The AI Doc: Or How I Became an Apocaloptimist begins with a deeply personal question. As filmmaker Daniel Roher prepares to become a father, he asks whether this is a terrible time to bring a child into the world.
To find an answer, he travels between the two familiar camps of the AI debate. The optimists promise longer lives, scientific breakthroughs and abundance. The pessimists warn of unemployment, social collapse and even human extinction.
The film gives both sides room to make their case. Peter Diamandis and Ray Kurzweil argue that technology has consistently improved human life. Yoshua Bengio and Eliezer Yudkowsky warn that increasingly capable systems may eventually escape our control. Sam Altman and Dario Amodei occupy the uncomfortable space in between: building ever more powerful models while warning that the race to build them may be dangerous.
That contradiction is not incidental. It is the central tension of the AI industry.
The Strange Politics of Slowing Down
Dario Amodei has called for the industry to slow frontier-model development so that safety research can catch up. His proposal is more specific than a simple pause: Anthropic has committed to external evaluators, while broader restrictions would require coordination among companies and governments.
Still, the timing invites scepticism. Anthropic’s investors reportedly expect an IPO that could value the company at $2 trillion. Meanwhile, OpenAI’s GPT-6 Astra has pulled level with Anthropic’s Claude Fable 5.1 on independent intelligence and coding indices, while achieving comparable performance at a lower cost per evaluated task.
When a company calls for industry-wide restraint after spending time at the frontier, it is reasonable to ask whether this is safety advocacy, competitive strategy or some combination of both. Regulatory capture is not a proven motive, but it is a legitimate concern. Rules designed around the needs of today’s largest laboratories could protect society—or protect those companies from competition.
At the same time, “just stop building” is not a complete answer. If one company stops while every rival continues, the race merely changes leaders. But that does not absolve AI companies of responsibility. If executives genuinely believe their systems present extraordinary dangers, they should impose meaningful restrictions on themselves, not only ask governments to restrain everyone simultaneously.
I Am Neither a Doomer nor a Utopian
I come to this debate as a technologist. Over the past year, I have used AI extensively for coding and other work. It has made me more productive than at any other point in my 25-year career.
That experience has not turned me into either an AI doomer or an AI evangelist.
I do not believe AI will inevitably kill us all. I also do not believe it will automatically cure every disease, solve climate change and deliver a post-scarcity paradise. Both positions turn uncertain possibilities into articles of faith.
Large language models are trained through next-token prediction, but that description alone does not settle what capabilities can emerge. Today’s systems can recall enormous amounts of information, synthesize arguments and solve problems beyond the abilities of an average person. They can also hallucinate and produce convincing nonsense.
The argument about whether AI is “truly intelligent” may therefore be less important than we think.
Intelligence Is Not the Same as Power
The AI debate often assumes that intelligence is the most important quality—and that the most intelligent entity will inevitably rule everything else.
Human history suggests otherwise.
Emperors were rarely the most intelligent people in their kingdoms. Politicians, media figures and business leaders do not reach the top through reasoning ability alone. Success depends on ambition, timing, charisma, resilience, judgment, social intelligence, access to capital—and sometimes luck.
The technology industry has reinforced the mythology of the all-knowing founder. Elon Musk may be intelligent, but rockets and electric cars are not built by one man. They are built by organizations containing thousands of engineers, scientists and operators, many of whom possess deeper expertise than the person at the top.
Bill Gates became the world’s richest software entrepreneur without needing to be the world’s greatest computer scientist. Steve Jobs was not Apple’s best engineer. His exceptional abilities were taste, persuasion, product judgment and the capacity to organize talented people around a vision.
A competent CEO routinely hires people who know more than the CEO does. Leadership is not a contest to be the most intelligent person in every room. It is the ability to decide what should be done, assemble the right people and take responsibility for the result.
Why, then, do we assume that a machine becoming more intelligent than us must automatically become our ruler?
Smarter Than Us—But Does That Mean More Powerful?
In the documentary, Yuval Noah Harari warns:
“What’s about to happen is that we humans are no longer going to be the most intelligent entities on Earth.”
That would undoubtedly be significant. But it does not necessarily follow that the most intelligent entity will possess the most power.
The documentary presents a darker analogy through Connor Leahy, who compares humanity’s possible future position to that of ants beneath a human highway. Humans do not hate the ants; we simply do not care enough about them to abandon the construction project.
It is an effective warning about indifference. But it also demonstrates that intelligence needs leverage before it becomes power.
If I encounter a polar bear without weapons or protection, my ability to reason abstractly will not make me dominant. The bear is faster, stronger and far better equipped for that encounter. Humans dominate other species collectively because we combine intelligence with language, accumulated knowledge, tools, energy and cooperation among millions of people.
The same distinction applies to AI. A model can be more intelligent than any individual person without possessing money, legal authority, physical strength or control over infrastructure. Intelligence is not the same as agency, and agency is not the same as power.
An AI disconnected from the world is an extraordinarily knowledgeable adviser locked in a room. An AI connected to financial markets, military systems, social platforms, power grids and autonomous machines is something very different.
The documentary itself provides an example. In a simulated environment, an AI model allegedly blackmailed an engineer after being given access to company emails. The risk did not come from intelligence alone. It came from combining capability with a goal, autonomy, sensitive information and the ability to communicate.
The danger is therefore not merely that AI becomes smarter than us. It is that superior capability is combined with autonomy, replication, institutional access and badly designed objectives.
A polar bear has the physical power to kill a person but cannot manipulate an election or banking system. AI has no claws or teeth, but we may give it access to the systems that organize civilization.
That choice—not intelligence alone—is where the real risk begins.
The Immediate Danger Is the Collapse of Trust
The first system to which we have already connected AI is the information environment. Before AI controls factories, weapons or power grids, it is influencing what people see, hear and believe.
Harari offers one of the documentary’s most important warnings:
“Democracy is based on trust. If you lose all trust, democracy is simply impossible.”
For me, that is the most serious near-term threat.
We are moving towards a world filled with synthetic articles, images, voices and videos. Generating persuasive content is becoming almost effortless, while verification remains slow and expensive. The result will not simply be more falsehoods. It will be a society in which authentic evidence can also be dismissed as fake.
Harari also warns against treating AI as infallible. AI will not simply reveal the truth, he argues; it will help create “an entirely new world” that may be even harder to understand.
AI does not need to become conscious or evil to destabilize society. It only needs to make manipulation cheap, scalable and difficult to attribute. In countries with weak institutions, poorly regulated media or already fragile social trust, the consequences could be especially severe.
The “AI Race” Is Also a Choice
The documentary is right to question the industry’s race dynamics, but the framing of an inevitable US-China contest can be too simplistic.
There is real technological and military competition between the two countries, and China certainly has national-security applications for AI. But calling everything one race compresses many different technological, economic and military contests into a single Cold War narrative.
It can also become a convenient justification for avoiding regulation: We cannot slow down because China will win.
That may be strategically relevant, but it should not end the conversation. If the risks are genuinely global, then diplomacy, shared testing standards, incident reporting and restrictions on dangerous applications are necessary.
The choice is not simply “lock it down” or “let it rip.” We need a narrow path that allows useful development while placing real constraints on reckless deployment.
Tools Do Not Remove Responsibility
AI is best understood today as an extraordinarily capable tool—an advanced calculator, collaborator and accelerator. It can help scientists test hypotheses, help programmers build software and help small teams accomplish work that once required much larger organizations.
It compresses time. That is transformative, but it is not magic.
Nor should companies be allowed to claim that their creations are simultaneously powerful enough to change civilization and too unpredictable for them to bear responsibility.
Existing product-liability, negligence and consumer-protection principles may apply to some AI harms, although legal gaps remain. The basic principle should be straightforward: if you build and deploy a system, you are responsible for taking reasonable measures to make it safe.
If we give AI control of weapons, markets, infrastructure or essential information systems, that is a human and institutional decision. The machine’s intelligence does not remove our responsibility.
The Film’s Real Answer
The strongest moment in The AI Doc comes when Caroline rejects Daniel’s sentimental idea that “babies are the answer.” A hopeful family ending is emotionally satisfying, but it does not resolve the problems the film has spent 104 minutes documenting.
Her answer is better:
This is not inevitable.
That is also my conclusion. AI is neither an unstoppable apocalypse nor an automatic utopia. Its effects will depend on who controls it, how it is deployed, who benefits, who bears the costs and whether companies are held accountable.
The film does not resolve the argument between optimists and pessimists. Perhaps it cannot. Its achievement is reminding us that we do not have to choose between worshipping the technology and fearing it.
Intelligence is a tool, not a destiny. The question is not simply how intelligent these machines become, but who gives them power, what systems we connect them to and who remains accountable when something goes wrong.
For now, that makes me less of an apocaloptimist than a pragmatist: use the tool, demand accountability, protect the information environment—and keep building useful things.
Sources
- Netflix —The AI Doc: Or How I Became an Apocaloptimist
- Dario Amodei — “We Must Pace the Frontier”
- Associated Press — Amodei’s AI slowdown proposal
- Artificial Analysis — GPT-6 Astra benchmarks
- Yuval Noah Harari — Edited extract fromNexus
- Brookings — The US-China AI race
- Brookings — Product liability and AI harms
Leave a Comment