In lieu of an abstract, here is a brief excerpt of the content:

In order to understand how intentions are encoded in the human brain, one should first consider the general principles for identifying neural representations of any kind of mental state. One of the key assumptions of modern neuroscience is that every mental state is realized by brain activity. Instead of a dualistic model that would allow an independence of mental from neural, neuroscience postulates that for each mental state it is possible to identify a specific neural state that “encodes,” “represents,” or “correlates with” it. One can think of the brain as the “neural carrier” in which the mental states occur, and the different states of the carrier (i.e., brain) encode different mental states (Haynes, 2009). Several theoretical approaches have been proposed for identifying the neural correlates of specific mental states. These typically distinguish between “enabling” or background conditions and “content-specific” conditions .1 An enabling condition would be a necessary condition for a mental state, but it would also be necessary for a number of other mental states. For example, wakefulness (along with its neural correlates in the brainstem) is necessary for sensory percepts, memories, and intentions alike. Similarly, activity in the inferior frontal junction might be necessary for intentions (Brass et al., 2005), but it is active in task switching across various different intentions and thus does not distinguish between the specific intentions that are being implemented (Haynes et al., 2007). Then there are specific conditions that are necessary only for a subset of mental states or even one individual mental state to occur. In the visual system, specific patterns of neural activity are necessary for the conscious percept of a very specific image. Similarly, it is reasonable to assume that intentions are also coded by specific patterns of activity. Such specific conditions have been termed the “core neural correlates of consciousness” (NCCs). This means a minimal set of neurons with a “direct correlation” or “tight mapping”2 with a specific class of experiences 4 The Neural Code for Intentions in the Human Brain: Implications for Neurotechnology and Free Will John-Dylan Haynes 158 John-Dylan Haynes (Chalmers, 2000; Block, 2007; Koch, 2004; Haynes, 2009). Importantly, every class of mental states (sensory percepts, memories, intentions, etc.) can have a different core NCC. The big question is: What is the core NCC for conscious intentions? What is needed is a way to translate the general formula mentioned above into a scientific research program. How would one be able to identify the core NCC of a specific intention?3 How would one know which of the approximately 85 billion neurons in the human brain (Williams & Herrup, 1988; Azevedo et al., 2009) are relevant? And what is the coding format? This paper will focus on coding principles for intentions rather than providing an overview of the entire cognitive neuroscience of intentions, for which several excellent reviews are available (e.g., Andersen, Hwang, & Mulliken, 2010; Blankertz et al., 2006; Brass & Haggard, 2008; Burgess, Gilbert, & Dumontheil, 2007; Haggard, 2008; Miller & Cohen, 2001; Passingham, Bengtsson, & Lau 2010; Ramnani & Owen, 2004; Sakai, 2008). As with any mental representation, various neural coding formats could be possible for intentions (figure 4.1). First of all, the neural code for lossless encoding of six intentions in a neural carrier could in principle be either univariate or multivariate. Univariate means that it would be based upon a single aspect of neural processing—say, the mean activity level of a single neuron or small group of neurons. Multivariate means that it would be based on multiple parameters of neural processing, such as the activity of a set of individual neurons. In figure 4.1 (left) the gray scale indicates a hypothetical univariate code where each intention is represented by a specific level of firing of cells. In another domain, perception , a univariate code is used to code the perceived intensity of a sensory stimulus (Haynes, 2009). For intentions, one might envisage specific graded dimensions to be encoded in this way (say, e.g., the degree of commitment to an intention or the time delay after which an intention will be implemented). An alternative is to use a multivariate code where a pattern of activity in a group of neurons encodes the different intentions. Multivariate codes come in different flavors. The most important distinction is between sparse codes and distributed codes. Sparse codes use one (and only one) neuron for each intention, like a labeled line (or “cardinal cell”) that is active only when this intention occurs. Examples of labeled line codes from other...

Share