspects of the people who lose their jobs,” said Mr. Barnant.

Hey, at least they are honest.