The following HTML text is provided to enhance online
readability. Many aspects of typography translate only awkwardly to HTML.
Please use the page image
as the authoritative form to ensure accuracy.
COLLOQUIUM ON VISION: FROM PHOTON TO PERCEPTION
is weak, information arrives slowly and the better strategy is to integrate over a long period of time (21,41). Even if the optimal strategy is followed, however, the monkey is apt to be less certain of its decisions at low coherences than at high coherences. Thus the dynamics expected of the decision process correspond to the dynamics of the neural signals illustrated in Fig. 3, and the certainty of the monkey's decision appears correlated with the probability level achieved by LIP neurons by the end of the stimulus period.
We therefore suggest that the evolution of predictive signals in LIP comprises a neural correlate of decision formation within the central nervous system. In the context of a discrimination task like ours, the decision process is simply a mechanism whereby sensory information is evaluated so as to guide selection of an appropriate motor response. To use a legal analogy, the decision process is akin to the events that occur inside a jury deliberation room. Sensory signals, in contrast, are analogous to the evidence presented in open court, while motor signals are analogous to the verdict announced after the jury has completed its deliberations. The neural events in LIP are suggestive of the process of deliberation—sifting evidence and forming a decision—as indicated by the gradual evolution of the signals over time, the dependence of the time course on stimulus strength, and the dependence of predictive activity on stimulus strength (i.e., certainty of the decision).
Practically speaking, such distinctions are difficult to make unless the accumulation of sensory information and formation of the decision can be spread out in time and cleanly isolated from execution of the motor response. If, for example, our monkeys viewed only 100% coherent motion and were allowed to make an eye movement as soon as a decision was reached, then sensory, decisional, and motor signals would be densely entangled in only a few hundred milliseconds of neural activity. Distinguishing among these signals may be virtually impossible under such conditions.
Importantly, we are not proposing that decisions in our task are actually formed in LIP. LIP may simply follow afferent signals from another structure or group of structures where decisions are initiated. We are, however, suggesting that neural signals in LIP may reflect the dynamics of decision formation and the certainty of the decision, regardless of where the decision is initiated. If so, neural activity in LIP provides a window onto the decision process that will permit interesting manipulations in future experiments. Obviously, we have not yet addressed the critical question of whether LIP plays a causal role in performance of this task. Microstimulation and inactivation techniques may allow us to investigate this possibility in future experiments.
Finally, we note that the present analyses leave several interesting questions unexplored, mostly because the population histograms in Figs. 3 and 4 may obscure interesting heterogeneity in the data. Are some cells influenced more strongly than others by sensory or motor signals? Are the firing rates of individual cells modulated smoothly, as suggested by the curves in Figs. 3 and 4, or do rates change abruptly at different times on different trials, thus yielding the smoothly increasing probability values in the population curves? These questions will be addressed in future analyses.
A Look at the Future
If the effort to identify neural substrates of a decision process is ultimately successful, a host of fascinating questions will be brought into the realm of physiological investigation. If, for example, LIP integrates motion signals to form a plan to move the eyes in our psychophysical task, a precise pattern of circuitry must exist between the direction columns in MT and MST and the movement fields of LIP neurons. In essence, LIP neurons with movement fields in a particular region of visual space should be excited by columns in MT and MST whose preferred directions point toward the movement field. Columns whose preferred directions point away from the movement field should suppress the response of the LIP neuron. The latter columns should, of course, excite LIP neurons whose movement fields are located elsewhere in space.
Realize that this is merely a restatement of the logic of the task: for the monkey to perform correctly, saccade-related neurons anywhere in the brain should be activated only when directional columns in the motion system signal a preponderance of motion toward their movement fields. Realize also that we are not implying that this circuitry must connect MT, MST, and LIP directly; motion signals could be processed through frontal cortex or other structures before activating parietallobe neurons. The logic of the task demands, however, that such connections exist regardless of the length of the pathway. Tracing such precisely patterned connections with physiological techniques would be a major step toward identifying the circuitry underlying the decision process in our task. Experiments that combine microstimulation of MT and MST with unit recording in LIP may shed light on the circuitry connecting the structures.
Monkeys can be trained to base eye movements on a wide range of sensory signals. For example, our animals could be trained to make rightward or leftward saccades depending upon the color of the random dot pattern rather than its direction of motion. In this case, LIP may continue to contribute to the formation of oculomotor plans, but the sensory signals that differentially activate one or the other pool of LIP neurons must originate from color-sensitive neurons rather than direction-selective neurons. Thus a different, but no less precise, pattern of connections from occipital to parietal cortex would underlie the decision process.
Raising the ante a bit further, monkeys should be able to learn both the color and direction discrimination tasks, and to alternate tasks from one block of trials to the next (perhaps, even, from one trial to the next). In this situation, the effective connectivity between the occipital and parietal cortices must be flexible. One pattern of connections should operate in the color version of the task, but a very different pattern should operate during the motion version. Obviously, higher-level control signals, probably related to visual attention, must engage and disengage these connections on a fairly rapid time scale if the monkey is to perform appropriately. Development of physiological techniques for monitoring the formation and dissolution of such circuits with fast temporal resolution is a high priority for future research (e.g., refs.42 and 43).
To conclude, systems neuroscientists have unprecedented opportunities to make significant discoveries concerning the neural basis of cognition. Though we currently fall short of Mountcastle's vision cited at the outset of this paper, the future promises substantial progress toward this goal.
We are grateful to Daniel Salzman for assistance in design of the experiments and to Jennifer Groh and Eyal Seidemann for participating in some of the experiments. These colleagues as well as Drs. John Maunsell, Brian Wandell, and Steven Wise provided helpful comments on the manuscript. We thank Judy Stein for excellent technical assistance. The work was supported by the National Eye Institute (EY05603) and by a Postdoctoral Research Fellowship for Physicians to M.N.S. from the Howard Hughes Medical Institute.
1. Hubel, D. H. & Weisel, T. N. ( 1977) Proc. R. Soc. London B198, 1–59.
2. Zeki, S. M. ( 1978) Nature (London)274, 423–428.
3. Kaas, J. H. ( 1989) J. Cognitive Neurosci. 1, 121–135.
4. Allman, J., Jeo, R. & Sereno, M. ( 1994) in Aotus: The Owl Monkey, eds. Baer, J. F., Weller, R. E. & Kakoma, I. (Academic, New York), pp. 287–320.
5. Felleman, D. & Van Essen, D. ( 1991) Cereb. Cortex1, 1–47.
6. Wurtz, R. H., Goldberg, M. E. & Robinson, D. L. ( 1980) Prog. Psychobiol Psychol.9, 43–83.