Three of AI's best-known researchers used a rare joint appearance at the Ai4 conference in Las Vegas to draw sharply different lines around how advanced AI should be released, regulated and used at work, Data Center Knowledge reported.

Geoffrey Hinton, Fei-Fei Li and Andrew Ng agreed that AI will reshape nearly every industry.

They split over the questions that now matter to companies building and financing AI capacity: how soon models could exceed human intelligence, whether they will replace routine intellectual labor, whether open-weight models spread opportunity or risk, and how far governments should go before more capable systems are deployed.

The session, billed as “The Architects of Intelligence: A Historic Convergence,” connected a policy dispute to an infrastructure buildout already under way.

Data centre developers, utilities and investors are committing capital on the assumption that demand for AI compute will keep rising for years.

The panel showed that the deployment rules behind that demand remain unsettled.

Hinton gave the most urgent warning.

He said artificial intelligence could surpass human intelligence within five to 20 years and argued that jobs built mainly around routine intellectual labor could be automated.

“If AI can do routine intellectual labor, any job that consists mainly of routine intellectual labor is going to be done by AI,” Hinton said.

His concern extended beyond employment.

Hinton warned that more capable AI systems could enable cyberattacks, concentrate power among leading AI companies and require stronger government oversight before deployment.

Ng challenged the premise that AI is already producing broad job losses.

He described current systems as tools that automate tasks and make workers more productive, rather than replacements for whole professions.

In that framing, the operating question for companies is not whether entire occupations disappear at once, but how jobs change as employees use AI to take on broader responsibilities.

Li pushed for a less polarized debate.

She said public discussion has become dominated by extreme narratives that obscure practical questions about deployment, education and policy.

Rather than treating AI as a single category needing sweeping restrictions, she argued that existing regulatory frameworks should be updated in sectors such as healthcare, transportation, finance and education.

The clearest policy conflict came over open-weight models.

Ng defended them as necessary for innovation, competition and broader access, warning against a future in which a small number of companies control advanced AI technology.

Hinton argued that releasing model weights could make it easier for malicious actors to adapt powerful systems for cyberattacks and other harmful uses.

Li rejected a simple open-versus-closed framing.

Different applications, she argued, require different levels of openness depending on risk.

She also called for greater public investment in AI research and education, describing AI as foundational infrastructure whose long-term development should not be driven only by private companies.

For AI infrastructure planning, the result is a less uniform deployment path.

Closed frontier systems, open-weight models and sector-specific deployments may each bring different compute, compliance and security requirements.

The panel reached no consensus, leaving guardrails, access rules, workforce adaptation and public research investment as policy variables that could shape how quickly financed AI capacity turns into production workloads.