Naomi Bashkansky, who previously conducted research at OpenAI, has announced her move to the startup Conduit, where she will lead one of the key areas — the development of models for "non-invasive" reading of human brain activity. The San Francisco-based company has set itself an ambitious goal: turning thoughts into an interface for interacting with artificial intelligence.

In her blog, Bashkansky outlined a vision according to which, by 2027, users will be able to transmit their thoughts to their AI agent via Bluetooth, and the system will interpret them into commands and actions. According to her forecasts, by 2030 Conduit's telepathic technologies will become the industry standard, and by 2035 AI will no longer be perceived as something alien.

The philosophy of "augmentation," not replacement

The researcher's key thesis is a humanistic approach to technology development. She emphasizes that such systems should not turn into "super-smart automated interns" that humans try to keep up with. Instead, in her words, it will be "a natural, joyful extension of yourself." Bashkansky cites forecasts from industry leaders, including Elon Musk and Sam Altman, who predict the inevitable convergence of the biological brain and digital intelligence.

Technically, Conduit's approach looks elegant: on the input side — data on brain activity, on the output side — a semantically accurate reconstruction of what the person was doing at that moment, whether it was writing text or performing an action. However, this approach has obvious skeptics. In the discussions under Bashkansky's post, a number of users doubted the very possibility of "non-invasive" mind reading, noting that without direct access to neurons, the accuracy of such systems remains highly questionable.

Notably, the idea itself is not new: in June, Meta already presented a system capable of reconstructing typed text from brain activity recordings without surgical intervention, achieving 61% accuracy at the word level compared to 8% with previous methods. However, Conduit goes further, aiming at full-fledged "telepathy."

My comment: Claims about "mind reading" by 2035 sound extremely optimistic, given the current state of neuroscience. Even the most advanced non-invasive methods, such as fMRI or EEG, provide only coarse spatial resolution. Nevertheless, the development vector itself — from keyboard-based interfaces to intention-based interfaces — is inevitable. The only question is how many decades will actually be required to overcome physical limitations.