A researcher leaves OpenAI to work on telepathy.
Naomi Bashkansky joined Conduit as a founding researcher at the beginning of August. The startup is training models to infer linguistic content from non-invasively measured brain signals.
When I read that, my own reaction bothered me more than anything else.
I had only one question: When?
A few years ago, this would have been the plot of a mediocre science-fiction film. Today, it is a job description.
And we keep scrolling.
What Conduit is actually building
Conduit does not have a finished telepathy product. The company describes a system in development. That distinction matters, because there are often a few uncomfortable years between an impressive demo and a reliable product.
According to Conduit, the team has collected around 10,000 hours of neuro-language data from several thousand people. Participants wear non-invasive research helmets and speak or type while their brain signals are recorded.
The models are supposed to detect semantic content in the seconds before a thought is spoken or typed. The published examples are still imprecise. “The room felt colder” becomes something closer to “There was a slight breeze.”
That is remarkable. It is not the same as complete access to someone's private inner monologue.
“Mind reading” is the big phrase. What we currently see is a statistical reconstruction of linguistic meaning from measured signals. Impressive enough. We do not need to make it bigger than it is.
Still, real people are sitting in a San Francisco basement wearing helmets that weigh almost two kilograms, supplying data 20 hours a day, seven days a week.
Improvised, uncomfortable, and very real.
“If” became “when”
The bigger change is happening in our own heads.
I have worked in digital business since 1999. I have seen enough technology promises arrive much later, in a completely different form, or not at all.
Scepticism was usually a sensible default.
AI is shifting that position. Plenty of promises remain hard to believe. At the same time, the speed of the last few years has pushed our personal boundary of what feels imaginable further and further out.
Text, images, voices, software, and autonomous workflows. Every new capability trains us to dismiss the next one as impossible a little less quickly.
Now thought itself is becoming an input method.
After keyboards and voice, even that suddenly sounds like the next logical step.
Science fiction has become a roadmap.
2035? Maybe. Perhaps sooner, perhaps much later.
But hardly anyone says: This will never happen.
The strange part is our reaction
This news should knock us off balance for a moment.
A former OpenAI researcher is working on translating thoughts directly into text and AI commands.
Telepathy.
Yet most of us react with surprising calm. We do not laugh. We do not treat it as impossible. We think: Yes, that will probably happen.
Maybe not in 2035. Maybe later.
But it will happen.
And I find that almost stranger than the technology itself.
Our idea of what is technically possible has shifted completely within a few years. Telepathy is no longer an absurd idea. It is a development project with a startup and a timeline.
We adapted to the unimaginable so quickly that even telepathy now sounds like the next product announcement.
The technology is not the only question
The more plausible thought-to-text becomes, the less useful it is to ask only whether it works. Three other questions matter more.
- What does the system actually detect?
A rough semantic intention, a formulated sentence, or statistical patterns that happen to look good in a lab? - Who controls the data?
Brain signals are not ordinary usage data. You can change a password. Your neural patterns, less so. - Who decides how it is used?
A voluntary communication aid is very different from expected productivity monitoring at work.
These are not reasons to stop the research. They are reasons not to hand our judgement over to it.
AI First does not mean using every new capability immediately. For me, it means understanding early enough what it can do, where its limits are, and which decisions must remain human.
May you live in interesting times
There is a saying often sold as an ancient Chinese curse:
“May you live in interesting times.”
It probably did not come from China at all. Perfect for our time: questionable origin, repeated millions of times, and somehow still true.
Interesting times are rarely comfortable. They are fast and full of developments whose consequences we only understand after we are already in the middle of them.
That is exactly where we are.
A researcher leaves OpenAI to work on telepathy. A startup collects 10,000 hours of brain data. And our surprisingly calm reaction is:
Sounds plausible. Let us see when it works.
These times are not easy. They are frightening too. But they are certainly interesting.