Personnel reshuffles in the world of AI continue to set trends. Naomi Bashkansky, who previously conducted research at OpenAI, has announced her move to the startup Conduit, where she will lead fundamental development. The project sets itself an ambitious goal — to train neural networks to "non-invasively" read human thoughts.
Bashkansky has publicly stated that within a few years, thoughts will become the dominant interface for interacting with artificial intelligence. In her concept, she emphasizes: this format should expand human capabilities, not replace them. According to her, it will not be a soulless automated assistant, but a "natural, joyful extension of yourself."
The timeline the researcher outlines looks quite concrete. By 2027, she predicts the emergence of technology that allows transmitting thoughts to one's AI agent via Bluetooth. The system will interpret signals into prompts and actions. By 2030, in her estimation, Conduit's telepathic technologies will begin mass adoption in industry, and by 2035, AI will no longer be perceived by humans as "something alien."
In her reasoning, Bashkansky relies on forecasts from industry leaders — Elon Musk and Sam Altman — who predict the inevitable convergence of the biological brain and digital intelligence. She notes that the level of "superhumanity" of future AI directly depends on how superhuman we ourselves decide to make them, betting on symbiosis rather than replacement.
Technically, Conduit intends to train models based on brain activity. The principle, according to Bashkansky, is simple: brain activity goes in, and a semantically close result of what the person was doing at that moment (e.g., written text) comes out. The startup is based in San Francisco and focuses on neural interfaces that do not require surgical intervention. Data on the company's funding and valuation have not yet been disclosed.
However, there has been no shortage of skepticism. Under Bashkansky's post, users have already criticized the term "non-invasive," rightly noting that reading thoughts without direct access to the brain is an extremely challenging task. This echoes Meta's recent breakthrough, which in June presented a system for reconstructing text from brain activity recordings without surgery. Word-level accuracy reached 61% versus 8% with previous methods, confirming that the path from laboratory experiments to a real "telepathic" interface is still fraught with difficulties.
Analyst's comment: Conduit's idea looks futuristic, but it is precisely such projects that shape the long-term market agenda. However, it is worth remembering: between demonstrating "mind reading" in controlled conditions and a stable commercial product lies a chasm. Investors and enthusiasts should view Bashkansky's statements as a vision, not a guaranteed roadmap. Commentators' skepticism about "non-invasiveness" is absolutely justified — without clear technical details, this is for now just loud marketing.