Exploring the Neurophysiology of Decisions

not another. However, unlike a motor neuron, its response should not oblige an immediate movement—just as we can decide what to do without acting impulsively. Matthew I. Leon and Michael N. Shadlen* Department of Physiology and Biophysics and Regional Primate Research Center Decision-Related Activity on the Sensory Side University of Washington Neurons in the visual and somatosensory cortex are Seattle, Washington 98195 known to encode the sensory variables that govern performance in discrimination tasks (reviewed by Parker and Newsome, 1998). Can such neurons teach us anyA prerequisite for any animal’s survival is the ability to thing about the decision process, beyond the simple fact make decisions about its sensory world. For a monkey that they encode the sensory data? It is often possible to respond to a call signaling imminent danger, or a to find sensory neurons that respond differently to the human to determine when to cross the street based on variety of stimuli shown in an experiment, but one that the flow of traffic, the brain must analyze the influx of responds differently to the same stimulus depending on fresh sensory information, integrate it with knowledge the animal’s interpretation might lend insight into the acquired recently or remotely, and select the approdecision process. priate behavior from perhaps many possible options. For example, neurons in the motion-sensitive regions This nonreflexive linkage between sensory input and of the extrastriate visual cortex have been shown to behavior involves interpretation and behavioral selecencode the signals that permit a monkey to discriminate tion, what we refer to as a decision process. In this between two directions of random dot motion (Britten review, we discuss recent developments in brain elecet al., 1992). In the paradigm depicted in Figure 1A, a trophysiology that lend insight into the neural basis of weak motion cue instructs one of two behaviors (e.g., decision formation. a saccadic eye movement to one or another target). Near psychophysical threshold, the same visual stimuA decision can only be studied in alert animals that lus often gives rise to two different interpretations (e.g., can perform sufficiently complex tasks. In addition to leftward and rightward decisions). Interestingly, from this basic requirement, there are several key elements trial to trial, neurons in MT, MST, and the superior that are essential for any psychological task devised to temporal polysensory area (STPp) exhibit variation in study a decision process. First, the sensory stimulus the number of spikes evoked by identical visual stimuli, should require an interpretation. Second, there must be and this variation has been shown to correlate with the some option as to whether one or another behavior monkey’s judgment of direction (Celebrini and Newensues. Finally, the conversion from sensory processing some, 1994; Britten et al., 1996; Thiele and Hoffmann, to behavioral intention should take place over an identifi1996). able epoch of sufficient duration to permit a physiologiThe data in Figure 2A were obtained from an MST cal dissection. neuron during 60 trials in which the monkey was shown Turning to physiology, what characteristics might one a random dot noise pattern containing no net motion. seek in neurons that play a role in a decision process? These trials represent a small fraction of the trials in In principle, the brain could contain neurons whose sole which stimuli of a variety of motion strengths were purpose is to form pronouncements about interpretashown to the monkey in random order. The responses tions, liberated in a sense from the fluctuating detail in tended to be larger, on average, when the monkey dethe sensory stream or the particular behavioral option cided in favor of the neuron’s preferred direction. This to be exercised. In fact, we know of no structure in suggests that the monkey’s decisions are influenced the brain that contains an abstract representation of an by the trial-to-trial variability in the responses of single interpretation or decision that is not either tied to an cortical neurons. It is a remarkable observation that coneffector system or dependent on the continued presstrains theories of cortical organization. It implies that ence of a sensory stimulus. Rather, the decision process the decision mechanism reads out from pools of sensory seems to emerge at the nexus of sensory and motor neurons whose variable discharge covaries weakly— processing—where sensory data give rise to a plan to otherwise the variability of any one neuron would have enact some particular behavior. The neural elements only a random relationship with the behavior (Shadlen therefore tend to one side or the other of what appears to et al., 1996). be a sensory–motor continuum. Accordingly, a decisionBinocular rivalry provides another example in which related neuron should modulate its response during the the same sensory stimulus gives rise to different interacquisition of sensory information that leads to one inpretations. Rivalry and other bistable perceptual pheterpretation or another. However, unlike a sensory neunomena are not usually construed as involving a deciron, it should continue to respond after the cue is resion process, but they are relevant. As shown in Figure moved—just as our decisions can persist after a sensory 1B, two visual stimuli, which are viewed separately by cue has vanished. Its response should also herald a the left and right eyes, give rise to an alternating percepparticular action that would indicate one outcome and tion in which one or the other eye’s image is seen. Scheinberg and Logothetis (1997) trained monkeys to indicate its perceptions by pressing one lever or another depending on which stimulus it sees. This is a challeng* To whom correspondence should be addressed (e-mail: shadlen@