Heads, Continuity. Tails, Singularity.

Continuity Singularity

A friend of mine is currently going back and forth on whether to stay at university for a PhD or start working. On the one hand: three more years on a junior researcher’s wage, but then a steeper lifetime earnings curve and a pension that more than makes up for the later start. On the other: three years of income now, savings compounding earlier and a professional head start while peers are still writing. After walking me through his options, he asked what I think.

I said something sensible and forgettable. Advice no better than he would have got from any free-tier AI model today. What I actually think about decisions like this is that he is weighing the wrong thing. Both options are totally justifiable, but I suspect his decision hinges on the wrong premise.

Long before his pension, possibly even before he would submit that thesis, there may be research labs with the lights off. Not because their overvalued IPO failed, but because nothing inside needs to see: AI conducting thousands of experiments a night, a singular intelligence deciding which ones merit pursuit, and the finding my friend would have laboured over for three years, solved by breakfast, superseded by lunch. In that world, a doctorate may not be justifiable economically, only by passion and curiosity.

I ponder this scenario with every model release, new AI product and the daily multi-billion-dollar data-centre deal announcements. It hides in ordinary discussions about pensions, national debt, migration, what to study or where to invest. Above the debates, a single question may be decisive for all of them:

Will we have artificial superintelligence within ten years?

Suppose that in five years AI capabilities are merely astonishing: models ship whole software products from a one-paragraph brief, audit a multinational’s books overnight, negotiate complex contracts agentically and find the protein human specialists missed. That is not the scenario I mean. Odd as it sounds, that is continuity. However powerful, that AI still fits the oldest category we have, a tool that makes human work go further. The economy it produces would be faster and stranger, but recognisably ours. Humans remain the protagonist.

Superintelligence is a different category. Science, invention, the development of the next generation of AI, even the oversight of the thinking machines themselves could carry on without us. There is a reason the arrival of superintelligence is often called the Singularity. Like the event horizon of a black hole, it marks the point we cannot see past. However, the advent of superintelligence will certainly be neither luck nor chance, but the product of unprecedented capital investment and the work of highly driven researchers and engineers. On top of both, recent advancements in recursive self-improvement dramatically accelerated this scenario and make it harder to judge how quickly it could diffuse through the economy. A tool is defined by the hand that guides it, and no hand can sensibly steer what it cannot understand. I fear we will follow superintelligent agents the way a dog follows dinner-table conversation, alert to the tone and deaf to the content. A mind beyond ours will not wait for chat instructions.

Most of the ways we tell the story of a life were built on the assumption that human thinking is scarce. We introduce ourselves by our jobs. We judge an education by where it gets you, and a career by the expertise it piles up. For centuries, asking someone what they do meant asking who they are. As thinking stops being scarce now, a large part of personal identity is in question. Students already ask why they should learn anything the hard way. The entry-level roles it once led to are disappearing, as every release cycle brings AI deeper into knowledge work. Superintelligence would merely finish what today’s tools have begun. So the change I mean is much larger than a tougher job market.

This binary outcome reframes how we should think about political and economic decisions. Three recent debates, picked from many, show how far it reaches:

If the coin shows continuity, these are still the right arguments, perhaps even eased by a productivity boom. If it shows singularity, human skill is no longer the bottleneck, and 2060 has no taxpayers in any sense we would recognise. None of these debates mention the binarity of the future, today. Western societies, which have built much of their self-image on planning for the long run, would be shaken faster than they expect.

The coin hangs over smaller choices too, like giving up weekends to study for the CFA, spending years getting good at your job, or picking a coding elective at university over a literature course. Every act of learning is a deposit into the asset you and I can neither sell nor transfer: human capital. If the outcome is continuity, it remains an income-generating asset. If it shows singularity, it may be a write-off.

Will we find new jobs, purpose and identity after superintelligence?

Will humans go the way of the horse? It is an old question, older than the language model and older than the computer. Every generation of machines has revived it, and so far every generation of workers has answered it by transitioning to more skilled labour. Each time machines and tools took over a task, the job left for the human carried greater agency and more accountability. We are paid less and less for our hands, and more and more for our judgement.

In the next decade, this pattern will protect only a few professions. Repeatable work on a screen is already at risk, with routine digital administration, writing, coding, data analysis and more junior roles feeling this today, while electricians, nurses and painters, who work with their hands in unpredictable rooms, are safe for longer. Digital superintelligence cannot yet wire physical cables, but the robotics singularity seems to be catching up. Only roles built on trust and liability, like the surgeon, the judge, the politician or the executive, will probably last longest, protected by law and habit more than true necessity.

Share of tasks exposed

But moving up only works while there is a rung above. Once the minds are superintelligent and the robots have caught up, there is no next rung. At that point the question is no longer which job we move into next, but who we are when we are no longer needed for anything. That is a question of identity.

1. 05???
2. 04Trust & liabilitySurgeons, judges, politicians, executives
3. 03JudgementKnowledge work on a screen
4. 02HandsCraft and factory work
5. 01MuscleThe work of the horse

My friend’s career spreadsheet had no row for any of this, and neither does mine. We just don’t know what to do with this scenario. I am not asking anyone to bet “heads or tails”. But wait and see is not an option either. This could be the defining development of our lifetime, and I dare say the most consequential in human history.

So what can you do?

Picture what you yourself truthfully feel AI will be capable of in ten years. It will probably be wrong, but as the technology and its diffusion through the economy develop, you will have an anchor for measuring how far we have come, and how you have positioned your professional and personal life:

• Weigh your largest professional commitments (degree, career, savings, city) and ask on which “side of the coin” each pays out. Capital accumulated now is one of the few stakes that pays on both sides, while a credential that only begins paying out in 2032 may be at risk.
• Check for immediate return. The fulfilling side projects, the language that opens up new social connection, work that is interesting today rather than safe in ten years.
• Value optionality. As the variance of future outcomes increases, the value of optionality increases.
• Update your mental model constantly.