Three of the most well known names in AI research shared the stage at the Ai4 conference in Las Vegas last week. Geoffrey Hinton, Fei-Fei Li and Andrew Ng spoke about open source AI and whether it helps or hurts the industry.
The topic has become a point of tension. Open weight models let anyone download and use a trained AI system without much oversight. Some labs see this as a safety problem.
Andrew Ng Warns Against Gatekeepers
Ng focused on what happens when a small number of firms control access to a technology. He compared it to how Apple and Google control mobile phone operating systems.
"I don't want there to be gatekeepers," Ng said. He argued this kind of control limits how people can use AI.
Ng said the best path forward is to keep multiple companies and models competing. He said openness lets more people benefit from AI rather than just a few large firms.
He also raised concerns about global competition. Ng said if China's open weight models spread across Africa and other developing regions, they could shape how billions of people learn about ideas like democracy and human rights.
"AI is a tremendous source of soft power," Ng said. He added that lobbying and fear in the United States could slow down American open source AI efforts.
Hinton Draws a Line Between Open Source and Open Weights
Hinton separated open source software from open weight AI models. He said open source lets people inspect code line by line and fix bugs.
Open weights are different, he said. Companies release the trained parameters of a model without sharing how it was built.
"That's very different," Hinton said. He explained this makes it easier for people to retrain large models for harmful purposes like cyber attacks.
Even so, Hinton said the shift toward open weight models has already happened. He said the high cost of training a model used to be a barrier, but that barrier is gone.
"I think that battle's been lost," Hinton said. He still believes AI overall will bring benefits like better healthcare and education.
Li disagreed with framing the issue as a simple choice between open and closed systems. "It's very dangerous to make this a dichotomy," she said.
She compared AI to nuclear physics, where scientific papers are published openly but materials like uranium are tightly controlled. Lab work sits somewhere in between those two extremes.
Li also pointed to the Human Genome Project as an example of open collaboration. She said the shared data allowed drug companies to profit while helping scientists and the public at the same time.
"We need some levels of openness," Li said. She added that closed systems still have a place alongside open ones.
Despite their disagreements on methods, all three researchers agreed that regulation will play a role in how AI develops. Hinton said the goal should be building AI that helps people.
"You can't leave it to people like Elon Musk and Mark Zuckerberg to decide how AI should be done," Hinton said."You can't leave it to people like Elon Musk and Mark Zuckerberg to decide how AI should be done," Hinton said.
