the lost game of the university against the private sector

Lab life. “Such growth, in such a short time, is unheard of in any field”, observes Neil Thompson. This specialist in artificial intelligence (AI) and economics at the Machusetts Institute of Technology (MIT, Boston) does not talk about the crazy growth of users of the ChatGPT talking machine developed by OpenAI. Nor even the media omnipresence of the concept of artificial intelligence since the release, in November 2022, of this conversational agent and its derivatives or applications. It evokes a deeper and more worrying fact: the clear imbalance between the academic world and the industrial world in the field of artificial intelligence. The second folded the game in just ten years.

In the American scientific journal ScienceMarch 2, with Nur Ahmed, also at MIT, and Muntasir Wahed (Virginia Tech University, Blacksburg), Neil Thompson quantified this victory. In 2020, in the United States, the overwhelming majority of doctoral students in AI, 70%, were hired by the private sector, whereas in 2004 the proportion was only 20%. The researchers also found that in 2018 the poaching of professors by large companies had increased by 10%, while hiring at the university has remained stable for fifteen years.

On the front of computing resources, same imbalance. Industry “models”, i.e. programs trained on vast amounts of data, such as ChatGPT, Bard, Dall-E, etc., are now, with hundreds of billions of parameters, thirty times bigger than those in the academic world. However, in many cases, a larger size is synonymous with greater quality.

As a result, in 2020, 40% of presentations at conferences come from private laboratories, double what they represented in 2012. In their annual report on the subjectin April, their colleagues at Stanford made similar observations: the industry had thirty-two models “important” against three in the academic world. In terms of publications, the report noted that, even on the ethical aspect of artificial intelligence, companies “submit more than ever”. For example, three times more in 2022 than in 2021 and their quantity is equivalent to a third of what the academic world produces.

“Even at MIT, we can no longer fight”

Brief, “on the three key parameters for the success of AI models, data, computing power and human resources, it is unbalanced”summarizes Nur Ahmed, who recalls that in 2021 companies spent 340 billion dollars on artificial intelligence when American funding agencies had supported research for 1.5 billion dollars that year. “Even at MIT, you hear colleagues say they can’t fight anymore”testifies Neil Thompson.

You have 35.6% of this article left to read. The following is for subscribers only.

Source link

Leave a Reply