Report

Teaser, summary, work performed and final results

Periodic Reporting for period 2 - ACTINIT (Brain-behavior forecasting: The causal determinants of spontaneous self-initiated action in the study of volition and the development of asynchronous brain-computer interfaces.)

Teaser

In everyday life, decisions to act are often initiated spontaneously, without any specific external imperative indicating whether or when to perform an action. The neural origins of spontaneous self-initiated actions and their relation to conscious intentions pose a...

Summary

In everyday life, decisions to act are often initiated spontaneously, without any specific external imperative indicating whether or when to perform an action. The neural origins of spontaneous self-initiated actions and their relation to conscious intentions pose a challenging problem for basic neuroscience research as well as for the engineering of neuro-prosthetic devices (brain-computer interfaces, or BCIs). How do spontaneous decisions-to-act emerge from the tangled complexity of the brain, and what determines whether and when an action is initiated - especially in the face of absent, incomplete, or noisy evidence? The “when” question has received less attention in the literature on decision making, even though it is of profound significance to our understanding of human behavior.

Recent work has introduced formal computational models to the study of self-initiated action, and used them to account for the slow buildup of neural activity that is known to precede self-initiated actions. This buildup, primarily found in pre-motor areas of the brain, evolves over the last one second or so before movement and is called the Bereitschaftspotential or “readiness potential” (RP). The RP has always been presumed to reflect a process of “planning and preparation for movement”. The RP became a source of controversy when, in the 1980s, Benjamin Libet argued that the conscious decision to act came about 300 ms or more *after* the onset of the RP. This was widely taken to imply that the brain had already “decided” to initiate the action well before one became conscious of having decided, casting doubt on our intuitions about conscious volition (the conscious decision should come first).

The introduction of computational models to this field of research exposed a very different interpretation of the RP: According to the “stochastic decision model”, when the external imperative to produce a movement is weak, as is the case in the kind of experiment that Libet performed, then the precise moment at which the decision threshold is crossed leading to movement is partly determined by ongoing sub-threshold fluctuations in brain activity. Time locking to movement onset ensures that these slow fluctuations are recovered in the event-locked average in the form of a gradual buildup. By this account, the real decision may come about near the end of the RP (quite close to the onset of movement) rather than at the onset of the RP, allowing for the possibility that the decision to initiate movement and the conscious feeling of having decided come about at the same time. This new interpretation of the RP is important in large part because of the significance that attaches itself to the topics of “free will” and personal responsibility, but also because it gives us a promising new vantage point from which to approach the study of volition and self-initiated action.

The overall objectives of the project are the following:
• Develop a new variant of the stochastic decision model that accounts for the spectral properties of neural data.
• Test specific divergent predictions that emerge depending on whether the early tail of the RP is modelled as the output or the stochastic input to the accumulator.
• Use machine-learning to map the time course of movement-preceding neural activity that is specifically and causally related to the initiation of movement.
• Refine the brain-behavior forecasting methodology and use it to determine the causal relationship between known antecedents of uncued movement and the fact of performing the movement.
• Determine the precise relationship between the readiness potential, the lateralized readiness potential, spontaneous fluctuations in cortical activity, and cortical mechanisms of evidence accumulation.
• Empirically probe deeper questions of causality and spontaneity through two high-risk / high-gain experiments.

Work performed

The project is divided into three main initiatives:
Initiative 1: theoretical and hypothesis-driven studies
Initiative 2: systematic mapping / exploratory studies
Initiative 3: deeper questions of spontaneity and causality

The summary of work performed so far will be grouped into these three initiatives.

Initiative 1: theoretical and hypothesis-driven studies

1.1 Develop a new variant of the stochastic decision model that accounts for the spectral properties of neural data:

Integration-to-bound models are currently among the most widely used models of perceptual decision-making due to their simplicity and power in accounting for behavioral and neurophysiological data. They involve temporal integration over an input signal (“evidence”) plus Gaussian white noise. However, brain data shows that noise in the brain is long-term correlated, with a spectral density of the form 1/f^α (with typically 1 < α < 2), also known as pink noise or ‘1/f’ noise - i.e. it is NOT Gaussian. Surprisingly, the adequacy of the spectral properties of drift-diffusion models to electrophysiological data has received little attention in the literature. One of the main goals of this initiative was to develop a generalization of the leaky stochastic accumulator model that can account for the spectral properties of brain signals. This goal has been achieved and we are now in the final stages of peer review. We developed a generalization of the leaky stochastic accumulator model using a Langevin equation whose non-linear noise term allows for varying levels of autocorrelation in the time course of the decision variable. We derive this equation directly from magnetoencephalographic data recorded while subjects performed a spontaneous movement-initiation task. We then proposed the simplest model of accumulation of evidence that accounts for the ‘1/f’ spectral properties of brain signals, and the observed variability in the power spectral properties of brain signals. This is a drift-diffusion equation with nonlinearities in both the drift and diffusion coefficients. This work was done in collaboration with mathematician Ramón Guevara Erra.

1.2 Test specific divergent predictions that emerge depending on whether the early tail of the RP is modelled as the output or the stochastic input to the accumulator:

Recent evidence suggests that the readiness potential may reflect subthreshold stochastic fluctuations in neural activity that can be modeled as a process of accumulation to bound. One element of accumulator models that has been largely overlooked in the literature is the stochastic term, which is traditionally modeled as Gaussian white noise. While there may be practical reasons for this choice, we have long known that noise in neural systems is not white – it is long-term correlated across a broad range of spatial scales. I explored the behavior of a leaky stochastic accumulator when the noise over which it accumulates is temporally autocorrelated. I also allowed for the possibility that the RP, as measured at the scalp, might reflect the input to the accumulator (i.e., its stochastic noise component) rather than its output. These two premises led to two novel predictions that I empirically confirmed on behavioral and electroencephalography data from human subjects performing a self-initiated movement task. In addition to generating these two predictions, the model also suggested biologically plausible levels of autocorrelation, consistent with the degree of autocorrelation in our empirical data and in prior reports. These results expose new perspectives for accumulator models by suggesting that the spectral properties of the stochastic input should be allowed to vary, consistent with the nature of biological neural noise. This work is completed and has been published in the journal eNeuro.

1.3 Test other predictions made by the stochastic decision model regarding entrainment:

In physics “entrainment” refers to the synchronization

Final results

We currently have three studies under way and two in development, and we expect to have results from all of them before the end of the project. One study, being headed up by doctoral student Bianca Trovó, is looking at parametric variation in the shape of the readiness potential as a function of the amount of time allowed for making a spontaneous self-initiated movement. Normally, in a standard readiness potential experiment, subjects are allowed to take as long as they like to produce a spontaneous movement. In this experiment we impose a range of different time limits, from 2 to 16 seconds, and predict a higher amplitude early RP for trials with a longer time limit. This experiment tests a prediction of the stochastic decision model.

A second study along similar lines looks at how the amplitude of the RP changes when the precise time of movement onset is influenced by unnoticed sensory regularities. This is a follow-up to the “entrainment” experiment described above in section 1.3. We will bring this experiment into the M/EEG lab with a loudspeker hidden behind the MEG dewar. As in the previous experiment we will play faint noise theough the speaker, designed to mimic background noise that might be produced by the intercom. Subjects will perform a self-initiated movement task in the presence of this noise, which can be steady or modulated at a certain frequency. We predict a decrease in the amplitude of the RP when self-initiated movements are performed in the presence of the modulated noise. The reason for this is that the noise presumably influences the precise moment of movement onset, and thus movements are less synchronized with the internal fluctuations that, according to theory, give rise to the RP.

We will soon begin to collect data for study 3.2 (above) on brain-body coupling in the initiation of movement. This study asks the profound question of whether or not proprioceptive feedback from muscles is involved in the decision to initiate movement. A positive result would count as strong evidence in favor of the dynamical systems view of action initiation: Decisions to act are not made in the brain and then transmitted to muscles, but rather decisions to act are formed in the brain+body as a whole.

Another study tries to capture sub-threshold fluctuations in brain activity under parametrically varying levels of uncertainty. Subjects are asked to detect faint bursts of white noise against a background of steady “pink” noise. The amplitude of the target noise bursts will be fixed within each block of trials, but will vary randomly between blocks (the order being counterbalanced across subjects). The task will be to perform a simple finger extension immediately, each time a target sound is played. The interval between target sounds will be drawn from a Poisson distribution, and subjects will know how often (but not when) to expect target sounds based on a supra-threshold practice session. In light of prior research, we expect a difference in pre-stimulus activity, associated with the detection of weak stimuli. However, in contrast to these prior experiments, our experiment involves detection in continuous time, i.e. a decision about “when” to move. Thus we further predict that the amplitude of movement-preceding activity in the time-locked average over pre-motor areas will be inversely proportional to the strength of the imperative to act (given by the contrast between the target bursts of sound and the background noise).

We are just now preparing to start behavioral testing of a new study looking at the neural basis for randomness in choice and its relationship with stochastic variability in brain activity. A random component of choice has been suggested as an important mechanism for exploration, potentially complementary with a directed information-seeking counterpart. Similarly, in a competitive context, unpredictability becomes extremely relevant (and complementary with predicting the opponent). In