Personnel changes in the field of artificial intelligence continue to set new trends. Naomi Bashkansky, who previously conducted research at OpenAI, has announced her move to the startup Conduit, where she will lead fundamental developments in the field of neural interfaces. The project sets itself an ambitious goal — to train models to non-invasively "read" human thoughts.

In her analytical review, Bashkansky predicts that by the 2030s, thought activity will become the dominant channel of interaction with AI. In her view, such a paradigm is aimed at expanding human capabilities, not replacing them. "These will not be super-intelligent automated assistants that you try to keep up with, but a natural extension of your personality," the researcher emphasizes.

Technological roadmap: from Bluetooth to full integration

According to the presented roadmap, by 2027 users will be able to transmit their thoughts to their AI agents via a Bluetooth connection, where they will be converted into commands and actions. By 2030, the technology is projected to move beyond laboratories and become mainstream, and by 2035, AI will cease to be perceived as a "foreign" element.

Bashkansky refers to the vision of industry leaders, including Elon Musk and Sam Altman, who predict the convergence of neuroscience and machine learning. "How superhuman our AI will be depends on how superhuman we decide to make it. I am drawn to the scenario where the human remains at the center of the process, and AI expands our capabilities rather than replacing us," she notes.

How it works and challenges

Conduit, based in San Francisco, is developing neural interfaces without surgical intervention. Technically, the approach looks elegant: brain activity as input, and a semantically relevant result as output, such as text that a person intended to write. However, skepticism is already being voiced in the professional community: critics rightly point out that non-invasive thought reading without direct access to neurons is an extremely complex task.

Significantly, similar developments are already underway at major corporations. In June, Meta presented a system for decoding brain activity that restores typed text with 61% accuracy at the word level, which is many times superior to previous methods (about 8%).

My expert commentary: Bashkansky's announcement is not just a personnel story, but a signal of a shift in priorities in the AI industry. The transition from "conversational" interfaces to neurocomputer bridges could be the next big leap, but the path from laboratory prototypes to commercial products will take longer than optimistic forecasts. The key question is not only the accuracy of algorithms, but also the ethics of privacy of thought processes, which remains outside the scope of discussion for now.