Now Is Not the Time to Prove Intelligence

We are entering a strange period in our relationship with intelligence. The systems around us are becoming capable of writing, coding, analysing, researching, reasoning and solving problems that once demanded years of human training, yet a surprising amount of our attention is still going towards proving that we can do those things ourselves. Can you write without AI? Can you code without it? Can you solve the problem without asking the model? Can a human still produce something better? There are moments when those questions matter because we should know how to think independently, question an answer, recognise when something is wrong and make decisions without surrendering our judgement to a machine, but there is a difference between protecting your intelligence and performing it simply to prove that you still have it. We have to be careful not to turn intellectual independence into intellectual vanity. If I have access to an abundance of intelligence because of a tool, why should I spend this moment struggling to prove that I can perform the same intellectual work the tool can already perform? Imagine standing beside an excavator and insisting on digging the foundation with a shovel because you want everyone watching to know that you are strong enough to do it. Knowing how to use the shovel still matters and there may even be moments when you need it, but once the excavator is sitting beside you, the more important question is no longer whether you can dig, it is what you are going to build. That is the shift I think we need to make with AI: now is not the time to prove intelligence, now is the time to discover what intelligence makes possible.

For most of human history, intelligence has been something we carried mostly within ourselves, so the person who could calculate faster, remember more, write better, understand several fields or solve a difficult problem had an advantage because those abilities were difficult to acquire and difficult to reproduce. We built education, professions and even parts of our identity around that reality, but now imagine what happens when some of those abilities become available on demand through systems that continue getting more capable. The value of human intelligence does not disappear, but the question we ask of it has to change. Instead of spending all our energy saying, look what I can still do, we should increasingly be asking, look where I can now reach. If AI can help you write the code you could not write yesterday, your ambition does not have to stop at eventually proving that you can write the same code without it, because you can instead ask what you are now able to build because coding is no longer standing between you and the idea. If a model can analyse information faster than you can, there is little value in turning that into a race simply to prove that your analysis remains superior when you could be asking what problem you can now investigate because that analytical ability is available to you. And if a system can help you move between mathematics, software, design, science, history or fields you could never have mastered simultaneously, then maybe the important opportunity lies in what becomes possible when those abilities meet.

This is where I think our ambition has to rise with the intelligence available to us because when intelligence becomes abundant, ambition should become more expensive. The standard has to rise with the opportunity, and we cannot be given greater and greater cognitive power while continuing to set ourselves exactly the same targets. Productivity matters, efficiency matters and doing familiar work faster has real value, but if this new intelligence only helps us do yesterday’s work a little more efficiently, then we will have misunderstood the size of the moment because at some point, increased intelligence should change not only how quickly we answer our existing questions but the size of the questions we are willing to ask. Take writing, for example. A writer does not need to defeat the model in some imaginary writing competition when the more interesting question is whether access to that intelligence allows the writer to examine an idea more deeply, challenge assumptions harder, discover connections that might otherwise have remained invisible and eventually produce thinking that neither the human nor the model would have reached independently. The same is true for a scientist, an engineer, a founder or a researcher, and it matters just as much for someone who carries none of those titles because a person may understand a problem after living beside it for twenty years without possessing the technical language needed to investigate it, someone may carry an invention in their head without knowing how to program it, while another person may recognise a connection between two things because of an unusual life experience that no institution would think to place beside each other.

That person brings something into the relationship that matters because a model may contain extraordinary amounts of knowledge and capability, but it does not arrive with every question humanity could ever ask already formed, nor does it carry every human frustration, every local problem, every unusual curiosity, every obsession or every intention that might cause its intelligence to be pointed somewhere new. So maybe we should be careful about defining the frontier of AI only by asking what the model itself can now do. The frontier isn’t necessarily inside the human and it isn’t necessarily inside the AI; the frontier may exist in the interaction between them. I think that possibility deserves much deeper investigation of its own, but for the argument here it already changes the question we should be asking because we should not only ask these systems to do the things we already know they can do: write this, summarise that, build this website, analyse this document, produce this code. Those things are useful and they are already changing how people work, but what happens when we begin trying to achieve things nobody yet knows the tool can help us achieve, or when a human keeps pushing a question into territory that neither the human nor the model could have completely mapped before they began? Maybe there are things these systems can help us reach that even the systems themselves could not have listed as capabilities before somebody tried.

We may therefore be standing in front of an enormous new field of human achievement while spending much of our time asking whether the machine can write a better email, which is why I think we need to ask a much more ambitious question: what becomes achievable now that this intelligence exists? That question carries more weight for me than whether AI can outperform a human at one more benchmark or profession, and it takes us towards what I think of as the holy grail of invention, although I do not mean one miraculous invention waiting somewhere for AI to discover. The holy grail is the unexplored territory created when billions of different human intentions gain access to intelligence beyond their individual limits, because every person brings a different question, every environment exposes different problems, every field has blind spots that someone standing outside it may notice and every life produces experiences capable of pointing intelligence in a direction it has not been pointed before. Give those intentions greater intellectual reach and we do not know exactly what comes out of the other side, but that uncertainty is part of the opportunity.

There is another problem here, however, and the more I think and write about Personal Digital Intelligence, the harder it becomes for me to ignore it. We are building AI systems whose capabilities may continue increasing far faster than our own, while human beings are not suddenly going to become ten times, a hundred times or a thousand times more intelligent simply because the machines around us do. If future systems eventually move towards forms of intelligence far beyond ours, I do not think the sensible goal is to force ordinary humans into an impossible race to close that gap because maybe we never close it, and maybe we do not need to. The question I keep arriving at is much more practical: if future AI systems become extraordinarily more intelligent than the people using them, how do ordinary humans remain capable of using that intelligence meaningfully? This matters because available intelligence is not the same thing as accessible intelligence. You could place extraordinary intelligence in front of someone and still leave most of its value beyond their reach because they may not know what is possible, which question to ask, how to explain the problem in a way the system can fully work with, which assumption deserves to be challenged or how to judge the answer they receive. The AI could therefore become dramatically more capable while the person sitting in front of it experiences only a small part of that increase, and at that point simply making the model smarter is not enough because we also have to think about the distance between the intelligence of the system and the human trying to use it.

We have solved simpler versions of this problem before because we did not make everyone learn machine code before they were allowed to benefit from computers; instead, we built ways for ordinary people to express what they wanted while increasingly complicated machines handled more of what was happening underneath. But superintelligence, if we reach it, would present a much deeper version of that problem because this would not simply be about making software easier to operate, it would be about making extraordinary intelligence meaningfully reachable by ordinary intelligence.

That requires a bridge.

I have spent a lot of time thinking about that problem through my work on PDI, and I think the question itself deserves much more attention because the future of AI cannot only be about how intelligent the machines become, it must also be about how much of that intelligence an ordinary human can actually reach. We do not need to force humans to suddenly become extraordinarily intelligent just to remain useful beside extraordinary systems; we need a bridge that allows ordinary humans to remain capable in the presence of extraordinary intelligence. And notice what happens when we look at the problem this way: the goal stops being to make the human equal to the machine, but it also stops being to replace the human with the machine, leaving a much more interesting possibility between those two ideas, which is to give the human greater reach through the machine.

This matters most when we stop imagining only experts using these systems, so think about the ordinary person who understands a problem but lacks the mathematics to model it, the small founder with an idea but no ability to assemble six specialists, the researcher who sees a possibility outside her field but does not have years to retrain, or the young person without access to the institutions that historically gathered advanced knowledge behind their walls. The point is not to pretend expertise no longer matters, because it does, but rather that the distance between an idea and the intelligence required to explore that idea may be shrinking. If we can build the right bridges between ordinary human intention and extraordinary machine capability, then the promise of increasingly powerful AI becomes much bigger than making our smartest people even more capable because it becomes the possibility of ordinary people attempting extraordinary things.

Now look back at the argument we started with and our obsession with proving that humans can still write, code, reason or calculate without AI begins to look much smaller. Imagine someone looking back at this period from 2035 and discovering that the embarrassing outcome was not necessarily that humans became dependent on AI, but that we were given extraordinary cognitive leverage and mostly used it to reproduce things humans were already capable of producing: more emails, more essays, more websites, more presentations, faster code and better advertisements. All of those things are useful, certainly, but imagine being given this much additional intelligence and using most of it to make the existing world slightly more efficient. That would be a failure of imagination because we should be looking towards problems we previously considered too difficult to approach, scientific questions we could not investigate, businesses one person could never have built, connections between fields that were previously too far apart, new forms of creativity, new systems and new ways of understanding ourselves and the world around us, until eventually we begin reaching for things for which we do not yet have names because nobody has reached them.

Humanity has spent thousands of years building ways to move beyond the natural limits of the individual: writing extended memory, mathematics extended reasoning, telescopes extended sight, machines extended physical strength, computers extended calculation and networks extended communication. We did not look at those inventions and decide that using them somehow made the unaided human less worthy because the point was never to worship the limit, but to use what we created to reach beyond it. Artificial intelligence brings that old human instinct into unusually sensitive territory because this time we are extending something we have tied very closely to our identity: intelligence itself. I understand why that makes us defensive and why someone wants to prove that they can still write the paragraph, solve the equation, build the program or find the answer themselves, but this is where I want you to look beyond that competition. We have spent thousands of years trying to become more intelligent, and now that we are beginning to build intelligence greater than our own, it would be a strange failure of imagination if all we used it for was proving that we are still intelligent too.

We should protect our judgement, preserve our ability to think and know when to question the machine or when not to use it at all, but none of those things require us to make intellectual struggle the measure of human worth. The future does not need billions of people desperately trying to prove that they can think like superintelligent machines because we do not need ordinary humans to become superintelligent; we need ordinary humans to become capable of reaching extraordinary things through superintelligence. That, to me, is the larger opportunity: not proving that the human can still do what the machine can do and not surrendering the human to the machine, but giving human curiosity, judgement, experience and ambition access to intelligence beyond the limits of the person carrying them. Once that happens, the question in front of us becomes much larger than what AI can do because it becomes what humanity can now attempt.

We should be looking for things humanity has never reached before.

— Teff