On New York University Week: How do animals make inferences?
Christine Constantinople, assistant professor of neural science, tests rats with juice rewards.
Faculty Bio:
Christine Constantinople is an Assistant Professor in Neural Science at NYU. After obtaining a PhD in neurobiology from Columbia University, and postdoctoral research at the Princeton Neuroscience Institute, Dr. Constantinople joined NYU in 2019. Her lab studies the mechanisms by which neural circuits compute and represent cognitive variables for decision-making.
Transcript:
To survive in a changing and unpredictable world, animals cannot simply react to stimuli they
encounter, but must make inferences that help them predict future outcomes, like when they
might encounter food or predators. This process is among the most important cognitive
operations that brains perform. To study how brains perform inference, we trained laboratory
rats on a task in which they revealed how much they valued different amounts of juice rewards
based on how long they were willing to wait for them. The task had hidden reward states with
different average rewards. In high reward states, rats were offered 3-5 drops of juice, and three
drops was disappointing. But in low reward states, rats could only expect 1-3 drops of juice, so
three drops was a big win. Indeed, rats waited less time for three drops in high versus low
reward states. We showed that rats inferred the reward state when deciding how long to wait:
for instance, they immediately changed their wait times following a single reward that revealed a state transition.
We trained hundreds of rats to perform this task using a high-throughput training facility
developed in my lab. The facility, which I call the “rat factory,” uses computers to train about 100rats per day. A major goal is statistical power and rigor.
We found that inactivating a part of the brain called the orbitofrontal cortex or OFC made rats worse at inferring the reward state. We used state-of-the-art silicon probe technology to record electrical signals from thousands of neurons in OFC, and identified neural correlates of single trial inferences. These results show that OFC supports inference based on prior knowledge, and lay the groundwork for figuring out how the OFC interacts with other brain regions to mechanistically implement inference.
