Personnel reshuffles in the world of artificial intelligence continue to surprise. Naomi Bashkansky, who previously conducted research at OpenAI, has announced her move to the startup Conduit, where she will lead fundamental development. The new goal is ambitious to the point of audacity: to train AI models to non-invasively "read" the human brain.

Vision for 2035: Thoughts as an Interface

In her blog, Bashkansky outlined a radical forecast: as early as 2027, we will be able to transmit our thoughts to our AI agent via Bluetooth, and it will interpret them into concrete actions and prompts. By 2030, the technology will move beyond laboratories and become mainstream, and by 2035, AI will no longer be perceived as "something alien"—it will become a natural extension of humans.

"These won't be super-smart automated interns you're trying to keep up with. This will be a joyful, natural extension of your personality," the researcher emphasizes. In her view, the key principle is augmenting humans, not replacing them. She cites Elon Musk and Sam Altman, who predict the inevitable merging of biological and digital intelligence.

Technically, Conduit's approach looks elegant: brain activity goes in, and a semantically precise result comes out—for example, text that a person intended to write. The company is based in San Francisco and develops neural interfaces without surgical intervention, although funding data has not yet been disclosed.

Skepticism and Context

The community has already reacted to Bashkansky's statements with a fair amount of criticism. The main question is how realistic "non-invasive" mind reading is without direct access to neurons. Opponents rightly note that this is an extremely complex engineering challenge requiring a breakthrough in neuroimaging.

Notably, Meta recently presented a system that reconstructs typed text from brain activity without surgery. Average accuracy reached 61% compared to 8% with previous methods—progress exists, but the distance to full-fledged "telepathy" remains enormous.

My view: Conduit's ambitions are impressive, but the market has already seen loud promises in the field of neural interfaces. The key risk lies not in a lack of talent, but in the physical limitations of non-invasive methods. If the project truly solves this problem, we will witness a tectonic shift in HCI. For now, though, it's a beautiful hypothesis with a decade-long horizon.