This Stanford computer teacher went to written exams 2 years ago because of the AI. He says his students insisted

But two years ago, when the last wave of artificial intelligence began to reshape education, said Leskvec Fortune He was shaken by the explosion of his field in the dominant current. He said that Stanford had such a prestigious computer program that he has the impression that he “sees the future as he was born, or even before the future was born”, but the public liberation of GPT-3 was shocking.

“We had a great existential crisis, among the students a few years ago, when it was not clear what is our role in this world,” said Leskocc.

He said that it seemed that the breakthroughs in the AI ​​would be exponential to the point where “it will simply do research for us, when we do?” He said that he had spent a lot of time talking with students at the doctorate level on how to organize himself, even on the fact that their role in the world in the future. It was “existential” and “surprising”, he said. Then he received another surprise: a request led by students for a test change.

“He came out of the group,” he said, in particular educational assistants, the previous generation of undergraduate students in computer science. Their idea was simple: “We do a paper exam.”

AI as a catalyst for change

Leskovec, an eminent researcher in Stanford, whose expertise lies in the structured data of graphics and AI applications in biology, told the pivot with a mixture of surprise and reflection. Historically, his courses were based on open -air examinations, where students could take advantage of manuals and the Internet. They could not use the code and solutions of others, but the rest was a fair game. Like models of large languages ​​such as GPT-3 and GPT-4 of Openai exploded on the stage, students and educational assistants have started to wonder if the assessments should be managed differently.

Now, it’s much more work for him and his heaps, he said, saying that these exams take “much more time” to note. But they all agreed that it was the best way to test the knowledge of students. The age of AI for Leskovec, an AI veteran, surprised him by putting a higher workload on himself and other humans. In addition, there are “fewer trees in the world” of all the paper it prints, he said that AI had just created “additional work”. His lessons in 400 people feel like an audience during a “rock concert”, but he insisted that he did not turn to AI to help synthesize and analyze all the exams.

“No, no, no, we have the hand,” he insisted.

A student -based solution

The Leskvec solution is squarely in the middle of an unleashed debate on the way in which AI modifies higher education, because the rampant cheating reports have led many colleges to prohibit the use of AI. Other teachers come back to the paper exam, restart the famous blue books from many memories of the 1990s of the school. A professor at New York University even suggested obtaining a “medieval”, embracing the old forms of tests such as oral and written examination. In the case of Leskovec, the solution of the AI ​​teacher for the age of AI is also to turn away from AI for tests.

When asked if he was worried that students cheat with AI, Leskvec asked another question: “Are you worried that students cheat with calculators?” Admiring the AI ​​to a calculator, he said that AI is an incredibly powerful tool that “has simply emerged and surprised us all”, but it is also “very imperfect … We must learn to use this tool, and we must be able to test humans able to use the tool and humans able to think by themselves.”

What is a skill AI and what is a human competence?

Leskovec is struggling with a question that affects everyone on the labor market: what is a human competence, what is a competence of AI and where do they merge? MIT professor David Autor and SVP Google James Manyika discussed in The Atlantic Tools such as a calculator or an AI generally distribute in two buckets: automation and collaboration. Think of the dishwasher, on the one hand or a word processor, on the other hand. The collaboration tool “requires a human commitment” and the problem with AI is that it “is not careful either (bucket)”.

The job market sends a message on the implementation of the AI ​​which is equivalent to something like a response from the magic ball 8: “Reply misty. Reverse again later. ” The federal job report has revealed anemic growth since the spring, more recently disappointing expectations with an impression of only 22,000 jobs in August. Most economists attribute the lack of hiring to the uncertainty concerning the pricing regime of President Donald Trump, which several courts ruled illegal and seems to be heading towards the Supreme Court. But the implementation of the AI ​​does not go well at the company level, with a MIT study (not linked to the highway) noting that 95% of the generative pilots of the AI ​​fail, followed shortly by a study by Stanford finding the start of a collapse in hiring at the entrance level, in particular in jobs exposed to AI automation.

For another perspective, the independent market Upwork has just launched its inaugural monthly job report, revealing what timely jobs are rewarded by the market. The answer is that “IA skills” are super on demand and, even if companies do not hire full -time employees, they accumulate in very paid and highly qualified independent work.

Despite a softer global labor market, Upwork finds that companies “take advantage of flexible talents to fill the temporary gaps in the workforce”, large companies resulting in 31% growth in what Upwork calls for high value work (contracts greater than $ 1,000) on the platform. Smaller and medium -sized companies accumulate in “AI skills”, AI and automatic learning demand jumping 40%. But Upwork also sees growing demand for the type of skills that are between the two: a human who is good to collaborate with AI.

Upwork says that AI is “to amplify human talents” by creating a demand for expertise in greater value work, the most visible in creative and design, writing and translation categories. One of the main skills hired in August was verification of the facts, given “the need for human verification of the AI ​​results”.

Kelly Monahan, Director General of the Upwork Research Institute, said that “humans are coming back immediately” to work with AI.

“We actually see human skills of superior quality,” she said, adding that she thinks that people realize that Hallucine too much time to completely replace human participation. “I think what people see, now that they use content generated by AI is that they need to verify the facts.”

By extending this line of thought, Monahan said that the evolutionary landscape of “IA skills” shows what it calls “the expertise of the domain” is increasingly precious. Legal is a category that increased in August, she said, stressing that legal expertise is necessary to verify the facts in legal drafting generated by AI. If you do not have advanced skills in a particular field, “it is easy to deceive yourself” by the content generated by AI, and companies hire to protect themselves against this.

Leskvec agreed when he was questioned about the skills gap which seems to face entry -level workers trying to be hired, on the one hand, and companies that have trouble implementing AI effectively.

“I think we almost need to restart the labor market. Human expertise has much more than ever (before). ” He added that the entry-level problem is “the node of the problem”, because how are young workers supposed to obtain the expertise of the field required to collaborate effectively with AI?

“I think it comes back to teaching, to reskulling, to rethink our programs,” said Leskvec, the addition of colleges has a role to play, but organizations do it too. He asked a rhetorical question: how are they supposed to have qualified workers for seniors if they do not take young workers and do not take the time to train them?

When asked by Fortune To study the landscape and assess where we are currently using AI, as students, teachers and workers, Leskvec said that we were “very early in this area”. He said he thought we are in the “to come with solutions”. Solutions such as a hand -graduated examination and a teacher find ways to verify the knowledge of his students.