Research project P7/33 (Research action P7)
The overarching goal of this project is to contribute to our understanding of the mechanisms of conscious and unconscious learning. Learning, that is, the ability to respond adaptively to changing circumstances is a fundamental ability for any organism. Thanks to recent advances in imaging methods, it has now become clear that the brain is a fundamentally plastic organ, the functional architecture of which is continuously modified through experience. From this perspective, one could thus argue that learning is a mandatory consequence of information processing: We learn all the time, whether we intend to or not. Learning takes many different forms. For instance, contrast learning the fact that Steve Jobs has just passed away with learning how to perform the complex movements involved in dancing the flamenco. Consider the differences between learning how to solve an arithmetic problem with learning a second language. Contrast a baby learning how to walk with an adult learning to play tennis, or a rat learning to avoid an electric shock with a human learning about the Hundred-years War.
In all such cases, one can see similarities, but also differences. All such cases involve changing the representational and behavioural repertoire of an agent, but each seems to appeal to fundamentally different processes. In particular, a long-standing distinction is that between associative theories and cognitive theories of learning. An important consequence of this diversity is that research on learning continues to be unproductively segregated into distinct subfields that entertain little communication with each other. For instance, research on implicit learning — the process whereby one learns without intending to do so and without awareness that one has learned, has so far made little contact with research on high-level, conscious learning such as involved in causal reasoning or in problem solving. Likewise, research dedicated to understand the basic mechanisms of learning in animals such as rodents remains almost completely disconnected from research dedicated to understanding basic mechanisms of learning in humans.
The domain as whole also remains very controversial. At least three such continuing controversies can be identified. The first concerns whether learning depends on associative mechanisms, on effortful, intentional, propositional-like reasoning processes or on a combination of both. Experimentally, recent, controversial evidence has indicated that even animals such as rats can exhibit inferential processing, thus questioning one of the fundamental tenets of associative theories. Conceptually, some theories in the domain assume that all learning is based on associative learning (e.g., connectionism), others assume that all learning is based on the manipulation of propositional symbol structures, and yet others assume that the two kinds of processes operate jointly or that they compete with each other. The second controversial issue is the role that awareness plays in learning, and in particular, the extent and limits of what can be learnt without awareness. The third controversial issue concerns the respective role of top-down and bottom-up learning mechanisms and the nature of their interactions (i.e., are phenomena such as conditioning penetrable to instructions?) Crucially, the poles of these different distinctions are often cast as correlated. Thus, we have one system that learns associations, automatically, in the absence of awareness, and that involves mostly bottom-up processes. The second system, by contrast, learns through hypothesis testing and inference, results in propositional representations that are available to consciousness, and involves top-down mechanisms.
Here, we propose to fundamentally reconsider the distinction. Instead of assuming that associative learning is always unconscious, automatic and bottom-up and that cognitive learning is always conscious, effortful and top-down, we propose instead that mechanisms of change operate continuously, at all levels of the cognitive hierarchy as well as over different times scales (i.e., over the time course of a single trial, over learning, and over development). From this perspective, the brain is continuously and unconsciously learning to anticipate the consequences of action or activity on itself, on the world, and on other people. There is considerable evidence for such predictive mechanisms in the human brain. This idea, in fact, forms the core of the Bayesian perspective on information processing and is at the heart of Friston’s free energy principle, according to which the brain continuously attempts to minimize “surprise” or conflict by anticipating its own future activity based on learned priors.
In this light, we will focus on exploring three central lines of research, as follows:
The first issue concerns the computational mechanisms and the neural correlates that subtend associative and cognitive learning, as well as their interactions. One set of questions concerns the extent and limits of each type of learning. Do associative learning mechanisms have sufficient power to account for all learning? Humans and animals share much of their neural organization, but also differ in many ways, most significantly perhaps through the fact that the former can leverage the expressive power of language to use and share symbolic structures through culture, so that they can, for instance, learn much more efficiently through instruction. Conversely, is there evidence for the involvement of symbolic, propositional-like representations in organisms that have typically been considered unable to carry out inferential processes? A second set of questions concerns the dynamics that underlie the transition between associative and cognitive learning (e.g., insight ; the role played by the sleep-wake cycle in consolidating memories ; the mechanisms of automatization in skill learning). There is a genuine puzzle involved in understanding how one can go from associative, subsymbolic learning to full-fledged cognitive learning.
The second issue concerns the relationships between awareness and learning. There continues to be considerable debate about the extent to which humans can learn without awareness, particularly in domains such as conditioning or implicit learning. Here, we will systematically probe the limits of what can be learned without awareness. The role that consciousness plays in learning, and, conversely, the role that learning plays in shaping the contents of consciousness, are thus fundamental, yet wholly unsolved issues. Are the mechanisms involved in conscious and unconscious learning subtended by the same or by distinct neural structures? What are the limits of learning without awareness? What is the influence of high-level, conscious processes on lower-level phenomena such as conditioning or habituation? How do we best characterize the differences and commonalities between human and animal learning?
A third issue concerns the respective influences of top-down vs. bottom-up processes and their interactions. Functions like executive control and attention are typically considered to involve “top-down” mechanisms associated with awareness, but there is now both evidence for the possibility of unconscious executive control as well as evidence for the fact that attention can dissociate from consciousness. Particular emphasis will be put on understanding (1) how high-level processes such as reasoning, instruction-following and awareness can modulate lower-level, associative learning, and (2) how low-level, unconscious learning can shape further conscious, intentional processing, such as involved in decision-making or in action.
These lines of research will be addressed over a series of eight interconnected work packages that are specifically aimed at leveraging the respective expertise of the partners. The network comprises experts on consciousness (P1 ULB—Cleeremans), on sleep and memory (P1 ULB—Peigneux), on language development (P1 ULB—Content), on literacy (P1 ULB—Kolinsky), on associative learning and evaluative conditioning (P2 UG—De Houwer), on intentional action and cognitive control (P3 UG—Brass), on animal learning (P4 KUL—Beckers) and on vision and perception (P5 UCL—Rossion).
Further, the network has solicited the expert collaboration of two foreign partners: Pr. Patrick Haggard (INT1, University College London) for his expertise on volition and action, and .Pr. Zoltan Dienes (INT2, University of Sussex) for his expertise on implicit learning and unconscious processes. The partners know each other very well, having often already collaborated with each other. They not only share a deep interest in the importance of learning and plasticity in their respective domains but also have complementary skills and areas of expertise that will be leveraged to their full effect in this project. All have already received the full support of their respective institutions.
COOL is structured in eight workpackages (WP), each placed under the responsibility of one of the partners. The proposed research is strongly driven by a coherent novel perspective on how one should conceive of the traditional dichotomies described above, and addresses the fundamental role that conscious and unconscious learning play in different domains (e.g., memory, face perception, perceptual learning, literacy, animal learning, conditioning, decision-making, habituation, implicit learning, subliminal perception, volition). This innovative vision will result in an important step forward in understanding the fundamental ability of humans and other organisms to adapt to an ever-changing environment.
