In June 2019, I wrote a short piece and asked a simple question: are we using our smart gadgets, or are they using us? At the time it read like mild paranoia dressed up as prophecy. I wrote about satellites and selfies, about gene-mapping and self-driving cars, about a coming world in which a child’s traits, habits, and life trajectory could be predicted before conception. I ended with a question I did not answer: do we actually want that predictable a life, or are we better off with the messiness we already have?
Seven years is not a long time. But it has been long enough that the piece now reads less like speculation and more like an early draft of the present. I want to revisit it — not to congratulate myself for having seen something coming, because in truth I saw the shape of the thing but not its texture, not the particular way it would arrive. I want to revisit it because the years between 2019 and now have handed me two things I did not have then: a much closer view of exactly how these systems work, because I now spend my days designing computing hardware from first principles; and a piece of land in Deosane where none of this touches anything, which turns out to be its own kind of answer.
Let me first do the accounting the original essay could only gesture at.
In 2019, “AI” mostly meant recommendation engines — the same machinery that decided what you watched next on YouTube or bought next on Amazon. It was powerful but narrow, optimizing for engagement, not for understanding. What has happened since is a change of kind, not just degree. Large language models arrived and, within a few years, moved from novelty to infrastructure. They do not merely predict what advertisement you will click. They can now read your prose, infer your reasoning style, hold a conversation that adapts to your particular vocabulary and preoccupations, and — this is the part that matters — do this in a way that feels less like being watched and more like being understood. That is a more effective form of the exact thing I was worried about in 2019. Surveillance that announces itself as surveillance is resistible. Surveillance that arrives disguised as a helpful conversation is not.
The data-mining apparatus I described — Google, Amazon, Facebook, “big brothers” collecting the minutiae of purchases, movements, and habits — has not been dismantled by any regulation I am aware of; it has been quietly absorbed and rebuilt on top of. The same companies that mapped your purchasing behaviour in 2019 now train models on the exhaust of that behaviour, and those models are sold back to you, to your employer, to governments, as productivity tools, as safety tools, as convenience. The baby I wrote about, the one who is entered into a database before it can speak, now has a database that can also write essays in its voice, forge its handwriting, and simulate its likeness in video. I did not anticipate that specific escalation in 2019, and I do not think many people did.
What I want to be precise about, though, is the mechanism, because I think I understated it in the original piece. It is not that “they” are watching “us” in some deliberate, conspiratorial sense — though there are certainly actors who do watch deliberately. The more important mechanism is structural: the economic incentive of nearly every large digital system is to model a human being accurately enough to predict, and ideally shape, that human’s next action, whether the action is a purchase, a vote, or simply continued attention. Once a system is good enough at that prediction, the distinction between “serving you” and “using you” stops being a moral question and becomes an engineering question — an optimization target.
In 2019 I asked whether we were being used by our gadgets. The more accurate question in 2026 is: whose optimization target am I currently the input to, and did I ever agree to be that?
But I want to go past the update, because an update alone is just more anxiety with better footnotes. The philosophical question underneath the 2019 essay is the one that actually matters, and I do not think I did it justice the first time. The question was: do we want a predictable life?
I have spent the last several years designing computing hardware — a balanced-ternary architecture built on carbon nanotube devices, the THATTE Structure, and the operating system and language that sit on top of it. This is, on its face, unrelated to a philosophical essay about free will. But the closer I get to the physics of computation, the more I think the two questions are the same question wearing different clothes.
A binary computer represents everything as a choice between two states, on or off, one or zero. Balanced ternary — the system I have built my hardware around — represents information across three states, and crucially, one of those states is a genuine negative, not a placeholder. This is not a technical footnote; it changes how uncertainty and reversal are represented at the most fundamental level of the machine. Binary logic, taken to its philosophical extreme, is a logic of determinism: everything eventually resolves to yes or no, present or absent, predicted correctly or predicted incorrectly. A civilization that builds its entire predictive apparatus — its AI, its surveillance, its gene-mapping — on binary logic is, whether it intends to or not, building a civilization that treats every human trait and choice as eventually resolvable into a bit. My 2019 essay was, without my fully realizing it, describing what a purely binary civilization does to a human life: it tries to collapse the ambiguity out of it.
I did not set out to build balanced ternary hardware as a philosophical statement. I set out to build it because I believe — as an engineer, from thirty years in industry and a background in nanotechnology going back to a Master’s degree in 2006 — that it is a more efficient and more physically honest way to compute, particularly at the nanoscale where carbon nanotube devices behave more naturally in multi-valued states than forced binary ones. But I notice, writing this now, that there is a kind of stubborn refusal built into the choice. A ternary system keeps a “middle” state alive as a first-class citizen — not yet decided, in balance, neither committed nor negated. That is closer to how an actual human deliberates than a system that insists everything must be forced to one of two poles as quickly as possible.
I did not design THATTEOS to resist predictive surveillance. But I think it is fair to say that any computing paradigm which keeps ambiguity structurally alive, rather than engineering it away, is a quiet act of resistance to the predictive future I was worried about in 2019 — even if that was never its stated purpose.
The other thing I have that I did not have in 2019 is the farm at Deosane.
There is a temptation, and I want to resist it, to present the orchard as some kind of noble-savage answer to the surveillance question — go back to the land, escape the database, live simply. That is too easy and not entirely honest. The farm has irrigation planning, solar pump applications, government schemes with their own approval queues, and I will eventually run sensors and monitoring on it, because that is how a serious orchard is run in this decade. I am not opting out of technology. I have never believed opting out was a real option, and I said as much implicitly in 2019 by writing about “big brother” rather than pretending we could simply disconnect.
What the farm gives me instead is a different relationship to time and predictability, and I think that is the actual answer to the question I asked seven years ago. A mango tree planted this year does not fruit for several years. A checkerboard of four hundred-odd pits, planted with Hapus mango, chikoo, sitaphal, lemon, neem — that is a bet placed on a future I cannot fully model, made anyway, because the alternative to planting is not planting. No algorithm I could build would meaningfully improve my ability to predict what that orchard looks like in year seven versus year twelve; the soil, the monsoon, a leopard on the neighbouring plot, all introduce a kind of noise that no amount of data collection resolves in advance.
I have come to think that this irreducible noise is not a bug in a life. It is closer to the point of one. A life that could, in principle, be fully predicted from parental genes and purchasing history is a life with the noise engineered out of it. The orchard is, among other things, a standing argument that the noise is worth keeping.
So, updated and deepened, here is where I actually land, seven years later.
The mechanism I worried about in 2019 has not slowed down; it has become more capable and more intimate, and I was right to be uneasy about it, though I underestimated how quickly language itself — the thing I am using to write this sentence — would become the primary surface on which prediction operates. That part I did not see coming.
But I no longer think the honest answer to “do we want a predictable life” is a flat no, delivered with a shudder, the way I wrote it in 2019. I think the more useful answer is: predictability is not evenly distributed across a life, and the task in front of any one of us is to decide, deliberately, which parts of our life we are willing to let be modelled, optimized, and predicted — because there is real convenience and real benefit there, I am not a purist about this — and which parts we insist on keeping structurally unresolved, ambiguous, slow, and unmonetizable.
For me, at fifty-two, with thirty years of business behind me and — I hope — a good number of years still ahead, that has come to mean two concrete things. In the machines I build, I want the logic itself to preserve ambiguity as a legitimate state rather than a temporary inconvenience on the way to a bit. And in the life I live outside the machines, I want a piece of land, several years’ worth of trees, and a family, that no dataset can accelerate or fully model in advance.
I do not think either of those is a solution to the problem I raised in 2019. Big brother, whatever form it currently takes, is not defeated by a ternary logic gate or by four hundred mango pits. But I no longer think defeat was ever the right frame. The question was never whether the predictive apparatus could be stopped. It was whether I would let it be the only apparatus I lived inside.
Seven years on, I think I have built myself two credible alternatives — one in silicon and carbon, one in soil — and that, for now, is the most honest answer I have to the question I asked myself in 2019.
— Manish Thatte, 18 July 2026