Naomi Bashkansky leaves OpenAI to develop thought-to-text AI at Conduit
Naomi Bashkansky left OpenAI to join Conduit as a Founding Researcher, where the company is developing thought-to-text models trained on non-invasive neural data. Bashkansky presents a vision of the future from 2027 to 2035, in which human-AI interaction increasingly shifts toward direct thought communication via neural headbands, progressing from simple decoding to more complex scenarios involving brain writes and seamless AI integration as an extension of human consciousness.
This project is far from unique. You likely know Elon Musk's Neuralink, but there are many more. Well, it looks like the era of better self-understanding and perhaps understanding of others is inexorably approaching.
How do thought-to-text models work and what data does Conduit collect to train them?
Thought-to-text models decode neural signals captured by non-invasive sensors (typically neural headbands) into text commands and thoughts. Conduit collects large quantities of non-invasive neural data to train models that recognize how the brain encodes various types of intended communications. The non-invasive approach requires no surgical intervention, unlike potential invasive alternatives.
What is the difference between reading and writing in neural technology according to the text?
Reading means extracting information from the brain (intentions, emotions, internal monologue, memories), while writing enables intervention in the brain (such as delivering audio or altering neural function). Non-invasive approaches require no surgery, whereas invasive methods do. Conduit is exploring invasive variants of both directions.
How should the role of AI evolve in its relationship to humans according to Bashkansky's vision?
Rather than being perceived as an autonomous tool or coworker, AI should become a natural extension of human consciousness—a sixth sense and additional limb. In the optimal scenario, human and AI operate in tight symbiosis where AI directly empowers human capabilities rather than replacing them, with decision-making remaining within human control.
- Understanding is the new bottleneck — geoffreylitt.com 80 % match
- An Alien Mind: OpenAI's Chief Scientist on Intelligence We Don't Fully Understand — openai.com 77 % match
- Mark Zuckerberg outlines a vision for the era of artificial intelligence in rare personal essay — meta.com 75 % match
- Naomi Bashkansky
- OpenAI26
- Conduit
- Neuralink
- Elon Musk5
- GPT-7