Responsible AI may prove to be one of the defining challenges of the AI age. As artificial intelligence grows more capable and more deeply woven into everyday life, the question is no longer simply what AI can do.
It is whether we can keep AI as a tool that serves us without gradually slipping into the habit of letting it think for us.
Technology ethicist Tristan Harris believes the future is still ours to shape. As co-founder of the Center for Human Technology, he argues that responsible AI is not an inevitable result of technological progress. Rather, those outcomes depend on the choices we make today about incentives, accountability, transparency, and, above all, restraint.
How important is it that we get those choices right?
Vastly important.
Few questions are more consequential. Yet many people assume the trajectory of AI is already set—that increasingly powerful systems will simply continue advancing as quickly as markets and technology allow.
Harris challenges that assumption. He believes we cannot afford to accept what he calls the myth of inevitability.
Harris first gained prominence by warning that social media’s race for attention would deepen political polarization, undermine well-being, and erode public trust. Looking back, that forecast was remarkably accurate.
Now, in his 2024 TED Talk, Why AI Is Our Ultimate Test and Greatest Invitation, he argues that we may be repeating the same pattern with artificial intelligence—only this time with technology powerful enough to reshape nearly every aspect of society.
His central question is both simple and unsettling:
What if the greatest danger isn’t what AI can do, but what we convince ourselves we can no longer stop?
A power without precedent
According to Harris, AI’s potential is unlike anything humanity has created before.
Imagine, he says, “a new country shows up on the world stage, and it has a million Nobel Prize-level geniuses in it.” Plus, these geniuses never sleep, work at superhuman speed, and cost less than minimum wage. “That is a crazy amount of power,” he concludes.
Applied wisely, such power could unlock extraordinary abundance: life-saving medical discoveries, cleaner energy, scientific breakthroughs, and innovations we can scarcely imagine.
Those possibilities tell only half the story.
The more difficult question is what becomes most likely once unprecedented capability collides with market incentives, competitive pressure, and the race to deploy ever more powerful AI systems.
Will organizations consistently choose caution over speed? Will transparency, accountability, and public safety receive the same priority as innovation and profit? Or will the strongest incentives continue rewarding whoever moves first?
Those are the questions at the heart of Harris’s argument—and the beginning of his case for restraint.
What “the probable” could look like
Harris distinguishes between what AI could do and what it is most likely to do under today’s incentives.
AI could help cure diseases, accelerate scientific discovery, improve education, and unlock cleaner sources of energy. Those possibilities are real.
But Harris asks us to look beyond what is technically possible and examine what markets are most likely to reward. Powerful technologies rarely shape society on technical merit alone. They also respond to incentives—financial, political, and social.
What might those incentives produce?
One plausible example illustrates the point.
Online engagement without guardrails
Imagine a video platform whose success depends almost entirely on keeping viewers engaged. Now give that platform an AI capable of generating convincing videos, cloned voices, and realistic deepfakes in seconds.
On platforms with strong moderation, clear disclosures, and reliable methods of tracing where content originated, much of this material can be identified or removed before it spreads widely. But on faster-moving networks with weaker safeguards, the incentives look very different.
The recommendation algorithm quickly learns a simple lesson.
Outrage attracts attention.
Novelty keeps people watching.
And attention becomes profit.
The result could be an endless stream of personalized deepfakes: a fabricated celebrity confession, breaking “news” delivered by someone who doesn’t exist, or a friend’s cloned voice sharing a believable half-truth. Each clip is designed to capture attention just a little longer than the last.
As advertisers follow viewers, the financial rewards increasingly flow to the content that generates the most clicks and watch time—not necessarily the content that is most accurate or trustworthy.
In other words, the system rewards engagement more reliably than truth.
This isn’t a prediction that every platform will evolve this way. Many companies are investing heavily in safeguards, provenance tools, and content authentication. Harris’s point is more fundamental: unless incentives reward responsibility as strongly as they reward engagement, the pressure to cut corners will remain.
And that is precisely why restraint matters.
The safety net that never lets go
Imagine a different kind of future.
A sophisticated AI trading system triggers an unexpected financial crisis, wiping out trillions of dollars in value within hours. Public confidence collapses. Governments, banks, and technology companies come together to build a global AI system designed to detect emerging threats before they spiral out of control.
They call it the Safety Cloud.
At first, its mission is narrow. It watches for financial fraud, cyberattacks, and market instability. The results are encouraging. Fraud declines. Major disruptions become less frequent. Confidence begins to return.
As new challenges emerge, the system expands.
Healthcare organizations connect their diagnostic networks. Transportation systems begin sharing data. Social media platforms contribute signals intended to identify coordinated misinformation campaigns before they spread widely.
Each new layer is added for a sensible reason.
Each promises greater safety.
Each works a little better than the one before it.
Over time, the Safety Cloud becomes less a collection of individual systems than a single network quietly coordinating decisions across much of modern life.
At first, few people notice.
Transactions clear more quickly. Emergency warnings become more accurate. Infrastructure failures become less common. AI seems to be making everyday life safer and more efficient.
Then the changes become more personal.
A journalist’s security clearance is delayed without explanation.
A physician’s license renewal sits in review for weeks.
A social media post questioning an official recommendation quietly disappears after an automated review.
Support requests receive the same response every time:
“Your case has been evaluated according to established safety protocols.”
No individual made the decision.
No individual can reverse it.
The reasoning behind each action exists somewhere inside the system, but few people understand how it arrived there.
Gradually, people adapt.
They avoid subjects likely to trigger additional scrutiny.
Businesses redesign websites to stay within recommended guidelines.
Writers begin checking their work against AI risk filters before publishing.
No law requires these changes.
No government orders them.
People simply learn that life becomes easier when they stay within the system’s invisible boundaries.
Why this is a plausible concern
Nothing in this scenario requires malicious intent.
Many of its building blocks already exist.
Financial institutions use AI to detect fraud.
Hospitals use predictive models to help prioritize patient care.
Online platforms increasingly rely on automated systems to identify scams, misinformation, and coordinated manipulation.
Each serves a legitimate purpose.
The question is what happens if these systems become increasingly connected—and increasingly trusted to make decisions that people once made for themselves.
Harris argues that this is why the myth of inevitability is so dangerous.
The greatest risk isn’t necessarily that someone deliberately designs an AI system to limit human freedom.
It’s that, step by step, we begin treating AI’s judgments as the default because they are faster, cheaper, and more convenient.
No one intends to hand over that much responsibility. It simply becomes easier to let the system decide.
Responsible AI asks more of us than building better technology. It asks us to keep exercising the judgment and responsibility that only people can bear. We can delegate many tasks to AI, and even some decisions under appropriate human oversight. But we cannot delegate responsibility for the consequences. That responsibility remains, and always will remain, our own.