I’m leaving OpenAI to build telepathy
A former OpenAI researcher’s plan to build “telepathy” via non‑invasive brain–computer interfaces, using headset‑collected neural data to turn thoughts into text and back, is drawing intense scrutiny. Commenters question the technical plausibility of decoding rich thoughts from noisy EEG-style signals, and worry more about surveillance, thought policing, advertising, and interrogation uses than about convenience gains over keyboards. Many see the project as emblematic of a broader AI hype cycle in which ambitious visions outpace both current science and serious consideration of long‑term social and ethical risks.
Technical feasibility of non‑invasive “telepathy”
- Many doubt that a non‑invasive headband/helmet (EEG/fNIRS-style) can get clean enough signal for thought‑to‑text; described as mostly “noise” and analogous to listening to a conversation from outside the stadium.
- Reports of current data collection (Craigslist recruits, rented basements, janky setup) reinforce skepticism about data quality.
- Others note that brain‑imaging has progressed (e.g., reconstructing images from scans) and think noisy data plus scale and better hardware might be enough for incremental progress, though still far from literal telepathy.
- Some argue invasive methods are likely required for high‑bandwidth, reliable interfaces. Non‑invasive is seen as more scalable but technically limited.
- Timelines like “a couple years” for mainstream thought‑based interaction are widely viewed as highly optimistic.
Privacy, surveillance, and dystopian risk
- Strong concern that streaming thoughts to corporate/LLM infrastructure enables unprecedented surveillance: thought crimes, automated interrogation, upgraded “polygraphs,” corporate productivity monitoring, border screening.
- Advertiser access to mental streams, thought‑triggered purchases, and “ads in your dreams” are recurring fears.
- People with intrusive or unwanted thoughts (e.g., OCD) see this as nightmarish.
- Satirical comments imagine on‑device “compliance” models, electric shocks for unapproved thoughts, and subscription‑tier thought allowances, underscoring fear of abuse.
Value proposition and possible benefits
- Some see obvious value for people who cannot speak or type (locked‑in, severe motor disabilities), and for faster, lower‑friction interaction with computers and AI agents.
- Enthusiasts imagine “typing by thought,” bypassing tedious tasks, and richer human–machine symbiosis.
- Skeptics question whether this is actually better than speech‑to‑text, given output reading speed and the role of speaking/writing as part of thinking and self‑editing.
Ethical, cultural, and philosophical concerns
- Many find the vision viscerally disturbing or authoritarian, describing it as cyberpunk, Black Mirror, or “torment nexus” territory.
- There is criticism of “AI safety” people building mind‑reading tech, and of an “inevitabilist” mindset that justifies moral hazards.
- Some see the project as primarily a fundraising narrative rather than a realistic near‑term product.
- Use of terms like “telepathy” and “AI as a natural extension of me / sixth sense” is challenged as misleading or bordering on “AI psychosis,” while others frame it as standard tool‑as‑body‑extension.
Law and regulation
- Discussion touches on whether such tech would implicate the Fifth and Fourth Amendments in the US, with references to existing precedent on thermal imaging and data collected by private vendors.
- Several argue brain‑interface tech itself—not just AI models—needs strong regulation before it normalizes.