Elon Musk is the most bold Broligarch When it comes to brain-computer interfaces. But Mark Zuckerberg followed closely behind.
Musk was co-founded shortly after Neuralink – company Put fries into the brains of three peopleAnd Count – 2016, Yuan (and then Facebook) Adventure into Neurotechnology Researchannounced plans to build technology that will enable people Typing with your brain and Hearing language through their skin.
Since then, Meta-funded researchers figured out how Decode voice Activity recorded from surgical implantation electrodes inside people’s brains. While brain surgery may feel worthwhile for paralyzed people who want to restore communication skills, invasive devices like this are a tough sale for those who just want to type faster. The average person with commercial equipment may actually need to wear wearable and removable, rather than permanent.
Sign up here Explore the major and complex problems facing the world, and the most effective ways to solve them. Sent twice a week.
Yuan made efforts Building a consumer’s brain computer interface a few years ago: Brain reading headbands are not ready for prime time. Instead of directly developing new gadgets, the company invests in neuroscience research that burns slower. Their hope is that studying the brain will help them build better AI in what humans are good at, such as processing language. Some of this research still focuses on thinking readings: Specifically, decoding how the brain produces sentences.
But this month, The RMB has made a breakthrough.
and Basque Cognitive, Brain and Language Centerresearchers in the Meta Basic Artificial Intelligence Research (FAIR) laboratory were able to accurately decode unapplication sentences of brain signals recorded outside the skull – no surgery required.
This is Just in the laboratorycertainly. But these findings mark an important step towards wearable thinking devices Zuckerberg promised eight years ago.
And, with brain-to-text devices approaching commercial viability in the near future, we need to work hard to deal with what it means to the gatekeeper of the meta. In the lab, mind reading technology is expected to reveal previously unknown information about how our brains build ideas, make decisions and guide our actions. But worldwide, tech companies can abuse our brain data unless we create and enforce regulations to prevent them.
Meta can decode unmentioned sentences from the brain’s magnetic field
Until a few years ago, researchers could not decode self-evident language without implanting electrodes into the brain, which required surgery. In 2023, University of Texas scientists used fMRI and added a version of the AI model to power Chatgpt Key points for decoding unspeakable sentences From brain activity. But fMRI machines cost millions of dollars, Better A fully grown elephant confined their use outside the laboratory.
Since neuroscientists are often reluctant to attach recording devices to the inside of a person’s brain, most research in the brain involves the agents that measure certain neural activity itself. The fMRI scanner measures how much blood flows to the brain cells during its operation, which requires a bit of lag. Another method called magnetoencephalography (MEG) is used to measure magnetic field brain cells that produce when sending electrical signals. While neither of these techniques can track what a single cell does, they can both provide a rough snapshot of the brain’s activity pattern when someone performs a task, such as reading or typing. It’s cool that unlike fMRI, Meg can record his brain in nearly real time.
So the meta-researchers recruited 35 volunteers, sat in a MiG scanner, and typed sentences on the keyboard, which looked like a Hair loss chair in outer space. Some people also have an EEG (EEG) electrode gel on their face and scalp to record electrical signals radiated from brain cells through the skull.
Everyone’s brain activities help Training AI models Guess what they entered. Essentially, part of the model learned to match brain activity patterns with letters typing at the time. The researchers provided a bunch of Wikipedia articles for another part of the model to teach how their sentences work, and words that appear frequently in different situations. With this information, if someone intends to type “I love you” but their brain signals say “I love yoi” – probably because their brains actually cause them to make typos – the model can effectively automate that prediction , because it knows how to know letters and words should work in the context.
Using EEG, it is easier to portable than fMRI or MEG scanners, meta-researchers are able to use AI to decode the exact letters someone enters in one third of the time. This doesn’t sound particularly impressive until you consider that the EEG records brain cells through electrodes outside the skull, which are many separations away from the brain itself. Just like trying to stand outside and Put the glass against the wall: Given all the noise, even grabbing a third of the conversation is already challenging.
MEG captures brain activity more accurately than EEG, because magnetic signals from brain cells are not twisted by the skull like electrical signals. By feeding MEG data to their AI model, the meta-researchers accurately decode what 70% to 80% of people type, blowing the previous model out of the water. So if Meta wants to build a headband of thinking, Record magnetic field Probably their best choice.
Like fMRI, the MEG devices used in this study were huge and expensive. But a MEG scanner like a helmet weighs only a few pounds. Already exists yes More sensitive More than non-portable scanners. These portable MEG devices are only a few pounds heavier than a few pounds Mission 3Meta’s latest VR headsets, and looks stupid. Although these MEG devices still cannot work outside of special magnetic rooms (not yet used to the public), it is not difficult to imagine a future where they can.
Technology companies don’t protect brain data unless we make them
Meta is not the only technology giant to invest in neuroscience research. Google and Microsoft Both have teams dedicated to studying the brain, and Nvidia and IBM Both work with neuroscience research institutions.
AI and neuroscience fields have long. The brain has many features that technical developers want to copy in their computers, e.g. Energy efficiency and Learning without a lot of training data. Tech companies have built tools that neuroscientists want to use. (Ideas for using non-invasive brain scans Diagnose mental illness Neuroscience has been fashionable for decades. After all, if the diagnosis of depression and Quick EEG scan)
Here, Meta uses the brain data they collect, while people input to study how the brain transforms abstract ideas into words, syllables, and letters, with its long-term goal of figuring out how to help AI chatbots do the same.
Data support Long-standing assumptions Held by neuroscientists and linguists, proposed We give speeches from top to bottom. When I was about to say something, my brain first painted the whole thing (“I’m going to have lunch soon”), then amplified a word (“going”), followed by a syllable (“go-“). When I type, my brain focuses on each specific letter (“g”, “o,” …) because it tells my fingers how to do it. Meta sees these representations – context, words, syllables, letters – overlap during language production, peaking and fading at different times.
In theory, understanding language production will also help metadata implement their State the goal “Fixed[ing] Communicate for those who have lost their ability to speak. “There are indeed millions of people from Traumatic brain injury, Strokeor another Nervous system disorders It’s hard to talk about. A wearable device that makes communication easy again can be a huge positive force in someone’s life.
But we know that’s not their only motivation. For Silicon Valley, the brain also represents the ultimate obstacle between humans and their devices.
Quick sanity check: Meta’s targets will never be merged with computers (That’s Musk’s thing), but to sell portable removable headphones that someone can use to type or play video games. To show such a device, the Meta needs to cross two huge technical barriers, one or even greater moral barriers.
First, they need to decode self-evident ideas from outside the skull. Check.
Second, they need equipment that someone can afford, to stay in their own house and wear on their head. At present, this is still far away.
Most importantly, once these devices are present, we will need strong protections of people’s cognitive freedom, which is our fundamental right to control our consciousness. These safeguards are not after they go to the store. Now.
“Facebook is already very good at gazing at your brain without any electrodes or fMRI or anything else. They know most of your cognitive abilities, and you can learn from the way you use it.” Neuroethicist Roland Nadler, Telling my colleague Sigal Samuel in 2019.
Meta has used AI to infer Your mental health From your digital footprint. They use AI Logo, sometimes deletedposts about self-harm and suicide, can trigger involuntary “Health check“When they detect messages on Messenger or WhatsApp.
Given how much convenience we get by giving away personal data (delivering food, working remotely, connecting with friends online), many people Give up digital privacy common. Even if many people are uncomfortable with the amount of personal information the company receives from us, they I believe they can’t control it To their privacy.
Last year, neuroscientists, lawyers and lawmakers Begin passing legislation Neural data is explicitly included in state privacy laws. Some smaller neurotech companies have Collect brain data from consumer products – More powerful protection is needed before large companies like Meta can do this too.
Zuckerberg has been here in the past two months Trump Dollar. It is unlikely that his company will handle our most private data carefully, at least not mentioned.
But in this world, the meta-brand brain to text headband is now as normal as a keyboard, Share brain data It may feel like a prerequisite for participating in a normal life. Imagine a workplace where in the office, instead of giving you a monitor and keyboard, they give you a text-coded helmet and tell you to tie it. If thinking type becomes the default setting for computer systems, avoiding brain-to-brain text devices will feel like avoiding smartphones: of course. But it is certainly not the path with the least resistance.
As our psychological safety becomes less guaranteed, we need to determine whether it is worth letting tech companies colonize our last truly private space.
A version of this story originally appeared in The future is perfect communication. Register here!
