Spring 2023
Instructor: Bonny Banerjee, Ph.D.
Contact Information:
Office: 208B Engineering Science Bldg
Phone: 901-678-4498
E-mail: BBnerjee@memphis.edu
Office Hours: By appointment
When: Wednesdays 2:20-5:20pm
There will be a regular lecture during 2:20-4:00 pm and a talk by an invited speaker on a topic relevant to the course during 4:00-5:20 pm. Everyone, including faculty, is welcome to attend the talks.
Where: FIT Bldg. Room 405
Course Description:
In the field of Artificial Intelligence (AI), anything that can perceive its environment and act on that environment is called an agent (Russell & Norvig, 2020). In this view, humans, most living organisms, autonomous vehicles, and thermostats are agents. A key question to understand for the design of agent models is -- why do living organisms act? In AI, there are two leading theories that strive to answer this question: (1) to maximize expected reward (a.k.a. reinforcement learning), which is the dominant view in AI, and (2) to minimize the brain's prediction error (a.k.a. predictive coding or free energy principle). A number of other theories have been extensively researched in the natural and applied sciences. In this Cognitive Science Seminar course, we will discuss papers and invite speakers from AI, robotics, psychology, neuroscience, biology, and philosophy to understand the influential theories spanning over a century of research regarding why living organisms act.
Required Text:
Readings from research papers and book chapters.
Evaluation and Final Grades:
This course requires high level of creative thinking from each student. Grading will include project and report (80%), and class participation (20%). Students will be given a project topic which will not involve computer programming and will be carried out throughout the semester.
External Speakers:
Jan 18: Dr. Bonny Banerjee, Associate Professor, Institute for Intelligent Systems, and Department of Electrical & Computer Engineering, University of Memphis, USA
Title: Reinforcement Learning and Predictive Coding Models of Agents in Artificial Intelligence
Abstract: I will present research from my lab on attention-based agent models applied to the task of handwritten numeral and alphabet recognition. I will present two models -- one reinforcement-based and the other based on predictive coding, and compare their performance on the task with respect to human performance.
References:
[1] M. Baruah, B. Banerjee, A. K. Nagar and R. Marois. (2023) "AttentionMNIST: A mouse-click attention tracking dataset for handwritten numeral and alphabet recognition", Scientific Reports, Nature, Vol. 13, Article 3305. Dataset: https://github.com/Murchana/AttentionMNIST .
[2] B. Banerjee and M. Baruah. (2023) "An attention-based predictive agent for handwritten numeral/alphabet recognition via generation", NeurIPS 2023 Workshop on Gaze Meets ML, December 16, New Orleans, LA. [Best Paper Award] Appears in Proceedings of Machine Learning Research, Vol. 226, pp. 4–20, 2024.
Jan 25: Dr. Susana A. M. Varela, Researcher at William James Center for Research (WJCR) of ISPA-Instituto Universitário (ISPA-IU) and at the Integrative Behavioural Biology Group (IBBG) of Instituto Gulbenkian de Ciência (IGC), Portugal
Title: The importance of social information and the modularity of social cognition
Abstract: Social information refers to the behaviour of others and allows learning about the environment without interacting directly with its agents. This process saves time and energy for decision-making and is generally accurate, because when observer and demonstrator individuals are conspecifics, they share the same ecological needs. Many other types of information also exist, such as genetic and epigenetic information, parental effects, habitat inheritance, and asocial information. The latter is the information that individuals acquire from their own experiences with the environment, informing animals about their individual abilities, preferences, and limitations. All types of information play important roles in the decision-making process, but the use of social and asocial information in particular requires a cognitive process of associative learning. Learning is considered adaptive when environmental conditions are predictive enough within a generation but not across generations, making the genetic inheritance of cognitive learning mechanisms more advantageous than the genetic inheritance of an extensive repertoire of specialized behaviours. However, we ignore how specialized learning mechanisms must be (special-purpose versus general-purpose cognition) to produce behaviours suitable for social and asocial environments. Knowing how social cognition can be distinguished from other cognitive modules in the brain is crucial for us to know what truly social behaviours are and understand the evolution of sociability. To answer that question, I have been studying the genetic architecture of social and asocial learning in Drosophila melanogaster, the model organism with the most powerful genetic tools for studying the nervous system. The results we obtained so far suggest a mixed mechanism of cognition, which includes a domain-specific genetic architecture for social learning.
Feb 1: Dr. Vittorio Gallese, Professor of Psychobiology and Cognitive Neuroscience, Università di Parma, Italy
Title: The shared manifold of emotions and sensations. An embodied perspective
Abstract: Cognitive neuroscience can provide new insights on topics like intersubjectivity and the self by emphasizing the crucial role of the body, conceived as the constitutive source of pre-reflective consciousness. The body, as being addressed in this talk manifests itself in two different, complementary and closely intertwined ways: 1) it is a Leib, a lived body entertaining experiences of self and others, and 2) it is a Körper, the somatic object of our sensations and perceptions. This dual nature of the experienced body can be fully understood and its genesis revealed by investigating at the sub-personal level its sensorimotor and interoceptive neurophysiological underpinnings within a comparative and evolutionary perspective. In my talk I’ll review and discuss empirical results on how we relate to and recognize others’ emotions and sensations. On emotions I’ll posit that we should abandon the idea that experiencing emotions is a purely sensory activity, totally independent from their motor expression. By endorsing Dewey’ s and Mead’s account of emotions and capitalizing upon recent empirical findings I’ll propose an alternative perspective: the behaviour connected to a specific emotion is part of the emotion itself. In my talk I will present and discuss recent neuroscientific studies showing the link between emotion experience and expression in social cognition. On sensations I’ll show that we do not just “see” a sensation experienced by someone else and then understand it through an inference by analogy. By means of embodied simulation we can map others’ sensations by re-using our own premotor and somatosensory representations to recognize others’ sensations from within.
Feb 8: Dr. Heather Lench, Professor of Psychological and Brain Sciences, Texas A&M University, USA
Title: Testing functional accounts of emotion through anger and boredom
Abstract: Functional theories posit that emotions are elicited by particular goal-related situations that represented adaptive problems and that emotions are evolved features of coordinated responses to those situations. We apply five criteria that can be used to judge whether features are adaptations. There is evidence that sadness, anger, and anxiety relate to unique changes in physiology, cognition, and behavior, those changes are correlated, situations that give rise to emotions are consistent, and emotions are complex. A key assumption of functional accounts of emotion is that emotions are functional and improve outcomes for people. Across multiple studies, we test this key assumption with the emotion of anger and demonstrate that anger improves goal attainment in the presence of obstacles in situations that involve different types of goals and obstacles to goals. We have begun to develop and test a similar model for the emotion of boredom, with evidence that boredom prompts the pursuit of new experiences that elicit emotional responses. The study of these two emotions provides a frame to understand when and how emotions can be considered functional, as well as the type of evidence required to draw this conclusion.
Feb 15: Dr. D. Kimbrough Oller, Professor and Plough Chair of Excellence, School of Communication Sciences and Disorders, University of Memphis, USA
Title: The human infant as an agent in vocal and language development
Abstract: The notion of “agent” is important as we move increasingly away, but still gradually in the 21st century, from the radical behaviorism of decades past. Babies and all the rest of us were thought in those decades to be reactors rather than actors, and every action was described as a response. Perhaps surprisingly, the narrowness of behaviorism is far from over. It remains true that the predominant expectation of the field of child development is that infants acquire language and even acquire a first-year-of-life vocal repertoire by listening to their caregivers and imitating them. In fact, careful longitudinal observations in naturalistic settings suggest that vocal turn-taking with caregivers in the first year is extremely rare, constituting less (and perhaps far less) than 10% of the speech-like vocalizations (protophones) that infants produce throughout the first year. And vocal imitation is so rare that one must search recordings diligently to find even a few convincing examples, and those are overwhelmingly cases where the caregiver elicits infant protophones they know are already in the infant repertoires. Such cases provide no evidence that infant vocal repertoires are affected at all by caregiver utterances or imitation. No, the story is wildly different from the behaviorist expectation that continues into current time. The real story is that the human infant is an explorer of sound from the first day of life, producing ~3500 protophones a day, 4-5 per min every waking hour, and at least 90% of these are directed to no one. Why are infants so interested in overwhelmingly endogenous vocal activity? We have an explanation, to be presented in the lecture, and it requires a true shift of viewpoint, a shift the requires accepting the evidence that human infants are vocal actors, true agents of their own vocal development.
Feb 22: Dr. Vacslav Glukhova, PhD in Electrical Engineering from Stanford University, ex-Director of AI Research, JPMorgan Chase & Co., UK
Title: Reward is not enough: Can we liberate AI from the reinforcement learning paradigm?
Abstract: I present arguments against the hypothesis put forward by Silver, Singh, Precup, and Sutton of DeepMind in their paper Reward is enough: reward maximization is not enough to explain many activities associated with natural and artificial intelligence including knowledge, learning, perception, social intelligence, evolution, language, generalisation and imitation.
I show such reductio ad lucrum has its intellectual origins in the political economy of Homo economicus and substantially overlaps with the radical version of behaviourism.
I show why the reinforcement learning paradigm, despite its demonstrable usefulness in some practical application, is an incomplete framework for intelligence – natural and artificial. Complexities of intelligent behaviour are not simply second-order complications on top of re- ward maximisation. This fact has profound implications for the development of practically usable, smart, safe and robust artificially intelligent agents.
I also briefly discuss the sub-optimality of mammalian visual system and present unpublished simulations of a model of visual object recognition to support my arguments
March 1: Dr. Reuven Dukas, Professor, Department of Psychology, Neuroscience & Behaviour, McMaster University, Canada
Title: Evolution of sociability by artificial selection
Abstract: Sociability, defined as the tendency to engage in friendly activities with others, is ubiquitous among animals including humans. To examine the evolutionary biology of, and mechanisms underlying sociability, we artificially selected for low and high sociability in fruit flies. After 25 generations of selection, the high sociability lineages showed sociability scores 50% higher than did the low sociability lineages. The evolved lineages maintained their sociability scores after ten generations of relaxed selection. Experiments on the evolved lineages indicated that both males and females from the low lineages were more aggressive than males and females from the high lineages. There were, however, no differences in other fitness-related traits between flies from the low and high lineages. We collected RNA from the heads of flies and have identified about 100 genes that had significantly different expression between the low and high sociability lineages. We are now in the process of critically testing whether indeed these genes affect sociability.
March 8: No class due to spring break
March 15: Dr. Philip Kohlmeier, Assistant Professor, Department of Biological Sciences, University of Memphis, USA
Title: The neurogenetic mechanisms and evolution of division of labor in social hymenopterans
Abstract: Division of labor evolved multiple times on different levels of biological organization and is a major driver of biodiversity: Molecules cooperate and form cells; specialized cells built multicellular organisms; and organisms specialize in tasks and live in social groups. Thus, dissecting how division of labor between formerly independent units is regulated and how these regulatory mechanisms evolved from an ancestor that performed all tasks by itself is key to understand the emergence of biodiversity and biological complexity.
An excellent system to answer these questions are social hymenopterans, i.e. ants, some bees and wasps. In their colonies, queens monopolize reproduction while the sterile workers take over all other tasks. Whether a worker specializes in brood care or foraging is regulated by numerous factors including age, physiology, neuronal networks, and gene expression. I will present previous and ongoing research of our lab that focusses on how genes, hormones and sensory neurons that already existed in solitary insects gained additional functions and became regulators of task specialization in ant workers.
March 22: Dr. James Moore, Reader, Department of Psychology, Goldsmiths, University of London, UK
Title: What is sense of agency and why does it matter?
Abstract: In this talk I will be discussing the sense of agency; the feeling of being in the driving seat when it comes to our interactions with the world. I will first give a broad introduction to this research area, explaining in more detail what sense of agency is, how it is measured, and what neurocognitive processes are thought to underpin it. I will then consider real-world applications of this research, focussing on a) aberrant experiences of agency in certain psychiatric disorders (schizophrenia and autism), and b) the psychology of human-computer interaction (in virtual reality and other input modalities).
March 29: Dr. Michael D. Greenfield, MD, Research Professor, Department of Ecology and Evolutionary Biology, University of Kansas, USA; Equipe Neuro-Ethologie Sensorielle, Univ. Lyon/Saint Etienne, France
Title: Coordinated rhythms in animal species, including man: Entrainment from bushcricket choruses to the phlharmonic orchestra
Abstract: Repetitive cycles of an activity or state, generally known as rhythm, are ubiquitous in living organisms, occurring in all branches of the tree of life. In animal species rhythms play prominent roles in physiology and behavior, and they are commonly expressed as long-term (period > 12 h) cycles, entrained to recurring astronomical phenomena, at one end of the temporal spectrum or as very short (< 10 s) ultradian cycles at the other (Aschoff, 1981). In behavior, ultradian rhythms often involve movement or the broadcast of communication signals, and in some species neighbouring individuals coordinate their rhythms in time. Such coordination fascinates us because of its clockwork precision and complexity, the horde of individuals that may participate, and the resulting crescendo of the communal display. But aesthetic sensibilities aside, coordinated rhythms pose major questions concerning mechanisms and evolution (Greenfield et al., 2021a): How is coordination achieved, which individuals in a local population participate, why do neighbours bother to coordinate their activity and what is the phylogenetic distribution of these displays among animal taxa?
This seminar will cover the diverse ways in which animals coordinate relatively fast behavioral rhythms. We concentrate on the coordination of rhythms expressed in sexual and social communication, as these tend to be conspicuous to observers and their parameters can normally be measured unambiguously. I begin by presenting the various timing schemes found in rhythm coordination, noting the relative timing of the signaling by individual participants, how those participants tend to be arranged in space, and the signaling modalities used. From the receiver’s perspective, I ask who perceives the signals, where these potential receivers, whether within or outside of the coordinated group, are located and how receivers may respond to the group display. Because of moderate to extreme precision found in a great many coordinated group displays, I pay special attention to the mechanisms with which signaling rhythms are aligned in different animal species. Following, I show how an understanding of mechanism is critical for evaluating adaptive explanations. Although a survey of rhythm coordination among animals is certainly incomplete and probably reflects sampling bias favouring signals and species that are readily observed and that have been studied traditionally, some striking patterns emerge. Rhythm coordination of both modest and high precision appears mostly among the arthropods and certain ‘lower’ vertebrates in the chordates, but also in humans where it is central in music, dance and other group activities in all cultures. We consider this curious distribution in an attempt to elucidate the origins of music.
April 5: Dr. James Meindl, Professor, Department of Instruction and Curriculum Leadership, University of Memphis, USA
Title: The goals, methods, and philosophy of behavior analysis science: Prediction and influence of meaningful behavior.
Abstract: Behavior analysis is a robust natural science that seeks to understand and elucidate the principles of behavior shared by humans and nonhumans animals alike. Anything an organism does is considered behavior (e.g., walking, talking, thinking, breathing), and behavior analysts approach the subject of behavior from a specific worldview and guided by a particular philosophy. This talk seeks to describe the behavior analytic discipline, the goals and philosophy of the science, and the behavior analyst’s approach to predicting and influencing behavior.
April 12: Dr. Daniela Corbetta, Professor, Department of Psychology, University of Tennessee, Knoxville, USA
Title: How infants visually select where to reach
Abstract: In most situations, adults only need a cursory glance to select where they will reach for an object. For example, to grab a cup, they typically make a prior fixation to the handle of the cup and then direct their hand to that visually selected location to grasp the object. Even though infant reaching has been studied for many decades, it is not known if infants are capable of selecting a spatial location on the objects when aiming at them and it is unclear whether they can form a precise perceptual-motor match when reaching for them. The research I will present is using eye-tracking in the context of live reaching for objects with infants to begin to assess where infants look and where they reach when presented with an object to grasp. Specifically, I will answer the following questions in infants aged 5- to 11-months-old: (1) Where do infants direct their visual attention prior to reaching for an object? (2) Does the goal of reaching change visual attention to the scene, compared to a non-reaching situation? (3) When infants focus on the target object, do they visually select an object area prior to reaching? (4) And if so, can they direct their arm in that visually-selected object area successfully? (5) Finally, can we determine some factors that may contribute to the selection and formation of visual-motor matching in infancy?
April 19: Dr. Jennifer Mandel, Associate Professor, Department of Biological Sciences, University of Memphis, USA
Title: Diversity and evolution in the sunflower family
Abstract: The Asteraceae family, or sunflower family, is the largest and most diverse family of plants on earth, with considerable economic and ecological importance. Asteraceae are distributed worldwide, from nearly polar latitudes all the way to the tropics and occur across a diverse range of habitats from extreme deserts to swamps and from lowland rainforests to alpine tundra. The family represents an excellent system for addressing a broad range of questions in ecology and evolution including systematics, comparative genomics, and plant conservation. This seminar will discuss the genetic and genomic tools The Mandel Lab has developed and address questions including: What is the role of gene duplication and genome evolution in driving evolutionary novelty? What are the underlying processes that lead to repeatability of the evolution across lineages? What role does genetic variation play in the conservation of species?
April 26: Dr. Martin A. Nowak, Professor of Mathematics and of Biology, Harvard University, USA
Title: Natural cooperation
Abstract: Cooperation means that one individual pays a cost for another to receive a benefit. Cooperation can be at variance with natural selection. Why should you help competitors? Yet cooperation is abundant in nature and is an important component of evolutionary innovation. Cooperation can be seen as the master architect of evolution and as the third fundamental principle of evolution beside mutation and selection. I will present five mechanisms for the evolution of cooperation: direct reciprocity, indirect reciprocity, spatial selection, group selection and kin selection. Global cooperation and cooperation with future generations is necessary to ensure the survival of our species.