Neural decoding

Algorithms that translate recorded brain activity into the movements, words or commands a person intends.

• Calibration comes first: the user attempts or imagines known actions while activity is recorded, giving labelled examples to learn from.
• Models map patterns to outputs in real time; speech decoders predict speech sounds, then a language model assembles the likeliest words.
• In 2023 speech implants reached 60–78 words per minute, against about 150 for natural conversation.
• Signals drift as electrodes shift and neurons change, so decoders need frequent recalibration; they read attempted actions, not arbitrary thoughts.