TL;DR: Superintelligence, civilizational risk, mass unemployment and machine consciousness deserve serious attention, and the people who work on them are doing real work. Using current tools in your professional practice is a separate decision, the one how I use AI on a book deals with, and conflating the two produces the paralysis that keeps thoughtful people on the sidelines of AI.
I take the long-term questions seriously, and I still get impatient when they’re used as a reason to avoid a transcription tool. Turning down a chatbot doesn’t slow a single lab. It lets a thoughtful professional feel principled while their own work carries on the same way, and the people pushing to slow AI down get nothing from the gesture.
Why do AI takeover fears and today’s tools get confused?
The discourse around AI has two layers that sound similar but operate at completely different time scales and stakes.
The first layer is the long-term question about where the technology is going. Will systems become superintelligent? Will they pose existential risks? Could they restructure the economy in ways that produce mass unemployment? Will they become conscious in some morally relevant sense? Those questions are real and serious. The people working on them include some of the most thoughtful researchers in the field, and the answers will matter enormously for the next several decades.
The second layer is the short-term question about whether to use current tools in your professional practice.
Should you let an AI help you clean up a transcript? Is it OK to use it for first-pass research summaries? Should you let it draft connective sections of a document? Those questions are concrete, the answers depend on your specific work, and the stakes are measured in hours saved and quality maintained instead of in civilizational trajectories.
The tools on your desk today are bounded. They won’t wake up superintelligent over a weekend, and using one to tidy a transcript has no effect on where the frontier goes. I think most people know this, and the fear sticks anyway, because the companies selling the tools keep talking as if every subscription is a step toward the future they warn about.
Watch: What Is Intelligence? Where Does it Begin?, Kurzgesagt – In a Nutshell
Why does conflating AI fears produce paralysis?
The doubter who reads about the long-term questions and concludes that long-term concerns should guide the short-term tool decision is making a category mistake. The mistake is understandable. The rhetoric around AI slides between the two layers without warning, and the same companies promote the long-term vision while selling the short-term product.
The slide makes the doubter feel that engaging with the short-term product implicates them in the long-term trajectory. That produces moral discomfort and contributes to refusal.
The doubter who reads about the long-term questions and concludes that long-term concerns should guide the short-term tool decision is making a category mistake.
The implication isn’t there. The use of a chatbot to clean up an interview transcript doesn’t contribute to superintelligence in any direct way. Refusal to use it doesn’t slow the development of superintelligence either. The technical work that produces frontier capability happens at a small number of large companies whose research agendas aren’t affected by individual customer choices on consumer products.
The professional who refuses on existential-risk grounds has no influence over the underlying risk. I’d call that stance comfort with no effect, and I don’t think serious people should settle for it. If the risk is worth losing sleep over, it’s worth an action that reaches it.
What serious people working on long-term risk do
The researchers and policy people who take long-term AI risk seriously aren’t refusing to use current AI tools. Not as a group, anyway. Many of them use the tools heavily, because using the tools is part of how they think clearly about what the tools can and can’t do. The position “this technology poses long-term risks worth working on” is fully consistent with the position “current tools are useful for these specific professional tasks.” The researchers who hold both positions at once aren’t being inconsistent.
They’re distinguishing the long-term risk question from the short-term capability question, and the thoughtful doubter should make the same move. The serious work on long-term risk looks like safety research, policy advocacy, governance proposals and institutional design. If you care about the risk and want to act on it, support the policy work, give to research, and show up for the political fights over governance. Refusing a transcription tool doesn’t belong anywhere on that list.
The economic disruption layer
The middle layer between the long-term existential question and the short-term tool question is the economic disruption question.
Will AI restructure labor markets, eliminate categories of work, concentrate wealth in fewer hands, or produce massive social adjustment costs? Those questions are also serious, and their time scale is shorter than the existential questions. Some of these effects are likely to play out within the next decade. The professional who’s worried about them is responding to something real and worth thinking about.
This is the layer I’d spend my own worry on, because it reaches people’s paychecks first. The decisions driving it get made by executives, investors and legislators, and none of them are waiting to see whether you personally open a chatbot.
The same logic applies. Individual refusal to use current tools doesn’t preserve the affected categories of work or slow the economic effects. The effects are downstream of major commercial deployments, policy decisions, and labor-organizing responses, and one professional declining a productivity tool moves none of those. Engagement with the technology, combined with engagement with labor policy and industry norms, is a more useful posture for a professional worried about disruption than refusal. A piece on the realistic AI-and-jobs picture covers the economic question in more detail without conflating it with the existential one.
The “but what if you’re wrong about long-term risk” question
The doubter can reasonably press the point. What if the existential risk turns out to be real, and the people who used the tools and contributed to the discourse normalized something dangerous? That argument deserves a direct response. The response is that normalization of current tools and contribution to existential risk aren’t the same causal chain. Existential risk, if it materializes, will materialize from advances in capability driven by frontier research at specific labs, not from professional adoption of consumer products.
A professional who used ChatGPT in 2026 won’t have any meaningful causal role in any AI catastrophe in 2040, because the causal path runs through technical capability development that’s fully decoupled from consumer adoption rates.
Any professional who used ChatGPT in 2026 and also contributed to public discourse, policy conversations, and industry norms about AI governance might have a meaningful causal role in better outcomes, because those are the levers where individual professional voice matters. The real version of the existential-risk concern leads to engagement with policy and governance. Refusal often masks a deeper unwillingness to engage with the policy work that would matter, and the unwillingness is what the doubter should examine.
My take
Worry about superintelligence on your own time, and pick up the transcription tool anyway. The two decisions have nothing to do with each other. Your refusal won’t slow frontier labs or protect anybody’s job. Economic disruption is where I’d spend my worry, because it hits paychecks first. If the long term risk keeps you up, put your voice into the policy work and the governance fights.
What should a thoughtful AI doubter do?
The position that holds together for a thoughtful professional looks something like this.
Use current tools where they help your work, with discipline and clear-eyed assessment of their limits. Engage with the policy and governance conversations about AI, where individual professional voice has real use. Support the technical safety research that takes long-term risk seriously, financially and rhetorically. Refuse the conflation between the short-term tool decision and the long-term trajectory question, because the conflation produces refusal stances without changing the things the doubter cares about.
That position asks more of a person than refusal does, because it means working several layers of the AI question at once. It also fits the facts: the tools work, the long-term questions are real, and the two connect in messier ways than either camp admits. I’d much sooner hear from a skeptic who uses the tools and argues about governance than from one who does neither.
My verdict is that blanket refusal is the easy way out wearing the clothes of principle.
If you’re worried about where AI is headed, I’m worried about parts of it too, and that worry should send you to the policy fights and the research that can change something. Meanwhile, use the tools that make your work better today. Keeping those two decisions apart is harder than refusing both, and it’s the only version that does anybody any good.
Today’s tools are a separate conversation from superintelligence. The Birth of the Augmented Human is my book on what goes right with those tools when you use them properly.
