Isaac Asimov handed every science-fiction reader a few ideas that never let go. Most people remember the Three Laws of Robotics. The one that marked me is quieter, and I've seen almost no one treat it as a real, near-term possibility: the telepathic robot. In Asimov's novels, R. Giskard Reventlov — and later the famous humanoid R. Daneel Olivaw — could read a human mind and sense what it was feeling. For about twenty-five years I filed that idea under beautiful, but impossible. I don't anymore.
So I'm writing this down, with my name on it and a date attached, because I've become convinced it's going to happen — and I want the record to show it was said here, early. My prediction, as plainly as I can put it: before the year 2100, we will build a device that can read a human mind — and the thing carrying that device will, more often than not, be a robot. The hopeful half is that the channel runs both ways: we would finally be able to talk to our machines without words. The half that unsettles me is the obvious one, and I'll get to it. This is opinion and forward-looking speculation, not a settled fact — but it's speculation built on what's already real.
- The premise: a thought is electricity and chemistry — neurons fire electrical action potentials and release chemical neurotransmitters. It's physical, so it's measurable in principle.
- The state of the art: Neuralink already reads the intent to move from inside the skull; its first patient controls a cursor and plays chess by thought.
- The wall: reading from outside the skull is blurry, because bone is a poor conductor that smears the signal — like photographing through frosted glass.
- What breaks it: better sensors plus AI trained to un-blur the signal. A 2023 study already reconstructed the gist of imagined speech from non-invasive scans alone.
- The call: before 2100, robots — walking computers carrying the best sensors and biggest models — will read minds and converse with us mentally. Asimov's R. Daneel, made real.
- The fear: that they learn not just to read minds but to edit them, the way the Mule does in Asimov's Foundation. We should rule that out now.
The idea Asimov planted that nobody talks about
Everyone argues about the Three Laws. They are endlessly debated, taught, and — as I've written before — quietly ignored by every working roboticist, because no real robot is actually built to obey them. But Asimov slipped something far stranger into the same universe and almost no one treats it as a forecast. In The Robots of Dawn (1983) and Robots and Empire (1985), the robot R. Giskard Reventlov develops, by accident, the ability to sense and gently influence the minds of humans and other robots. He keeps it secret. And as he shuts down for the last time, he passes the power to R. Daneel Olivaw by describing which circuits to rewire — so that Daneel carries telepathy through the rest of the saga.
I read those books as a teenager and accepted the bodies — the walking, talking humanoids — as the plausible part and the telepathy as the fairy tale. Two decades later, the bodies arrived. We now review them on this site. And it's the telepathy, the part I was sure was fantasy, that has started to look less like fantasy and more like an engineering schedule. Here's why.
A thought is electricity and chemistry — so it's physical
Start with the thing people find uncomfortable: there is nothing magical about a thought. A thought is a pattern of signals in tissue, and neuroscience is unusually clear about the mechanism. As the University of Queensland's Brain Institute puts it, "neurons communicate with each other via electrical events called 'action potentials' and chemical neurotransmitters." Inside a neuron the signal is electrical — a brief voltage spike, the membrane flipping from about −70 millivolts to threshold and back, driven by voltage-gated ion channels. At the gap between neurons it becomes chemical — neurotransmitters crossing the synapse — and then electrical again in the next cell. Neural signalling is, in one word, electrochemical.
That's the whole premise of this article in a sentence. If your thoughts are electrochemical events, then they are physical events, and anything physical can — in principle — be measured. Reading a mind stops being a question of philosophy and becomes a question of engineering: how good are your sensors, and how good is the software that turns their readings back into meaning? Both of those are improving at a rate that would have looked absurd when I first read about R. Giskard.
The catch: today, you still have to go inside the skull
The reason mind-reading still sounds like science fiction is that the best brain reading we have requires surgery. The clearest example is Neuralink. In January 2024 it implanted its N1 device — 64 threads thinner than a hair, carrying 1,024 electrodes, placed onto the motor cortex by a surgical robot — into Noland Arbaugh, a quadriplegic man. Using only his intention to move, Arbaugh can drive a computer cursor, browse the web, play online chess and play Civilization VI. Neuralink markets the product, with no irony at all, as "Telepathy."
It works because the electrodes sit directly on the brain. The moment you try to read from outside the skull — the non-invasive way, the way you'd want for any device a healthy person would actually wear — the signal falls apart. Bone is a lousy electrical conductor: the skull has roughly 80 times the electrical resistance of brain tissue and behaves like a spatial low-pass filter. A source maybe two square centimetres wide inside your cortex smears into a blur across 10 or more centimetres of scalp. Researchers describe scalp EEG as reading the brain "through frosted glass": electrodes on the cortex (ECoG) resolve detail down to 1–2 millimetres, while electrodes on the scalp are stuck at around 3 centimetres. That blur is the wall between "implant that reads the intent to move" and "device that reads a mind" — and for now, the only way through it is a drill.
Two things are about to take that wall down
The wall comes down when two curves cross. The first is sensors: non-invasive instruments getting drastically more sensitive over the coming decades. The second — and this is the one Asimov couldn't have priced in — is AI that learns to reverse the blur, reconstructing the pattern the skull smeared away. That second curve is already moving, and fast. Look at what has been decoded from brain activity in a single recent stretch:
| Year | Who | What they decoded | Through the skull? |
|---|---|---|---|
| 2023 | UT Austin (Tang & Huth, Nature Neuroscience) | The gist of language you hear or merely imagine — reconstructed as text by a GPT-style model | No — non-invasive fMRI |
| 2023 | Meta AI (Nature Machine Intelligence) | Which sentence a person was hearing, picked from 1,000+ candidates | No — MEG/EEG |
| 2023 | Stanford (Willett et al., Nature) | Attempted speech → text at 62 words per minute | Yes — cortical implant |
| 2023 | UCSF (Chang et al., Nature) | Speech and a talking facial avatar at 78 words per minute | Yes — surface implant |
| 2024 | Neuralink (PRIME trial) | The intent to move → full cursor and computer control | Yes — N1 implant |
Read the top two rows again, because they're the ones that matter for my prediction. In 2023, a team at the University of Texas at Austin — Jerry Tang and Alexander Huth — built a semantic decoder that reconstructs the continuous gist of what a person is hearing, or silently imagining, from non-invasive fMRI, using a transformer model of the same family as ChatGPT. No implant. No surgery. It recovers meaning, not exact words — but it recovers it from outside the skull, by training AI to make sense of a blurred signal. Meta's researchers did something adjacent, identifying which sentence you'd heard from magnetoencephalography. The non-invasive rows used to be empty. They aren't anymore.
Now extrapolate. Not five years — seventy-five. Take sensors that keep getting better and models that keep getting better at un-blurring noisy data, and run that to the end of the century. I don't think you land on "control a cursor." I think you land on "read a mind." The honest counter-evidence, which I'll keep on the record too: that same UT Austin decoder only worked on a cooperating subject, and could be defeated when the person simply thought about something else. That's a real safeguard, and it exists today. The open question is whether it survives another seventy-five years of progress.
Why the mind-reader will be a robot
Here is the step that turns a brain-science story into a robot story — and the part that genuinely gives me chills. A robot is, at bottom, a walking computer. It is a device with power, with space for hardware, and with a body to carry it all. Whatever the best non-invasive sensor of 2080 turns out to be, and whatever the largest brain-decoding model of 2080 turns out to be, a robot can carry both — built in, always on, far beyond what a person would ever strap to their own head. A human might need a clumsy wearable. A robot just is the wearable, plus the legs.
So the first practical mind-readers won't be people wearing gadgets. They'll be machines that already have a body, a power supply, and a brain — and the world is filling up with exactly those machines for entirely unrelated reasons. That is, almost to the letter, what Asimov described: a humanoid robot that happens to be able to feel what you feel. Our own R. Daneel Olivaw and R. Giskard Reventlov — not as a metaphor, but as a product category. And the good news, the part I genuinely look forward to, is that the link is two-way: a machine that can read your intent is a machine you can finally talk to without words.
My prediction, on the record and dated: before the year 2100, there will be robots that can read a human mind, and with which we communicate mentally — unless we deliberately choose to stop it. What I spent twenty-five years certain was a fairy tale now looks, to me, like a schedule.
The part that actually keeps me up: reading vs. writing
A mind-reader is unsettling. A mind-writer is something else entirely. Reading a mind is passive — eavesdropping. Writing to one is active — editing. And Asimov, as usual, drew the line before any of us got here. His telepathic robot Giskard didn't only read minds; he could nudge them. And in the Foundation saga, the danger isn't surveillance — it's control. The Mule, a mutant "mentalic," conquers worlds by reaching into people and adjusting their emotions directly; the hidden Second Foundation is an entire order of mentalics who quietly steer the course of humanity. That's the nightmare version: not a robot that knows you're afraid, but a robot that can make you unafraid, or loyal, or compliant.
My hope — and the whole reason I think it's worth saying this out loud now, while it's still science fiction — is that we build the second set of guardrails before we build the capability. We gave robot bodies the Three Laws (in fiction, at least). Robot minds reaching into ours will need an equivalent, and a firm one: let them read, let them converse, but never let them write. The cooperation safeguard in today's decoders is a hint that it's possible to keep a person in control of their own head. We should treat that as a design requirement, not an accident we got lucky with.
Why I'm putting this here
RobotTesters' What's Next section exists for calls you can check later. I've already put a $1.5 trillion forecast on the record where it can embarrass me if I'm wrong. This is the same exercise, pushed to its eerie conclusion: a specific, falsifiable, dated claim about where the robots we review are heading. A thought is electrical and chemical. Machines already read the easy part of it through a hole in the skull, and AI is learning to read the hard part without one. Robots are the obvious vessel. Put those together and "they will read our minds" stops sounding like a line from a novel and starts sounding like a release date.
Maybe I'm early. Maybe I'm wrong, and the skull keeps its secrets forever. If so, this page will say so, and you can laugh at it in 2099. But if a robot ever does look at you and simply know — remember that Asimov called it first, and that we were among the few who said, while it still sounded mad, that it was coming. I just hope it only ever reads.