Hinton, Li, and Ng at Ai4 2026: The Real AI Battle Isn't Open vs. Closed — It's Who Gets to Decide

In August 2026, at the Ai4 conference held at The Venetian in Las Vegas, three of AI's most respected figures — Nobel laureate Geoffrey Hinton, World Labs CEO Fei-Fei Li, and Coursera co-founder Andrew Ng — shared a stage for a session titled "The Architects of Intelligence: A Historic Convergence." Moderated by Yun-Hee Kim, the panel was expected to be a clash over existential risk. What emerged was far more interesting: a unified rejection of centralized AI control hiding deep, unresolved disagreements about how openness should actually work .

The Shared Case: Power Concentration Is the Real Danger

All three speakers rejected the idea that locking down AI inside a handful of well-capitalized labs is the right answer to safety concerns. Their core shared argument: concentrating power in a few companies is a greater danger than the risks of open distribution. They worried that gatekeeping would slow innovation, entrench incumbents, and let a few firms decide what the rest of the world can build .

Yun-Hee Kim framed the session as a historic convergence, and on this foundational point, the three pioneers did converge. But the moment the conversation moved from principle to practice, the consensus fractured.

The Five Points of Disagreement

1. Open-Weight Models: A Distinction That Divides

Geoffrey Hinton drew a sharp line between open-source code and open-weight models. "Open source is great. You show people the code, and lots of people look at the lines of code and say, 'Oh, there's a bug.' Open weights means you train a big model and then you give people the weights. That's very different," he said. He warned that released weights make fine-tuning for cyberattacks cheap and easy — but also conceded the battle is effectively lost: "I think that battle's been lost… It's too late" .

Andrew Ng was far more comfortable with open-weight distribution. He framed it as a competitiveness issue and noted that when his team needed to run a security review on their open-source agent tool OpenWorker, they turned to Chinese open-weight models like Kimi K3 and GLM-5.2 because closed US models refused to help. "From what I'm seeing, I think open-weight models seem safer to me than closed-weight models," Ng said .

Fei-Fei Li rejected the binary entirely, calling the open-versus-closed framing a false dichotomy. She argued for a more nuanced, sector-specific approach rather than treating openness as an all-or-nothing choice .

2. Gatekeeping: Ng's Central Fight

Ng was the most vocal on this point: "I don't want there to be gatekeepers. That limits how all of us can access AI." He accused large companies of hyping AI dangers specifically to lock in regulatory advantages and freeze out smaller competitors. His prescription: "promote openness… because AI is amazing technology and I want it to be in everyone's hands" .

Hinton and Li both agreed that a handful of labs should not control AI's trajectory, even though Hinton was far more worried about the risks of unfettered distribution .

3. Competition with China: An Uncomfortable Paradox

Ng explicitly framed openness as a soft-power and competitiveness issue against Chinese open-weight models. He argued that if the US restricts open models, China's rapidly improving open-weight ecosystem — including Alibaba's Qwen 3.8-Max and Zhipu's GLM-5.2 — will leap ahead . He pointed out a paradox: Chinese open models were actually helping US security researchers do work that American closed models blocked .

Hinton acknowledged the competitive pressure but remained more worried about misuse. He saw the genie as already out of the bottle regardless of US policy .

4. Regulation: How Much, and Who Benefits?

All three agreed some regulation is necessary, but they disagreed sharply on scope and approach .

• Hinton called for mandatory safety testing before release and full disclosure of results to government — he wants a minimum baseline enforced by regulators . He described regulation as a "steering wheel, not the brakes" of a car .
• Li pushed for sector-specific regulation rather than blanket rules, arguing that AI reshapes tasks differently across industries and that productivity gains won't automatically translate into shared prosperity .
• Ng was the most skeptical of top-down regulation. He warned it could be captured by incumbents and used to entrench their advantages. He preferred promoting competition and openness as the primary safeguard .

5. Is Openness an All-or-Nothing Choice?

Here the panel exposed its deepest division. Li explicitly called the binary framing false — she argued for a middle path where openness is calibrated by sector, risk level, and use case . Hinton effectively agreed it isn't binary, but from the opposite angle: he wanted a world where weights could be open with safety interlocks, even if he no longer believes that world is achievable . Ng leaned closest to an all-or-nothing posture — openness was the central remedy, and he resisted carve-outs that could be exploited by incumbents .

Bottom Line: The Same Destination, Different Maps

The three pioneers stood united against a future where a handful of labs control AI, but they split sharply on the vehicle. Hinton reluctantly accepted open weights as inevitable while fearing their misuse. Ng embraced openness as the primary cure for concentration of power and a strategic necessity against China. Li rejected the binary entirely and called for calibrated, sector-specific rules. The debate captured the honest tension in AI governance in 2026: the same open distribution that democratizes access also makes catastrophic misuse harder to contain.