As I take my cat for a stroll, I see a familiar form in the distance. Is the form I’m approaching Obama? Let’s forget about the low likelihood of passing Obama in San Diego and the myopia that cripples my ability to see far distances and go through a scenario of what might be going through my head as I make this judgment. Over time. I am accruing sensory information that either sways me to say, “well gee wilikers, that’s Obama!” or, conversely, “aww shucks, that’s not Obama.” According to the drift-diffusion model of decision-making, as I accrue evidence that the form is Obama, an internal likelihood running count is shifted towards the Obama decision, and as I accrue evidence against that scenario, it is shift towards the “some random person” decision. Thus, at any moment, I have to maintain a memory of the evidence I have already gathered and add evidence to that memory. At some point I reach a decision threshold at which enough cues have added up in one direction that I feel confident with making a decision. If this is real life, I have set my threshold much too low and impulsively yell, “Heyo Bama!” only to realize within the next 20 steps that I have been approaching a tree stump.
We make numerous judgments like these within a day, some of which are incorrect. There are two main possibilities that could underlie this imperfection. The first is that the sensory inputs one receives are imperfect either due to the inherent noisiness of the stimuli, the variability introduced during sensory processing in the brain, or the imperfection of adding the evidence to the estimate kept in memory. The second is that the memory of the signals one has received is noisy and drifts with time. Dr. Carlos Brody’s lab at Princeton University (Brunton et al., 2013) set out to distinguish between these two possibilities using both rats and humans. Subjects were asked if they heard more tone clicks on the right or left side in an auditory discrimination task in which tone clicks were presented at random time intervals but with a fixed overall rate.
The authors modeled each subject’s responses using nine parameters, two of which were sensory noise and memory diffusion noise. Sensory noise will add uninformative variance proportional to the sum of the amplitude of the tone clicks and memory diffusion noise will add variance proportional to the stimulus duration. Thus the variable timing allows the two possible noise sources to be pulled apart. By using best fit models given the data of each subject, Brody et al. found that in 13 out of 19 rats and for all of the human subjects (n=3), the best fit value for the memory diffusion noise was zero, implying that the subjects could maintain evidence perfectly and thus that decision making imperfections stem from variability in sensory processing or in adding evidence to the estimate.
Please join us in the CNCB Large Conference room on Tuesday, January 28th at 4:00pm to learn more from Dr. Carlos Brody about “Neural substrates of decision-making in rats.”
Margot Wohl is a first year student in the UCSD Neuroscience Graduate Program. She is currently rotating in the laboratory of Dr. Jeffry Isaacson.
Brunton B.W., Botvinick M.M. & Brody C.D. (2013). Rats and Humans Can Optimally Accumulate Evidence for Decision-Making, Science, 340 (6128) 95-98. DOI: 10.1126/science.1233912