14 de October de 2021
Brain-computer interface systems
The day Facebook will read minds
«The day Facebook will read minds». Originally published in agenciasinc.es
Several technological giants are in the race to decode brain activity into written text. This is not only a major breakthrough for optimising the quality of life of people with various motor diseases, but also a potential risk to our privacy.
Dear reader, if you have accessed this article attracted by its headline, I am sorry to tell you that you have fallen victim to the clickbait -a technique to get visits to a website- that has been flooding neuroscience communication lately. You’ve probably read headlines similar to this one over the last few years, but I’m happy to give you some good news: the day when Facebook (or any other tech giant) can read our minds has not yet arrived… and may never come.
This increasingly widespread belief probably comes from the ‘neurohype’ about the studies recently funded by this company to the team led by Edward Chang at the University of California in San Francisco (USA). This group has developed a brain-computer interface (BCI) system to decode brain activity related to spoken language production and translate it into written text.
The neural activity is collected using the electrocorticography technique (ECoG, an electrode plate on the surface of the cerebral cortex). The information is then processed – thanks to a machine learning method commonly used for translation between languages – and finally the text is generated.
The results obtained by Edward Chang and his collaborators are astonishing: they have managed to translate dozens of sentences with error rates even lower than those of professional transcribers.
The great novelty of this BCI lies in the fact that it considers brain activity as another language and, instead of focusing on the decoding of phonemes (individual letter sounds), it translates complete sentences processed as a single linguistic unit.
The results obtained by Chang and his collaborators are astonishing: they have managed to translate dozens of sentences (such as “Tina Turner is a pop singer” or “those thieves stole thirty jewels”, among others) with error rates even lower than those of professional transcribers.
This is certainly a major breakthrough in improving the quality of life of people suffering from ALS, locked-in syndrome or other motor diseases, and has already been successfully applied by Chang himself in a study published in the New England Journal of Medicine.
However, Facebook has recently announced that it will no longer fund the development of this BCI and will focus its efforts on an electromyography device that would be placed on the wrist (not on the head) and whose short-term success seems more likely.
A very long-term aspiration
The company’s main goal was to find an interface that would allow typing at a speed of 100 words per minute, that would be silent (i.e. operate by thought alone), and that would be non-invasive so that it could be purchased by all types of users. But achieving this goal turns out to be, at best, a long-term aspiration.
Facebook's goal was to find a fast, silent and non-invasive interface. But achieving this goal turns out to be, at best, a long-term aspiration. There are technological limitations
There are technological limitations, such as the fact that it requires a long training period prior to use and that large amounts of data must be collected from thousands of users in order for the machine learning algorithms used to be effective;
It is also a highly invasive interface, involving intracranial surgery to place the electrodes on the cerebral cortex. And finding a transcranial (i.e. external to the head) system that does not entail significant signal loss is certainly the biggest of these three challenges.
What exactly does ‘mind reading’ entail?
It is essential to clarify that both Facebook’s silent interface and similar interfaces currently under development (such as MIT’s AlterEgo) rely on decoding motor signals sent to the articulatory organs of language, located in the mouth.
In other words, they require the user to explicitly state the words, even if they do so mentally (i.e. they must pronounce them to themselves), and are not able to decode the process prior to the production of the words.
There are no interfaces currently available to decode thought in this way, nor is there likely to be in the foreseeable future. However, 'mind-reading' does not just involve reading explicitly articulated thoughts.
To consider that these BCIs can read deliberations or thought processes that precede the expression of ideas is a misunderstanding of their true scope: there are currently no interfaces that decode thought in this way, nor are there likely to be in the foreseeable future.
But ‘mind-reading’ is not just about reading explicitly articulated thoughts. In fact, current neuroscientific techniques make it possible, with increasing scope and precision, to decode individually various components and mental states intimately related to thought formation, such as emotions or images.
The dark side of decoding
Jack Gallant, one of the world leaders in neural decoding, is even of the opinion that not long from now (two to five decades) there will be non-invasive BCIs capable of decoding thoughts that are not being thought by means of explicitly articulated words. Although we do not know if he is being overly optimistic, we must take seriously the potential risks that decoding techniques may pose to our privacy.
For example, given that the application of this technology requires a period of prior training based on repetition, it seems that recurrent thoughts, addictions and compulsions may be particularly accessible to decoding. This does not necessarily have to be negative if we think in diagnostic and therapeutic terms, but it does generate debate about the revision of informed consent procedures on the part of patients.
Recurrent thoughts, addictions and compulsions may be particularly accessible to decoding. This need not be negative if we think in diagnostic and therapeutic terms, but it does generate debate about informed consent on the part of patients.
On the other hand, it is very important to bear in mind that the most accurate and effective decoding is currently carried out by neuroimaging techniques (especially functional magnetic resonance imaging or fMRI) that require the use of very bulky instruments and under controlled conditions in a laboratory or hospital, so that decoding thoughts without the user’s knowledge is still impossible.
However, certain non-invasive technologies are beginning to be developed that are portable and show great potential for neural decoding, such as functional near-infrared spectroscopy (fNIRS) or functional ultrasound (fUS).
More worrying than the future development of all these systems, however, is the fact that numerous neurotechnological devices have proliferated and are easily accessible to any consumer through well-known online shops.
These devices, which usually consist of helmets or headsets with electrodes, can decode information on parameters such as concentration and stress levels, or even process mental passwords. While it is true that these BCIs are still expensive and rather inaccurate, it is possible that, as is often the case when a technology is developed and consolidated, their price will decrease and their effectiveness will improve.
The right to our neurodata
As the use of consumer-accessible neurotechnological devices becomes more widespread, there will be increasing problems related to health (as they are acquired and used without technical or medical supervision) as well as privacy.
As recently explained by a group of experts from the International Neuroethics Society, there are certain characteristics of brain data (also called neurodata) that call for specific regulations: the data of numerous users are integrated and it is difficult to identify each user’s data; and users cannot control access to such data to decide for what purpose it can be used.
It is unreasonable to spread apocalyptic and sensationalist interpretations of the possible negative repercussions of brain decoding, but it is not reasonable to take the issue lightly.
These two features could clearly make it difficult for users to exercise their rights of access, rectification and erasure (digital oblivion). For too many years now, we have been giving away (or rather giving away) huge amounts of data through our mobile phones, our behaviour patterns on social networks and our internet searches.
In this regard, dear reader, I have some bad news: the day when Facebook will be able to read minds… is long overdue. This and other tech giants already know a lot about how we think and live, information that they exploit for their commercial and strategic purposes. Reversing the trend, despite growing social awareness, seems unfortunately very complicated.
It would be essential for us to be properly informed about the risks, limits and implications of the collection and use of this type of data through devices whose use may not take as long as we think to proliferate. Countries such as Chile and Brazil are even taking regulatory steps in this regard.
In short, it is not reasonable to extend apocalyptic and sensationalist interpretations of the possible negative repercussions of brain decoding, but it would be wise not to take this issue lightly.
About the author
José Manuel Muñoz is a researcher at the Mind-Brain Group of the Institute for Culture and Society (ICS) at the University of Navarra. He also works at the International Centre for Neuroscience and Ethics (CINET) created by the Fundación Tatiana.