For external observers, researchers on artificial intelligence are found in an enviable position. They are sought by the giants of technology. Dizzying salaries are taking home. And they are part of the most popular sector of the moment.
But all this involves strong pressure.
More than half a dozen researchers with whom Techcrunch talked, some of whom asked for anonymity for fear of retaliation, said that the dizzying rhythm of the artificial intelligence sector has put their mental health to the test. The fierce competition between artificial intelligence laboratories has fomented an atmosphere of isolation, they say, while the growing stake has increased stress levels.
“Everything has changed practically from one day to the next”, a researcher told me, “with our work – both the positive and negative results – which have had enormous impacts measured by things such as the exposure of the product and the consequences financial “.
Just last December, Openii hosted 12 live streaming during which he announced over a dozen new tools, models and services. Google replied with its own tools, models and services in a dizzying series of press releases, social media and blog posts. The exchange between the two giants of technology has been remarkable for its speed, a speed that according to researchers has a high cost.
Grind and bustle
Silicon Valley is not extraneous to the worst culture. With the boom of artificial intelligence, however, public support for the superlavoro has reached worrying levels.
In Openai, it is not uncommon for the researchers to work six days a week and well beyond working hours. The CEO Sam Altman pushes the company’s teams to transform the discoveries into public products according to exhausting deadlines. According to reports, the former manager of the research of Openai, Bob McGrow, mentioned Burnout as one of the reasons why he left the company last September.
There is no relief to find in the competing workshops. The Google Deepmind team that developed Gemini, the leading series of IA models of Google, at some point went from 100 hours a week to 120 hours to correct a bug in a system. And the engineers of Xai, the artificial intelligence company of Elon Musk, regularly publish posts on the working nights that extend until the early hours of the morning.
Why this incessant push? Today research on artificial intelligence can have a considerable impact on the profits of a company. Google’s parent company, Alphabet, has lost about 90 billion dollars in market value due to the aforementioned bug, which led Google’s Gemini chatbot to generate controversial representations of historical characters.
“One of the major pressures is competitiveness,” said Kai Arulkumaran, research manager at the Araya artificial intelligence service provider, “in combination with quick times”.
Mainly rankings
Part of this competition takes place very publicly.
On a monthly basis, and sometimes weekly, artificial intelligence companies aim to replace each other in the rankings such as Chatbot Arena, which classifies artificial intelligence models in categories such as mathematics and programming. Logan Kilpatrick, product manager for various Google Gemini development tools, said in a X post chatbot Arena “he had a non -trivial impact on the speed of the development of artificial intelligence”.
Not all researchers are convinced that it is a good thing. The speed of the sector is such, they say, that their work risks becoming obsolete even before it can be sent.
“This leads many to question the value of their work,” said Zihan Wang, a robotics engineer who works in an invisible artificial intelligence startup. “If there is a huge probability that someone will go faster than me, what is the meaning of what I’m doing?”
Other researchers complain that the attention to productivity has taken place at the expense of academic skull.
“One of the underlying causes of stress is the transition of researchers on artificial intelligence from pursuing their research programs in industry to going to work on (artificial intelligence models) and the supply of solutions for products,” said Arulkumaran . “The industry expected that artificial intelligence researchers could carry on academic research in industry, but it is no longer so.”
Another researcher has said that, with his great consternation and anguish, the open collaboration and the discussions on research are no longer the norm in industry, with the exception of some artificial intelligence laboratories that have embraced the opening as a release strategy.
“Now we focus more and more on marketing, on the reduction of Closed-Source and on the execution,” said the researcher, “without contributing to the scientific community”.
Manage the graduate challenge
Some researchers trace the seeds of their anxiety to their degree programs in artificial intelligence.
Gowthami Somepalli, a doctoral student who studies artificial intelligence at the University of Maryland, said that research is published so quickly that it has become difficult for graduate students to distinguish between passengers and significant developments. This matters a lot, said Somepalli, because it has seen artificial intelligence companies give candidates more and more priorities with “an extremely relevant experience”.
“A research doctorate is generally a rather insulating and stressful experience, and a doctorate of automatic learning research is particularly demanding due to the rapid progression of the field and the ‘public or perisper’ mentality,” said Somepalli. “It can be particularly stressful when many students in your laboratory publish 4 articles while you publish only 1 or 2 per year.”
Somepalli said that, after the first two years of his degree course, he stopped taking his holidays because he felt guilty for having moved away before having published any study.
“I constantly suffered from the imposter syndrome during my doctorate and I almost abandoned my studies at the end of the first year,” he said.
The path to follow
So what changes, if there are, could favor a work environment based on less punitive artificial intelligence? It is difficult to imagine that the rhythm of development slows down, not with so much money at stake.
Somepalli underlined small but of great impact reforms, such as normalization that gives voice to its challenges.
“One of the major problems … is that nobody openly discuss their struggles; Everyone takes on a courageous face, “he said. “I think (people) could feel better if they could see that others are struggling too.”
Bhaskar Bhatt, artificial intelligence consultant at the EY professional service company, says that the industry should work to build “sturdy support networks” to combat insulation feelings.
“It is essential to promote a culture that enhances the balance between work and private life, in which individuals can really disconnect from their work,” said Bhatt. “Organizations should promote a culture that enhances mental well -being as much as innovation, with tangible policies such as reasonable working hours, mental health days and access to consultancy services”.
Ofir Press, a post-dictate student in Princeton, has proposed a fewer number of conferences on artificial intelligence and “pauses” of a week on the presentation of documents so that researchers can take a break from the monitoring of the new work. And Raj Dabre, artificial intelligence researcher at the National Institute of Information and Communication Technology in Japan, says that researchers should be reminded kindly what is really important.
“We have to educate people from the beginning on the fact that artificial intelligence is only work,” he said as a dabre, “and we must focus on family, friends and the most sublime things of life”.
