I have not encountered such a misleading argument about artificial intelligence from a major political figure in a long time.
Barack Obama is right about one thing: AI companies want to make money. Of course they do. They are commercial companies with investors, enormous infrastructure costs and valuations to justify.
But reducing the development of agentic AI to the need to create something consumers will pay for misses the much larger technological transformation taking place.
Obama recently argued that if we want AI to cure cancer or develop better energy technologies, we can pursue those goals without allowing agentic AI to roam freely across the internet. According to him, companies are pushing toward such systems because they need marketable products.
That framing deserves serious scrutiny.
An AI that answers questions is not the endpoint
The distinction between today’s assistants and genuinely agentic systems matters.
A conventional AI can analyze information when prompted. An agent can potentially formulate a goal, search for relevant information, use software and databases, coordinate specialized models, test possible solutions, evaluate the results and continue working.
That difference could become enormous.
Scientific discovery itself is an iterative process: find information, formulate hypotheses, test them, analyze the results and modify the hypotheses. Researchers are already experimenting with AI systems that automate significant portions of precisely this cycle.
This is no longer speculation. A 2026 Nature paper described Robin, a multi-agent system capable of automating hypothesis generation and data analysis in experimental biology. Researchers used it to identify potential therapeutic candidates for age-related macular degeneration. Another Nature study introduced an AI “Co-Scientist” in which multiple agents generate, criticize and refine scientific hypotheses.
Medical researchers are separately investigating increasingly autonomous medical AI agents.
The important question, therefore, is not merely whether an AI can tell us something about cancer. It is whether increasingly autonomous systems can perform substantial portions of the intellectual work required to discover something about cancer.
Those are very different propositions.
Revolutions in nearly every field. Does he care?
The same principle applies across science and engineering.
Agentic systems could eventually coordinate simulations, analyze gigantic datasets, examine scientific literature, propose experiments, write and test software, operate research tools and collaborate with other specialized agents.
The potential applications extend to medicine, energy, climate science, materials science, engineering and many other fields. The UN’s scientific advisers already describe AI as an increasingly important enabling technology for health, sustainability, climate resilience and other global challenges.
AI will produce in years discoveries that otherwise would have taken humanity 1,000 years.
But the underlying possibility is real: if AI substantially accelerates the cycle of hypothesis, experimentation and refinement, scientific progress itself could accelerate. Technologies that automate parts of that process already exist.
Treating autonomy principally as a commercial gimmick therefore understates what is at stake.
Governments want advanced AI too
There is another problem with the idea that autonomy is fundamentally being driven by companies trying to sell products.
Governments themselves have enormous strategic incentives to obtain increasingly capable AI.
The current U.S. national-security strategy explicitly calls AI one of the most transformative national-security technologies in American history and directs the government to accelerate its adoption across intelligence and warfighting. It also calls for close partnerships with private companies to make frontier models available to national-security personnel.
It also involves scientific institutions, militaries, intelligence agencies, governments and geopolitical competitors.
AI capability is becoming economic power, scientific power and military power.
Commercial incentives are certainly part of this story. They are nowhere near the whole story.
Obama is right about the danger
Ironically, none of this means Obama’s concern about autonomous AI should be dismissed.
Quite the opposite.
The more useful autonomous agents become, the more seriously their risks must be treated. NIST reported this year that respondents to its consultation broadly agreed that AI agents create novel security threats requiring adaptations to conventional cybersecurity practices.
An autonomous system capable of accelerating scientific research could also accelerate cyberattacks or other harmful activities. A system capable of operating computers for legitimate purposes could potentially operate them for malicious ones.
Power and risk increase together.
So there is a perfectly serious argument for regulation, testing, containment, monitoring and restrictions on what autonomous AI systems may access.
But that argument is very different from suggesting that companies are developing agentic AI mainly because they need another product to sell.
The real debate is much harder
Obama’s broader concern — that commercial incentives can conflict with public interests — is legitimate. Companies can have incentives to deploy systems before society has adequately understood their consequences. Even Obama has described himself as occupying a middle position rather than being either an AI “accelerationist” or “doomer.”
But society should not misunderstand why increasingly autonomous AI matters.
We are potentially building systems that do not merely answer questions but perform increasingly long chains of intellectual and practical work.
The consequences could reach medicine, science, energy, economics, national security and warfare.
Profit is one incentive.
It is not the reason the technology matters.
And reducing this technological race to Silicon Valley companies desperately searching for another product to monetize risks obscuring the much bigger question:
What happens when machines become capable not merely of giving us answers, but of independently searching for answers that humanity has not yet discovered?
Leave a Reply