There is a reason why tech companies are clamoring for data. Artificial intelligence, or AI, is only as good as the data you feed it. Recommendation algorithms like those used by Netflix and Amazon need data from many users. Language models like ChatGPT start with massive amounts of data—crawling the internet from Wikipedia to Reddit for training. It isn’t just about quantity;  data quality and cortical coverage are important. You would not train a language model for healthcare workers using a chatroom for Harry Potter fanfiction, and Google Maps would be of limited usefulness if it only contained street information about major highways. 

Brain computer interfaces, or BCIs, are specialized medical devices that supply brain data for AI in order to enable new therapies. These devices stream real-time data from the brain and—together with AI—can allow people with severe paralysis to move robotic arms, control computers, and even speak again. The same devices can also read out brain data related to chronic pain and depression and may help people with those conditions better manage their symptoms. BCIs can also send data back to the brain through electrical stimulation.

Just like other AI applications, the quality and quantity of BCI is important:

Quality and Quantity

‍There is an old saying in computer science “garbage in, garbage out” and this is particularly true for BCI. The closer the sensors are to brain-activity data, the better the signal. And that means getting as close as possible to brain cells, or neurons, themselves. Devices that use tiny microwires to reach down into the brain and listen to the chatter of individual neurons have the highest quality data. Implantable devices located under the skull and on top of the brain have better, lower noise signals. Wearable devices can detect a little bit of information, but they need to run slowly to separate signal from noise.

‍

It's important to note that you can’t substitute quantity for quality. Millions of sensors outside of the brain do not equal 1,000 sensors on the surface of the brain, which are still not as powerful as 100 sensors that can listen to single neurons. From a data perspective, the best case scenario is quality and quantity: hundreds to thousands of sensors with single-neuron resolution. 

Not all BCI applications require an equal amount of data. Cursor control, for instance, does not require very much. This is the reason why the Neuralink device, after a publicized device failure, was still able to provide cursor control to their clinical-trial participant. Speech decoding and dexterous robotic hand control require more data. Specifically, several groups have shown that even with the highest quality data—single neuron data—the performance of these applications substantially increases with the use of additional sensors. 

The Future of BCI

The most successful BCI clinical trials to date are still using a brain-implant technology developed in 1989, called the “Utah Array.” Leading academic centers are using 1989 hardware with 2024 software to make incredible breakthroughs. Last year, several people who had lost the ability to speak regained the ability to communicate by using BCI plus AI. This is only the tip of the iceberg. The world is about to change. Next generation BCI companies like Paradromics and Neuralink are now building new hardware platforms that leverage 30 years of microchip innovations. The future of BCI will be modern hardware plus modern software, and just like competition between Intel, AMD, and Nvidia has powered modern computing, a succession of new BCI releases from competing companies will unlock hundreds of billions of dollars worth of new applications.

BCI is the ultimate “plus AI” technology. It connects our most powerful algorithms with our most powerful data source—the human mind.

As new BCI platforms emerge with even greater speed, quality, and coverage, the scope of what BCI can do will expand to include, and one day exceed, the present extent of human activities. The medium-term applications include a complete overhaul of mental health. The long-term applications will keep scientists, entrepreneurs, and ethicists busy for decades. 

If you have any questions, please reach to media@paradromics.com.