Personnel changes in the world of AI continue to set the tone for the industry. Naomi Bashkansky, who previously worked in research at OpenAI, has officially announced her move to the startup Conduit, where she will lead the research division as one of the founders. The new team has set itself an ambitious goal — to train neural networks in non-invasive "mind reading" of humans.

In her blog, Bashkansky outlines a bold forecast: by the end of the current decade, thoughts will become the primary interface for interacting with artificial intelligence. In her view, such technology is intended not to replace humans, but to become their natural extension, expanding cognitive capabilities rather than substituting for them.

Roadmap of the "mental" revolution

The researcher proposes specific timelines for development. According to her analysis, by 2027 users will be able to transmit their thoughts to an AI agent via a Bluetooth headset. The neural network will interpret these signals into commands and actions. By 2030, scaling of the technology and the introduction of Conduit's telepathic solutions into the mass industry are expected. And by 2035, in Bashkansky's view, AI will cease to be perceived as something alien, becoming an organic part of human thinking.

Notably, in her reasoning she references the vision of industry leaders, including Elon Musk and Sam Altman, who predict the inevitable convergence of the biological brain and digital intelligence. Bashkansky's key thesis: "How superhuman our AIs become directly depends on how superhuman we decide to make them. This is a future where humans remain at the center of the process."

Technical foundation and challenges

From a technical standpoint, Conduit plans to train models based on brain activity data. The logic is simple: the input is neural signals, the output is the semantic result of a person's actions, such as written text. The task is to predict the final outcome as closely as possible to the real intention.

It is worth noting that Conduit is based in San Francisco and focuses on developing neurointerfaces that do not require surgical intervention. However, the company has not yet disclosed data on funding and valuation, leaving questions about the practical implementation of such bold plans. Skepticism is already emerging in the comments on the announcement: critics rightly point out that high-quality "mind reading" without direct access to neurons is an extremely complex engineering challenge, despite progress in this area.

My view: the transition of a specialist of this level from OpenAI to a niche startup is a strong signal to the market. However, statements about "non-invasive telepathy" by 2035 look optimistic. Recent successes by Meta in decoding text from brain activity (61% accuracy versus 8% with old methods) show that the technology exists, but the distance to a commercial "mental" interface with a Bluetooth headset is enormous. Most likely, we will see hybrid solutions, but within the stated timelines this looks like an ambitious declaration rather than a working plan.