Years ago, when I started writing for Silicon Valley’s efforts to replace artificial intelligence workers, most technicians should have lied about it.
“We do not automate workers, we are.” extension Jim, “the executives told me.
Lines, such as those that were often supposed to assure nerve workers and provide plans for enterprises – of course, said more about the limitations of technology than the motifs of executives. At that time, AI was not good enough to automate most jobs, and it was definitely unable to replace university -educated workers in the white collars such as Tech, Consulting and Finance.
This is beginning to change. Some of today’s AI systems can write software, create detailed research reports and solve complex mathematical and scientific problems. Newer Ai Ai “AI agents are able to perform long sequences of tasks and control his own work as a person. And while these systems are still not reaching people in many areas, some experts are afraid that the recent increase in university graduates is a sign that companies are already using AI as a compensation for some basic level workers.
On Thursday, I looked at the future after working at an event organized in San Francisco by Mechanize, a new beginning AI, who has a bold goal to automate all jobs-my voice, mine, such, our doctors and lawyers, people who write our software and design our buildings and take care of our children.
“Our goal is to fully automate the work,” said Tamay Besiroglu, 29, one of the founders of the mechanize. “We want to get to a fully automated economy and do it as quickly as possible.”
The dream of full automation is not new. John Maynard Keynes, an economist, at the age of 30 predicted that machines automate almost all jobs, create a lot of material and leave people free to devote themselves to their passions.
Of course, this has never happened. However, recent advances in AI have ruled the belief that technology capable of mass automation is close. Dario Amodei, CEO of Anthropic, He recently warned that AI could move up to half of all basic jobs on a white collar over the next five years.
The mechanize is one of the many start -ups that allow this. The company was founded this year by Mr. Besiroglu, EGE Erdil and Matthew Barnett, who worked in the research company EPOCH AI, which studies the capabilities of AI systems.
Has attracted investment From well -known technological leaders, including Patrick Collison, founder Stripe and Jeff Dean, the main scientist Google. He now has five employees and works with leading AI companies. (He refused to say which of them, referring to confidentiality agreements.)
The access of AI task automation is focused on a technique known as learning amplification – the same method that was used to practice a computer to play board game at a superhuman level almost ten years ago.
Today, AI leader uses learning to improve the outputs of their language models by making further calculation before creating an answer. These models, often called “thinking” or “reasoning” models, have become impressively well in some close tasks, such as writing code or solving mathematical problems.
However, most jobs include more than one task. And today’s best AI models are still not reliable enough to handle more complicated workload or navigate comprehensive business systems.
To solve this, the mechanism creates a new training environment for these models – essentially complicated tests that can be used to teach models, what to do in a given scenario, and assess whether they managed or not.
For example, to automate software engineering, a mechanization creates a training that resembles a computer that a software engineer would use – a virtual machine equipped with e -mail inbox, a relaxed account, some encoding tools and a web browser. The AI system is asked to complete the task using these tools. If it succeeds, he gets the reward. If he fails he gets a punishment. Then they try again. If the simulation was well designed, with sufficient attempt and accidentally, AI should eventually learn to do what the human engineer does.
“It’s actually like creating a very boring video game,” Mr. Besiroglu said.
The mechanism starts with computer programming, which is a profession where the strengthening of learning has shown some promise. However, it hopes that the same strategy could be used to automate tasks in many other fields of white collar.
“We really know that we have succeeded when we created AI systems capable of taking almost every responsibility one could exercise on a computer,” company he wrote In a recent blog post.
I have some doubts about whether the access of the mechanism will work, especially for non -technical jobs where success and failure are not so easy to measure. (What would it mean, for example, to AI to succeed in that he is a high school teacher? Hack rewards By cringing students the right answer, in the hope that they will improve their test scores?)
The founders of the mechanism are not naive about the difficulty of automation of tasks in this way. Mr. Barnett told me that his best estimate was that the full automation would take 10 to 20 years. (Mr. Erdil and Mr. Besiroglu expect it to last 20 to 30 years.)
These are conservative timelines according to Silicon Valley standards. And I appreciate that, unlike many AI companies working on technology, by settling work behind closed doors, the mechanism is sincere about what they are trying to do.
But I also found that their playground strangely lacks empathy for people whose employment is trying to replace and is not interested in whether the company is ready for such a deep change.
Mr Besiroglu said he believed that AI eventually creates “radical abundance” and wealth that could be redistributed to released workers, in the form of a universal basic income that would allow them to maintain a high standard of living.
But like many AI companies that work on the technology of inventing work, the mechanization has no new political proposals that would help to smooth out the transition to the economy of controlled AI, no brilliant ideas of expanding the social security network or retraining workers for new jobs-the objective to make current jobs as soon as possible.
In one place during the questions and answers, I asked: Is it ethical to automate all the work?
Mr. Barnett, who described himself as a libertarian, replied that he was. It believes that AI will speed up economic growth and stimulate breakthroughs to save life in medicine and science, and that a prosperous society with full automation would be more convenient than a low growth economy where people still had a job.
“If society as a whole becomes much richer, then I think it will only outweigh the disadvantage of people who lose their work,” Barnett said.
Hey, at least they’re honest.
