Four years ago, language transition startup Unbabel worked towards creating a “brain-to-communications interface” that would allow businesses to understand (and be understood) by their customers and clients in multiple languages. Its Language Operations platform blends artificial intelligence (AI) with humans that evolve through high-quality transactions and conversations over time.
“You Had an Americano”
As a startup with $90 million in VC funding and annual revenues around $50 million while simultaneously surviving the COVID-19 pandemic, Unbabel has concentrated its efforts on better understanding the ways in which our brains have evolved.
TechCrunch’s Mike Butcher spent an entire afternoon in New York City, sitting in Unbabel’s offices alongside its founder and CEO Vasco Pedro and got a first-hand IRL demonstration of how this technology works.
“Sitting in a meeting room in a startup office in Lisbon, I silently typed the answer to a question only the person opposite would know the answer to. What kind of coffee had I asked for when I’d arrived at the office? A short moment later, without even moving or opening his mouth, the reply came back via a text message: ‘You had an Americano.’”
– Mike Butcher
Unbabel’s innovation team, led by Paulo Dimas, VP of Product Innovation, told TechCrunch that in time, we will start to see “the creation of the ‘uber cortex’” that the company believes “will be AI-powered…and is going to be existing outside of your biological brain.”
“You have your limbic system, you have your neocortex. But they’ve actually evolved over millions of years. They’re actually separate systems,” Pedro told TechCrunch.
As part of this research, Unbabel initially began looking into electroencephalogram (EEG) systems, which could be invasive to the human body – similar to some of the devices that Elon Musk’s Neuralink is actively exploring – yet, chose to instead focus on an EMG system, or electromyography, which measures muscle response and/or electrical activity in response to a nerve’s stimulation of the muscle.
These types of systems are easily accessible in the marketplace, with Amazon playing host to a number of them.
Pedro told TechCrunch that they wanted to explore a “non-invasive” mechanism instead, pointing towards something that could “more reliably capture some of the signals.” In other words, they wanted to depict EMG systems as a “gateway to brain interaction directly.”
In doing this, Pedro and his team paired an EMG system with generative AI to create a personalized LLM, or large language model, that is trained on a wide array of specific words or phrases directly relating to how an EMG’s wearer would react when thinking of a particular word or phrase.
Dubbed “Halo” (named after “halogram”), Unbabel uses a mobile app that runs on the wearers phone that enables access to a central hub that allows for the receiving of inbound communications with the personalized LLM as well as outbound communications. It currently leverages OpenAI’s ChatGPT 3.5.
Putting this to the test, Butcher asked Vasco what kind of coffee he’d asked for that morning (in an unseen text message), Vasco was sent those words via Halo’s AI voice to his earbuds, thinking of words like “Black coffee.” As a result, Halo matched Vasco’s physical response to the word, measuring it against the possibility of “Black coffee” equating to an “Americano” – all through the audio Vasco was receiving through his earbuds. Halo’s AI then sent the answer “Americano” to Butcher via a Telegram text message.
“The LLM expands what you’re saying. And then I confirm before sending it back. So there’s an interaction with the LLM where I build what I want it to say, and then I get to approve the final message,” explained Pedro.
Pedro clarified that at all times, Butcher, as well as all wearers of the EMG device, have “absolute control” of what they are outputting – “it’s not recording what I’m thinking. It’s recording what I want to say,” Pedro explained. He distinguished this type of approach from Musk’s Neuralink, which he says attempts to measure “subconscious interactions” – an invasive method that is done without the permission or choice of the wearer.
With that said, Halo’s underlying technology still has a number of obstacles to overcome, Pedro noted, adding that the current EMG device could eventually “be miniaturized.”
Currently, Unbabel is working with the Champalimaud Foundation in Lisbon, working on advanced biomedical research and interdisciplinary clinical care with respect to some of the world’s most heartbreaking diseases, including ALS and Cerebral Palsy.
Pedro says that their Halo prototype is already endorsed by Portugal’s major ALS association and is expected to deploy to its first ALS users by Christmas. Current technology for individuals suffering from ALS currently rely on eye tracking, which adds to a frustrating calibration process for its wearer.