The Media Lab professor spoke with KSJ fellows about building wearable devices that can predict seizures and tell us when we’re sliding into depression.
For some, the words “artificial intelligence” conjure up a vision of a future where robots have taken all of our jobs. For others, it brings to mind emotional machines — robots programmed to demonstrate a semblance of humanity. Behind that first image lies the idea that computers can do things much better and faster than we can. Behind the second lies something less tangible — perhaps a desire to understand how we work, what makes us human.
Rosalind Picard, the founder and director of the Affective Computing research group at the MIT Media Lab, works in an area of AI that takes advantage of both of these features. Her projects utilize AI’s unique ability to gather lots of data and identify patterns, and apply it to read the often-indecipherable web of signals that our bodies and minds relay.
Picard recently spoke to KSJ fellows and guests about two such projects that she says have the potential to save lives.
The first project grew out of her group’s work building sensors that measure skin conductance, subtle electrical changes across the surface of the skin. Wired to a wristband, a sensor can produce a 24/7 feed of the wearer’s sympathetic nervous system activity.
Picard stumbled upon a discovery when someone wearing the bands had a seizure, and the measure of their skin conductance jumped through the roof. The peaks on the graph were higher than any Picard had seen before. She worked with the chief of neurosurgery at Children’s Hospital in Boston to investigate whether there was a connection between this huge sympathetic nervous system response and the seizure. In their study, 100 percent of Grand Mal seizures, the kind that cause a loss of consciousness and convulsions, showed the same massive peaks. Based on that research and subsequent studies, Picard developed a product called Embrace, a watch that automatically notifies a designated third party if the wearer is showing signs of a seizure. This kind of intervention could potentially prevent cases of Sudden Unexpected Death in Epilepsy, or SUDEP.
KSJ fellow Elana Gordon, who reports on addiction, asked Picard whether anyone has thought about using sensor technology to prevent drug overdoses. Picard said that her group has measured responses in cocaine users, and found patterns in their skin conductance and the temperature of the skin. She noted that factors unrelated to cocaine use can cause similar responses, but with machine learning, if you can gather enough data, the signals start to separate from the noise. As the patterns become more defined, prediction will improve.
This is the idea behind the second project Picard discussed: using AI to prevent depression. Our phones, the little computers that we carry around in our pockets, can learn an awful lot about us. What if we could use that information to alert us to when we’re at risk of sliding into depression?
She ran a study with MIT students, and in addition to skin conductance, she tracked data like amount of sleep, light exposure, exercise, drinking, texting behavior, and she surveyed how they were feeling throughout the day. The goal was to create a deep neural network that would make predictions: Based on your data from today, are you likely to feel happy or sad tomorrow? Are you likely to feel stressed or calm?
One problem with all of this incredible pattern-recognition is that we might not always want to know the results. Do people want an app that tells them they are going to be sad tomorrow?
Picard wasn’t sure. But if algorithms learn which habits lead to dreary moods, an app could give recommendations that might change the outcome, like “call your friend today.”
AI might be able to help us understand our complicated, messy selves, but as of yet there’s no replacement for human empathy.