Reardon believes that the company’s edge lies in its unusually strong crop of neuroscientists. These researchers will conduct lab experiments while the AI team builds models informed by their discoveries; the algo team, meanwhile, might unearth clues that help the neuroscientists. “You don’t really know if you understand something until you can build it, implement it in silicon,” says Josh Morgan, a Flourish neuroscientist. They say they’re open to publishing some of their original research.
“Fundamentally, the company is looking for the algorithms underlying intelligence,” says Jacob Vogelstein, who is managing partner of Catalio Capital.
Reardon tells me that his team has identified paths to near-term revenue that exploit recent brain research. They’re developing a hippocampus-inspired way to handle memory that will allow the company’s models to learn without extensive training data, he says. The algorithm team has built a model that can learn continuously and is working on embodying it in “the kinds of devices you carry in your pocket,” he says. He adds that he’s negotiating with a major chip manufacturer to put the model on silicon.
In early May, the Flourish scientists held an all-hands meeting in their New York office. Reardon and Williams are seated at a conference table with about a dozen others, including senior advisor Wayne, who’s visiting from London. The scientists are debating six potential experiments. These are big swings that require the purchase of multimillion-dollar microscopy machines and years of work.
The discussion of the experiments brings up biological phenomena as widespread as how rabies spreads in the cortex and the neurobiology of birdsong. They debate whether they should examine molecules and synapses or focus on bigger-scale cells or circuits. Are analyses of connectomes in the mouse brain sufficient for some purposes, or would only human brains do?
The verdict, for the moment at least, seems to be to try it all. “The takeaway is that we want to do data collection across the nano, micro, and meso scales to support the discovery of the core algorithm,” says Sean Bittner, a computational neuroscientist who also worked with Reardon at Meta.