wednesday, 14 march 2018—12:15
Computational models of decision making applied to spontaneous action initiation and volition
How does the brain decide when to act, or whether to act in
the first place, when decisions are not dictated by immediate external
imperatives? The mechanisms that govern so-called “self-initiated”
decisions-to-act remain poorly understood. One promising avenue for
shedding light on self-initiated action is the use of formal
computational models. Integration-to-bound models, which have long
been used in research on cued decision making, have recently taken
center stage in the study of self-initiated action.
Integration-to-bound models at a minimum involve temporal integration
over an input signal (the “evidence” or “imperative”) plus Gaussian
white noise. In the context of self-initiated action, the “noise” in
the model plays a more prominent role because the imperative to move
is often weak or absent. At the same time, neural data show that noise
in the brain is not white, but rather is long-term correlated, with a
spectral density of the form 1/f^ß (with ß close to 1.5) across a wide
range of spatial scales. I will discuss ongoing work using
computational models to account for self-initiated action and
“volition” (the conscious feeling of an urge to initiate movement)
while staying true to the spectral properties of noise that are
demonstrably present in real neural data. The results expose new
perspectives for models of the RP and for accumulator models in
general by suggesting that the spectral properties of the stochastic
input to the accumulator should be allowed to vary, in keeping with
the nature of real biological neural noise.