David C. Noelle

Letting structure emerge: connectionist and dynamical systems approaches to cognition

Connectionist and dynamical systems approaches explain human thought, language and behavior in terms of the emergent consequences of a large number of simple noncognitive processes. We view the entities that serve as the basis for structured probabilistic approaches as abstractions that are occasionally useful but often misleading: they have no real basis in the actual processes that give rise to linguistic and cognitive abilities or to the development of these abilities. Although structured probabilistic approaches can be useful in determining what would be optimal under certain assumptions, we propose that connectionist, dynamical systems, and related approaches, which focus on explaining the mechanisms that give rise to cognition, will be essential in achieving a full understanding of cognition and development.

Indirection and symbol-like processing in the prefrontal cortex and basal ganglia

The ability to flexibly, rapidly, and accurately perform novel tasks is a hallmark of human behavior. In our everyday lives we are often faced with arbitrary instructions that we must understand and follow, and we are able to do so with remarkable ease. It has frequently been argued that this ability relies on symbol processing, which depends critically on the ability to represent variables and bind them to arbitrary values. Whereas symbol processing is a fundamental feature of all computer systems, it remains a mystery whether and how this ability is carried out by the brain. Here, we provide an example of how the structure and functioning of the prefrontal cortex/basal ganglia working memory system can support variable binding, through a form of indirection (akin to a pointer in computer science). We show how indirection enables the system to flexibly generalize its behavior substantially beyond its direct experience (i.e., systematicity). We argue that this provides a biologically plausible mechanism that approximates a key component of symbol processing, exhibiting both the flexibility, but also some of the limitations, that are associated with this ability in humans.

A biologically inspired working memory framework for robots

The human brain includes a capacity-limited memory system devoted to the short-term retention of task-relevant information. This system is called working memory. Some computational neuroscience accounts of working memory have explained it in terms of interactions between the prefrontal cortex and the mesolimbic dopamine system. Inspired by these models, we have constructed a software toolkit for creating working memory components for robot control systems, based on the proposed mechanisms used by the brain. We report our design for this toolkit, as well as the results of a feasibility study, involving a robotic version of the delayed saccade task, and we discuss future plans to test this framework in the context of more complex tasks.

Relation between confidence in yes-no and forced-choice tasks

Yes-no and forced-choice tasks are common in psychology, but the empirical relation between reported confidence in the 2 tasks has been unclear. The authors examined this relation with 2 experiments. The general experimental method had participants first report confidence in the truth of each of many general knowledge statements (a yes-no task) then report confidence in them again when the statements were put into pairs where it was known that one statement was true and one was false (a forced-choice task). At issue was how confidence in the statements changed between the yes-no task and the forced-choice task. Two models, including the normative one, were ruled out as descriptive models. A linear model and a multiplicative model remain viable contenders.

Dopamine and the development of executive dysfunction in autism spectrum disorders.

Persons with autism regularly exhibit executive dysfunction (ED), including problems with deliberate goal-directed behavior, planning, and flexible responding in changing environments. Indeed, this array of deficits is sufficiently prominent to have prompted a theory that executive dysfunction is at the heart of these disorders. A more detailed examination of these behaviors reveals, however, that some aspects of executive function remain developmentaly appropriate. In particular, while people with autism often have difficulty with tasks requiring cognitive flexibility, their fundamental cognitive control capabilities, such as those involved in inhibiting an inappropriate but relatively automatic response, show no significant impairment on many tasks. In this article, an existing computational model of the prefrontal cortex and its role in executive control is shown to explain this dichotomous pattern of behavior by positing abnormalities in the dopamine-based modulation of frontal systems in individuals with autism. This model offers excellent qualitative and quantitative fits to performance on standard tests of cognitive control and cognitive flexibility in this clinical population. By simulating the development of the prefrontal cortex, the computational model also offers a potential explanation for an observed lack of executive dysfunction early in life.

Generalisation benefits of output gating in a model of prefrontal cortex

The prefrontal cortex (PFC) plays a central role in flexible cognitive control, including the suppression of habitual responding in favour of situation-appropriate behaviours that can be quite novel. PFC provides a kind of working memory, maintaining the rules, goals, and/or actions that are to control behaviour in the current context. For flexible control, these PFC representations must be sufficiently componential to support systematic generalisation to novel situations. The anatomical structure of PFC can be seen as implementing a componential ‘slot-filler’ structure, with different components encoded over isolated pools of neurons. Previous PFC models have highlighted the importance of a dynamic gating mechanism to selectively update individual ‘slot’ contents. In this article, we present simulation results that suggest that systematic generalisation also requires an ‘output gating’ mechanism that limits the influence of PFC on more posterior brain areas to reflect a small number of representational components at any one time.

Modular behavior control for a cognitive robot

We propose a method for a cognitive robot behavior control in which a small number of behaviors are loaded into a workspace, called working memory, where they are combined to generate actions during a task execution. We use the existing components in our cognitive robot architecture, such as the long-term memory, the short-term memory, with the addition of a working memory system and a control mechanism called the central executive agent to create a modular control system. This control method is used to drive the behaviors of our humanoid robot ISAC

Towards Instructable Connectionist Systems

At least three disparate channels have been used to install new knowledge into artificial intelligence systems. The first of these is the programmer channel, through which the knowledge in the system is simply edited to include the desired new knowledge. While this method is often effective, it may not be as efficient as learning directly from environmental interaction. The second channel may be called the linguistic channel, through which knowledge is added by explicitly telling the system facts or commands encoded as strings of quasi-linguistic instructions in some appropriate form. Finally, there is, for want of a better phrase, the learning channel, through which the system learns new knowledge in an inductive way via environmental observations and simple feedback information. These latter two channels are the ones upon which we wish to focus, as they are the hallmarks of instructable systems. Most instructable systems depend upon, or at least heavily favor, one of these two channels for the bulk of their knowledge acquisition. Specifically, symbolic artificial intelligence systems have generally depended upon the explicit use of sentential logical expressions, rules, or productions for the transmission of new knowledge to the system. In contrast, many connectionist network models have relied solely on inductive generalization mechanisms for knowledge creation. There is no apparent reason to believe that this rough dichotomy of technique is necessary, however. Systems which are capable of receiving detailed instruction and also generalizing from experience the both possible and potentially very useful.

Working memory and perception

The ability to teach a robot new skills and tasks without explicit programming is an important goal in robotics. Such capability tends to imply the ability to learn from experience, much like many biological creatures. Evidence suggests that working memory plays a pivotal role in this learning process, in part by focusing attention on the most relevant data. We describe ongoing research to study the utility of computational neuroscience models of working memory within robotic systems. A system comprised of working memory, short term memory, long term memory, spatial reasoning and perception modules is proposed. The paper focuses on the perceptual module and its interaction with the working memory. Results are given to show the current progress.

Modeling the development of overselectivity in autism

People with autism consistently demonstrate a lack sensitivity to the full range of important aspects of everyday situations. Often, an overly restricted subset of the information available in a given situation gains control over their behavior. This can result in problems generalizing learned behaviors to novel situations. This phenomenon has been called overselectivity. Indeed, many behavioral intervention techniques seek to mitigate overselectivity effects in this population. In this paper, we offer an account of overselectivity as arising from an inability to flexibly adjust the attentional influences of the prefrontal cortex on behavior. We posit that dysfunctional dopamine interactions with the prefrontal cortex result in overly perseverative attention in people with autism. Limiting attention to only a few of the features of a situation hinders the learning of associations between the full range of relevant environmental properties and appropriate behavior. Thus, a restricted subset of features gain control over responding. A simple neurocomputational model of the attentional effects of prefrontal cortex on learning is presented, demonstrating how weak dopamine modulation of frontal areas can lead to overselectivity.