Adam Safron’s scientific contributions

In this first set of explorations/meditations (of three), we examine major transitions in cognitive evolution and provides a framework for understanding different types of intelligence in naturally evolved systems. Drawing on synthetic perspectives from multiple theorists, we review several key taxonomies of cognitive sophistication, including Dennett's hierarchy of creatures (Darwinian, Skinnerian, Popperian, and Gregorian), Pearl's ladder of causation (association, intervention, and counterfactuals), and Tomasello's evolution of agency (goal-directed, intentional, rational, and socially normative). We place particular emphasis on Tomasello's account, which traces the development of increasingly sophisticated forms of agency from early vertebrates to modern humans. The paper also examines how "unlimited associative learning" may mark critical transitions in cognitive evolution. This review provides groundwork for two subsequent papers exploring social intelligence, human uniqueness, and artificial intelligence. We suggest that understanding these natural histories of intelligence may be crucial for developing artificial systems that can robustly engage in causal reasoning and social interaction. This theoretical synthesis offers insights into both the evolution of natural intelligence and potential pathways for developing artificial agents with sophisticated cognitive capabilities.

In this final set of explorations/meditations (of three), we examine the requirements for developing artificial general intelligence (AGI) through the lens of human cognitive architecture, with particular emphasis on the role of narrative selfhood and social cognition. Drawing on perspectives from cognitive science, philosophy of mind, and artificial intelligence research, we critically evaluate current claims about the capabilities of large language models, particularly regarding their purported achievements of theory of mind and self-awareness. We argue that genuinely human-like artificial intelligence may require more than sophisticated pattern recognition and language modeling, potentially necessitating the development of coherent narrative self-models and rich causal understanding. Special attention is given to the relationship between consciousness, conscience, and trustworthy AI systems, suggesting that meaningful artificial intelligence may require forms of richly-embodied and socially-embedded development to achieve robust and reliable functionality. We conclude by proposing that the path to artificial general intelligence may require recapitulating aspects of human cognitive development, particularly regarding the construction of narrative identity and social-moral reasoning capabilities. This analysis has implications for both the technical development of AI systems and the ethical frameworks through which we evaluate artificial minds.

In this second set of explorations/meditations (of three), we examine potential sources of human cognitive uniqueness through comparative analysis with non-human primates, focusing particularly on theory of mind, analogical reasoning, and social learning capabilities. We review evidence suggesting that while humans share many cognitive foundations with other primates, we may be distinct in our capacity for abstract relational reasoning and understanding others' false beliefs. Drawing on research in developmental and comparative psychology, we explore how these capabilities may emerge through the interplay of innate predispositions and social learning. Special attention is given to the role of imitation and over-imitation in cultural learning, as well as the development of mirror self-recognition and its relationship to broader social-cognitive abilities. We also consider neurobiological factors that may contribute to human cognitive uniqueness, including patterns of brain integration and the role of social neuropeptides in shaping affiliative behaviors. We conclude by suggesting that human cognitive capabilities may be best understood as emerging from the dynamic interaction between domain-general processes like analogical reasoning and species-typical social motivations, laying groundwork for subsequent discussions of artificial intelligence and consciousness. This synthesis provides insight into both the continuities and discontinuities between human and non-human minds, with implications for understanding the requirements for developing human-like artificial intelligence.

In this second set of explorations/meditations (of three), we examine potential sources of human cognitive uniqueness through comparative analysis with non-human primates, focusing particularly on theory of mind, analogical reasoning, and social learning capabilities. We review evidence suggesting that while humans share many cognitive foundations with other primates, we may be distinct in our capacity for abstract relational reasoning and understanding others' false beliefs. Drawing on research in developmental and comparative psychology, we explore how these capabilities may emerge through the interplay of innate predispositions and social learning. Special attention is given to the role of imitation and over-imitation in cultural learning, as well as the development of mirror self-recognition and its relationship to broader social-cognitive abilities. We also consider neurobiological factors that may contribute to human cognitive uniqueness, including patterns of brain integration and the role of social neuropeptides in shaping affiliative behaviors. We conclude by suggesting that human cognitive capabilities may be best understood as emerging from the dynamic interaction between domain-general processes like analogical reasoning and species-typical social motivations, laying groundwork for subsequent discussions of artificial intelligence and consciousness. This synthesis provides insight into both the continuities and discontinuities between human and non-human minds, with implications for understanding the requirements for developing human-like artificial intelligence.

In this final set of explorations/meditations (of three), we examine the requirements for developing artificial general intelligence (AGI) through the lens of human cognitive architecture, with particular emphasis on the role of narrative selfhood and social cognition. Drawing on perspectives from cognitive science, philosophy of mind, and artificial intelligence research, we critically evaluate current claims about the capabilities of large language models, particularly regarding their purported achievements of theory of mind and self-awareness. We argue that genuinely human-like artificial intelligence may require more than sophisticated pattern recognition and language modeling, potentially necessitating the development of coherent narrative self-models and rich causal understanding. Special attention is given to the relationship between consciousness, conscience, and trustworthy AI systems, suggesting that meaningful artificial intelligence may require forms of richly-embodied and socially-embedded development to achieve robust and reliable functionality. We conclude by proposing that the path to artificial general intelligence may require recapitulating aspects of human cognitive development, particularly regarding the construction of narrative identity and social-moral reasoning capabilities. This analysis has implications for both the technical development of AI systems and the ethical frameworks through which we evaluate artificial minds.

In this first set of explorations/meditations (of three), we examine major transitions in cognitive evolution and provides a framework for understanding different types of intelligence in naturally evolved systems. Drawing on synthetic perspectives from multiple theorists, we review several key taxonomies of cognitive sophistication, including Dennett's hierarchy of creatures (Darwinian, Skinnerian, Popperian, and Gregorian), Pearl's ladder of causation (association, intervention, and counterfactuals), and Tomasello's evolution of agency (goal-directed, intentional, rational, and socially normative). We place particular emphasis on Tomasello's account, which traces the development of increasingly sophisticated forms of agency from early vertebrates to modern humans. The paper also examines how "unlimited associative learning" may mark critical transitions in cognitive evolution. This review provides groundwork for two subsequent papers exploring social intelligence, human uniqueness, and artificial intelligence. We suggest that understanding these natural histories of intelligence may be crucial for developing artificial systems that can robustly engage in causal reasoning and social interaction. This theoretical synthesis offers insights into both the evolution of natural intelligence and potential pathways for developing artificial agents with sophisticated cognitive capabilities.

Last year, a controversial letter criticized a highly cited theory of consciousness, integratedinformation theory or IIT, as “pseudoscience”. The letter was penned by 10 prominentpsychologists, cognitive scientists, and neuroscientists, and signed by over 100 additional researchers,creating a group called “IIT-Concerned.” The pseudoscience accusation, while controversial, hassince been repeated by one of IIT-Concerned most vocal members. Herein, we will challenge IIT-Concerned’s criticism by providing a historical parallel. And yet, our goal is not to promote IIT over other theories of consciousness. As discussed below, we believe that IIT has both strengths and weaknesses as a theory of consciousness; we are not arguing against reasonable criticisms of the theory. Rather, our goal is to critique the reasoning of IIT-Concerned. In what follows, we provide a brief background on IIT for context before diving into the controversy surrounding the theory and, finally, drawing a historical analogy that we believe to be illuminating.

We are introducing a novel thermodynamic emotion model. In this model, emotions are regarded as deviations from equilibrium, akin to fluctuations in body temperature. This bipolar regulation maintains bodily and psychological homeostasis while spurring mental development. Emotional regulation typically occurs through expanding one's perception of time. Positive, low-information content emotions can reduce action drive, but stressful, information-rich conditions can heighten it. However, action accelerates time perception to facilitate fluid action performance, where a unique state of contentment and challenge represents flow. Therefore, time perception can control emotions’ capacity to control motivation. By anchoring psychological processes to the principles of energy and entropy, our model offers a comprehensive bipolar foundation for understanding motivation and behavior. Beyond its theoretical implications, this model also lays the groundwork for addressing mental health conditions resulting from the dysregulation of emotions. It can inspire potential interventions to harness the mind-body connections elucidated in our thermodynamic perspective.