Writing
What would it mean for this to go well?
Most discussion of AI's future dwells on what could go wrong. The opposite question is just as serious — and harder to answer honestly.
Ask people where artificial intelligence is heading and you will mostly hear about danger: lost jobs, runaway systems, a flood of convincing falsehoods, power gathered into a few hands. These concerns deserve to be taken seriously, and we do take them seriously. But a steady diet of worst cases quietly narrows the imagination. It can leave the impression that the only realistic futures are bad ones, and that the best we can hope for is to avoid catastrophe.
There is a different question, equally serious and asked far less often: what would it actually mean for this to go well? Not in a utopian, all-problems-solved sense — that is a fantasy, and fantasies are not useful for planning. We mean something more grounded: a future in which a powerful, general-purpose technology ends up, on balance, making human life freer, healthier, wiser, and more capable, without quietly costing us things we would not knowingly trade away.
This essay is an attempt to describe that future carefully, and to be honest about what reaching it would require. It is a sketch of a possibility, not a prediction and not a promise.
What “going well” is not
It helps to clear away two tempting but unserious pictures.
The first is the techno-utopia, in which AI cures every disease, ends scarcity, and ushers in effortless abundance. The trouble with this picture is not that progress is impossible but that it skips the hard part. Powerful tools have always created problems alongside benefits, and they distribute both unevenly. A vision that assumes the difficulties away cannot guide real choices.
The second is mere catastrophe avoidance — the idea that “going well” just means “nothing terrible happened.” This sets the bar on the floor. A future in which we narrowly dodged disaster but gained little is not a good outcome; it is a near miss. If we only ever ask “how do we avoid the worst?”, we never get around to asking “what is the best we could reasonably build?”
A good future lives between these: realistic about costs and risks, but actually aiming at something worth wanting.
A modest sketch of a good future
So what might that something look like? A few features seem to belong in any honest version.
Capability that stays in human hands. In a good future, AI dramatically expands what people can do — in science, medicine, education, creative work — while the important decisions, and the responsibility for them, remain with humans. The technology amplifies human judgment rather than quietly replacing it. People remain authors of their own lives, not passengers.
Widely shared benefits. A technology that makes a small group vastly more powerful while leaving most people worse off has not “gone well,” however impressive its output. A good future spreads the gains broadly — in access to knowledge, care, opportunity, and tools — rather than concentrating them. This is partly an economic question and partly a question of power, and neither resolves on its own.
A healthier relationship with truth. AI can flood the world with persuasive nonsense, or it can help people find, check, and understand reliable information. Which way this goes is not determined by the technology alone; it depends on the institutions, norms, and tools we build around it. A good future is one where it became easier, not harder, for ordinary people to know what is true.
Preserved human agency and meaning. Much of what makes life worth living comes from effort, skill, relationship, and contribution. A future in which AI does everything for us, leaving us with nothing meaningful to do, would be a strange kind of loss even if it were comfortable. Going well means using these tools to deepen human flourishing — to free people for better work and richer lives — not to make human effort obsolete.
Robustness, not luck. Finally, a good future is one whose goodness does not hang on everyone behaving perfectly. It has margins for error: systems we can correct, decisions we can reverse, power we can hold to account. Outcomes that require flawless execution to stay safe are fragile by design.
What it would take
The honest part of this essay is the recognition that none of the above is automatic. Technologies do not have destinies; they have trajectories shaped by countless human choices. A good future, if we get one, will be built rather than received. At least a few things seem necessary:
- Understanding, widely distributed. People cannot steer what they do not understand. The more broadly the public, and not only specialists, grasps what these systems are and do, the better the collective choices are likely to be. (This is, frankly, much of why an organization like ours exists.)
- Choices made deliberately, while they are still open. Many of the most consequential decisions about a technology get made early, almost by default, and harden into place. Treating them as choices — and making them on purpose — is how a society keeps its options.
- Institutions that can keep up. Law, governance, and norms tend to lag fast-moving technology. Closing that gap, without freezing useful progress, is genuinely difficult and genuinely important.
- A willingness to hold complexity. The future will not be all triumph or all tragedy. Keeping both the promise and the peril in view at once — refusing the comfort of pure optimism or pure doom — is itself a discipline worth practicing.
Why describe it at all
We are aware of the risk in an essay like this. Describe a good future too confidently and you start to sound like an advocate for one particular vision — which is precisely what we try not to be. So let us be plain about the intent.
We are not telling you this future will arrive, or that it is the only good one, or that you should want exactly this. Reasonable people will fill in the details differently, and some will disagree with the sketch entirely. The point is narrower and, we think, defensible: that “what would it take for this to go well?” is a real question, worth as much careful attention as “what could go wrong?” — and that asking it clearly is part of how a good future becomes more likely rather than less.
The worst outcomes are easier to picture than the good ones. That asymmetry is exactly why the good ones are worth the harder work of imagining well.