Mind Reading, Between Promise and Reality: Brain-Computer Interfaces

About sixty words a minute: that is the speech rate achieved by a participant with amyotrophic lateral sclerosis through electrodes implanted in the cortex of the brain, in a study published in Nature in 2023. The previous record for brain interfaces was 18 words a minute. Are we on the threshold of "mind reading"? Let us look at what the papers said.

What Did the Implant Do?

The implant recorded activity in speech-related regions of the cortex, and a learning system translated it into words. The average was 62 words a minute, but the error rate was 9.1% with a 50-word vocabulary and rose to 23.8% with a 125,000-word vocabulary, far above speech-to-text systems (about 4–5%). The researchers describe their work as a proof of concept, not a clinical system: it involved one participant, needed daily training on hundreds of sentences, and required recalibration as neural activity shifted.

The benefit is plain for those who have lost the ability to speak. Here artificial intelligence plays the translator between the language of neural signals and human language.

And Does It Read What We Have Not Said?

Another team worked with functional magnetic resonance imaging, without surgery, and their system reconstructed the gist of heard or imagined speech as phrases that approximate the meaning rather than the wording. But the results were bound by two conditions.

The first is cooperation: decoding worked only when the brain's owner took part in training and in use, and when he or she resisted, say by counting in multiples of seven or imagining another story, performance fell in some cases to near zero. The second is that each brain has its own language: a system trained on another brain recovered only 1–5% of segments, against 65–82% when trained on the same person. It is as if each brain had its own dialect. Training also takes several sessions in a large scanner.

So the scientific picture today is that general "mind reading", working covertly on anyone, does not exist, and nobody knows whether it ever will. The researchers themselves argue that the need for cooperation protects privacy for the present.

Why Are Laws Moving?

Caution nonetheless precedes the technology. On 25 October 2021 Chile published Law 21.383, amending Article 19 of its Constitution, the first constitutional text to protect neurorights according to the law firm Carey; it requires that a law set limits on the use of these technologies, with particular care to safeguard brain activity, and it extends to information derived from it. On 12 November 2025 UNESCO's General Conference adopted the Recommendation on the Ethics of Neurotechnology. It is non-binding; it describes neural data as private and potentially highly sensitive, and calls for banning its use in recommender systems that steer behaviour and for barring neuromarketing during sleep.

What Should We Hold To?

First, treatment before enhancement: the surest benefit today is to those who have lost speech or movement, and they most deserve the effort and funding. Second, free, informed and revocable consent, because what is gathered from brain signals is unlike other data. Third, limits that bar the use of such data in advertising and in influencing behaviour. These frameworks preserve human dignity and privacy, and we dealt with the foundations of privacy in "Privacy in the AI Era".

And if interfaces one day move from treatment to enhancement, the question of fairness will arise: who can afford the upgrade? It is a question we put now and cannot yet answer, and it should not wait until it happens. What is permissible in such enhancement, and what is not, is a matter for qualified Sharia scholars and medical specialists.