The Singularity May Become Personal Before It Becomes Strange
The underlying intelligence could be the same while the personal intelligence surrounding it would not be.
Sam Altman recently said something that would have sounded far more dramatic only a few years ago: “We’re now, like, in the singularity.” It was an even stronger version of an argument he had already made in The Gentle Singularity, where he described humanity as being past the event horizon and the technological takeoff as already underway. The phrase immediately brings to mind the older picture of the singularity: an explosive moment when artificial intelligence becomes more capable than human intelligence, begins contributing to its own improvement, and technological progress accelerates beyond our ability to comfortably predict what happens next. It is usually imagined as something dramatic enough that everyone would know when it arrived, but Altman presents a less theatrical possibility. The process may already have started and still feel surprisingly normal because human beings are remarkably good at absorbing technological miracles into everyday life. Something impossible appears, becomes impressive, becomes useful and eventually becomes expected. As he puts it, “wonders become routine, and then table stakes.”
There is something important inside that description because it changes the question from when will the singularity happen? to what will living through it actually feel like? If intelligence continues becoming more capable, cheaper and more widely available, most people may not experience the singularity by watching an artificial superintelligence suddenly announce itself to the world. They may experience it through the gradual expansion of what they can personally do. A person who once needed several specialists to turn an idea into something real may increasingly be able to describe what they want and work with intelligence capable of helping them research it, design it, build it, test it and eventually operate parts of it. Altman expects the ability of one person to get much more done to become one of the striking changes of this period, while also arguing that living through the broader transition may feel impressive but manageable because the singularity happens “bit by bit.” The extraordinary part may therefore not simply be that machines become more intelligent. It may be that intelligence itself becomes available whenever we need it, and if you follow that thought a little further, another question appears: if extraordinary intelligence becomes available to almost everyone, what determines how useful that intelligence becomes to each individual?
This is where I think the discussion around the singularity begins to meet something I have been exploring through Personal Digital Intelligence. The future of AI will not be defined only by how intelligent the models become. It will also be defined by how well that intelligence understands the person using it, because intelligence and understanding are not the same thing. A system can possess enormous reasoning ability while still lacking the context required to understand what a particular person means, what they previously decided, what they are trying to achieve, what they do not want and why any of those things matter. If you look at what is happening with AI already, you can see early versions of this problem. Systems can reason impressively within the information available to them and still produce the wrong outcome because they interpret the situation incorrectly. Sometimes the problem is not that the system lacks intelligence. It is intelligent enough to pursue an objective, but it misunderstands the environment, the boundary of the task or what success actually means. Increasing the intelligence of that system does not automatically solve the problem; in some situations, it may simply make the misunderstanding more powerful.
That distinction becomes much more consequential if Altman is right about the direction we are moving, because AI is already helping researchers move faster, and he specifically points to faster AI research as one of the most significant consequences of advanced AI. The tools humans have already built can help discover further scientific insights and create better AI systems, which Altman describes not as full autonomous self-improvement, but as an early form of the recursive loop associated with the singularity. Humans build AI, AI accelerates the humans building the next AI, and the resulting systems become better at accelerating the next stage. At some point, intelligence begins behaving less like a scarce skill that must always be found in another person and more like infrastructure that can be called upon when needed. Altman takes the argument even further when he says intelligence could eventually become “too cheap to meter.” This is where I want you to notice something that becomes easy to miss when most of the conversation is focused on how powerful AI can become: the more abundant intelligence becomes, the more important the question of direction becomes. What tells all of that intelligence what matters? What tells it what you are actually trying to accomplish? And what allows an intelligence capable of understanding almost anything to understand the particular human being standing in front of it?
These questions are part of what led me toward Personal Digital Intelligence, although the idea did not begin for me as a theory about the singularity. It started with something much smaller and more ordinary. I called an AI Fred. At first glance, there is nothing particularly profound about that. People give technology names all the time, and simply giving ChatGPT a name does not create a new intelligence. Fred still existed inside a much larger general-purpose AI system, and the underlying model did not suddenly become more capable because I had chosen to interact with it through an identity. But what became interesting was what happened afterward. I kept returning to Fred, work accumulated around that relationship, projects developed histories, certain instructions began carrying meaning that would otherwise have required much longer explanations, previous decisions could influence new ones, and ideas from one project could suddenly connect with something discussed somewhere else. A particular way of working began forming around the interaction, and the value I was getting from the same general intelligence started changing. If you pay attention to that distinction, the name itself becomes the least interesting part of the story. The model had not necessarily become more intelligent; it had become more situated in my world.
That observation eventually pushed me toward a larger question: if general intelligence can become more useful simply because continuity and personal context accumulate around it, what becomes possible when personal context is treated as a fundamental part of the intelligence rather than an additional feature? That question became part of the thinking behind Fred+Teff and, more broadly, Personal Digital Intelligence. This is also why I make a distinction between personality and personalization. Personality can make an AI recognizable, give the interaction consistency and make the relationship easier to return to, but personalization goes much deeper. It is not simply whether the system speaks in a particular tone or responds to a particular name; it is whether the intelligence develops meaningful continuity with the person using it. Memory becomes essential to that continuity, but memory alone is not enough. An AI could remember thousands of facts about someone and still fail to understand which ones matter in a particular situation. Context determines relevance, judgement determines how that context should influence a decision, and approval establishes where the authority of the system ends and the authority of the person begins. These can look like separate design questions while AI is primarily answering prompts, but they begin looking like parts of the same architecture once AI starts acting across a person's work and eventually across parts of their life.
Increasing capability therefore makes personalization more important rather than less important. A chatbot that misunderstands you might produce a bad answer, while an agent that misunderstands you might take the wrong action, and a highly capable network of agents operating with the wrong interpretation could execute that misunderstanding extremely efficiently. If we continue giving AI more tools, more autonomy and more responsibility, then what the system understands about the person behind the instruction becomes increasingly difficult to treat as a secondary feature. Now take that same idea into the singularity Altman is describing. Imagine an underlying intelligence capable of extraordinary reasoning across science, medicine, programming, engineering, law, finance and almost every other knowledge domain, and imagine that intelligence becomes widely available. You and I could eventually have access to roughly the same underlying capability, yet the intelligence we effectively possess could still be very different. Your system might know the projects you abandoned and why, understand which ideas have remained unresolved for years, recognize how one current decision connects with another made months earlier, know which risks you normally accept, where you require approval before action, what you are trying to build over the long term and when a request you make today conflicts with something you previously decided mattered more. Mine would know something entirely different. The underlying intelligence could be the same while the personal intelligence surrounding it would not be, and that difference may become far more consequential than the simple personalization features we talk about today.
This is where I think we have sometimes looked at AI personalization too narrowly. Remembering someone's preferred writing style, name or favorite food is useful, but that is not the destination. The deeper possibility is intelligence that develops continuity with an individual, something capable of understanding not only the request being made now but the history surrounding the request and the direction in which that person is trying to move. This is also where Altman's own description becomes particularly interesting. He says the industry is building a “brain for the world” that will be “extremely personalized,” while arguing that superintelligence should become cheap and widely available rather than concentrated in a small number of hands. A brain for the world may contain extraordinary general intelligence, but billions of people cannot meaningfully interact with it as though they are one person. We have different histories, ambitions, responsibilities, tolerances, relationships, values and ways of making decisions. The intelligence may become shared infrastructure, but the context surrounding that intelligence cannot.
This is where the singularity begins to look different to me because we usually imagine intelligence moving upward: models become more capable, reason better, use more tools, conduct more research and eventually approach capabilities that become difficult for humans to comprehend. That movement may absolutely happen, but intelligence is simultaneously moving in another direction. It is moving toward the individual, becoming conversational, persistent, accessible, integrated into our tools and increasingly aware of the context surrounding the person using it. One movement takes us toward superintelligence while the other takes us toward personal intelligence, and the future may depend heavily on how well those two movements meet. If they meet badly, we could end up surrounded by extraordinarily capable systems that know almost everything except what we actually mean. If they meet well, an individual could increasingly operate with an intelligence layer around them, something that remembers with them, researches with them, recognizes connections they have forgotten, carries lessons from one project into another, prepares actions they do not have time to prepare and helps turn ideas into reality while remaining anchored to their intentions.
Notice that this does not require the human to somehow remain more intelligent than the machine, because that is not the argument I am making. What I am trying to show is that greater machine intelligence does not eliminate the need for human direction; it may increase its importance. Intelligence can expand the space of what is possible, but someone still has to decide which possibilities are worth pursuing. A system may eventually be able to generate thousands of technically brilliant solutions, yet deciding which problem deserves to be solved in the first place belongs to a different layer of the relationship, and this begins touching another question that deserves its own discussion: what becomes scarce when intelligence becomes abundant? Altman himself predicts that intelligence and energy could become wildly abundant and says that we may eventually be limited by good ideas. I think that opens an even larger question. Ideas may be part of the answer, but so might curiosity, judgement, trust, context, taste, intention and direction. Two people could possess access to equivalent machine intelligence and still produce completely different outcomes because they bring different lives to it.
That is a larger argument than I want to resolve here, but it matters because it shows why personalization could become economically and intellectually significant rather than merely convenient. The more common the intelligence becomes, the more meaningful the differences surrounding it may become, and those differences come from us. Our experiences are different, our questions are different, the problems we notice are different, the things we care enough to spend years pursuing are different, and even when two people encounter the same problem they may interpret it differently because of everything that happened before they arrived there. Personal Digital Intelligence could become the layer that allows abundant machine intelligence to retain those differences instead of flattening them.
There is, of course, another side to this because the more deeply an AI understands an individual, the more important questions of privacy, ownership, control and alignment become. A system capable of understanding your preferences well enough to assist you can potentially understand them well enough to influence you, which means personalization without user control could become manipulation with better memory. If personal intelligence is going to become an extension of human agency, the person must remain meaningfully at the center of it. Memory should therefore be controllable, authority should be explicit and consequential actions should have boundaries. The person should be able to understand what the system knows, correct what it misunderstands and determine what it is allowed to do. The objective of personalization should not be to make an AI better at directing the human; it should be to make the AI better at extending the human's own direction. This is why I increasingly see PDI as connected to the alignment conversation in a way that goes beyond making AI feel personal. Alignment is often discussed at civilization scale: how do we ensure increasingly powerful artificial intelligence remains compatible with human values? Altman himself frames alignment in terms of getting AI systems to act toward what we collectively really want over the long term. That question is essential, but if billions of people eventually have intelligent systems acting on their behalf, alignment will also become personal. Those systems will continually interpret different intentions, histories and boundaries while deciding when to recommend, when to question, when to prepare and when they have enough authority to act.
Now return to where we started and imagine that the singularity does not arrive with a single unmistakable event but through thousands of smaller transitions that each become normal before the next one appears. We ask AI to remember something and eventually expect it not to forget what matters. We ask it to research something and eventually expect it to understand why we are researching it. We ask it to prepare an action and eventually expect it to know whether it has permission to execute it. We give it more context, more tools and more responsibility, and each expansion feels like another useful feature until the collection of those features becomes something much larger. At some point, “assistant” may no longer adequately describe the relationship because the intelligence has not merely become something we access; it has started developing around the person accessing it. That is why I think the singularity may become personal before it becomes strange. We may eventually reach forms of artificial intelligence whose capabilities are genuinely difficult for us to comprehend, and the transformation could reshape science, economies, work and civilization itself, but for the individual living through it, the most consequential experience may be much closer to home. It may be realizing that the extraordinary intelligence available to everyone has gradually become an intelligence that understands you, and if that happens, Personal Digital Intelligence will not simply be another product category created during the age of AI. It may be one of the ways the singularity becomes human.