Search arXiv⌕ Search

arXiv subjects

Adam Safron

Publications and source records attributed to Adam Safron.

10 recordsLinked to original sources
arXiv2025

The Conditions of Physical Embodiment Enable Generalization and Care

As artificial agents enter open-ended physical environments -- eldercare, disaster response, and space missions -- they must persist under uncertainty while providing reliable care. Yet current systems struggle to generalize across distribution shifts and lack intrinsic motivation to preserve the well-being of others. Vulnerability and mortality are often seen as constraints to be avoided, yet organisms survive and provide care in an open-ended world with relative ease and efficiency. We argue that generalization and care arise from conditions of physical embodiment: being-in-the-world (the agent is a part of the environment) and being-towards-death (unless counteracted, the agent drifts toward terminal states). These conditions necessitate a homeostatic drive to maintain oneself and maximize the future capacity to continue doing so. Fulfilling this drive over long time horizons in multi-agent environments necessitates robust causal modeling of self and others' embodiment and jointly achievable future states. Because embodied agents are part of the environment, with the self delimited by reliable control, empowering others can expand self-boundaries, enabling other-regard. This provides a path from embodiment toward generalization and care based in shared constraints. We outline a reinforcement-learning framework for examining these questions. Homeostatic mortal agents continually learning in open-ended environments may offer efficient robustness and trustworthy alignment.

cs.AI↗
arXiv2025

Relational Norms for Human-AI Cooperation

How we should design and interact with social artificial intelligence depends on the socio-relational role the AI is meant to emulate or occupy. In human society, relationships such as teacher-student, parent-child, neighbors, siblings, or employer-employee are governed by specific norms that prescribe or proscribe cooperative functions including hierarchy, care, transaction, and mating. These norms shape our judgments of what is appropriate for each partner. For example, workplace norms may allow a boss to give orders to an employee, but not vice versa, reflecting hierarchical and transactional expectations. As AI agents and chatbots powered by large language models are increasingly designed to serve roles analogous to human positions - such as assistant, mental health provider, tutor, or romantic partner - it is imperative to examine whether and how human relational norms should extend to human-AI interactions. Our analysis explores how differences between AI systems and humans, such as the absence of conscious experience and immunity to fatigue, may affect an AI's capacity to fulfill relationship-specific functions and adhere to corresponding norms. This analysis, which is a collaborative effort by philosophers, psychologists, relationship scientists, ethicists, legal experts, and AI researchers, carries important implications for AI systems design, user behavior, and regulation. While we accept that AI systems can offer significant benefits such as increased availability and consistency in certain socio-relational roles, they also risk fostering unhealthy dependencies or unrealistic expectations that could spill over into human-human relationships. We propose that understanding and thoughtfully shaping (or implementing) suitable human-AI relational norms will be crucial for ensuring that human-AI interactions are ethical, trustworthy, and favorable to human well-being.

cs.AI↗
arXiv2024

From Physics to Sentience: Deciphering the Semantics of the Free-Energy Principle and Evaluating its Claims

The Free-Energy Principle (FEP) [1-3] has been adopted in a variety of ambitious proposals that aim to characterize all adaptive, sentient, and cognitive systems within a unifying framework. Judging by the amount of attention it has received from the scientific community, the FEP has gained significant traction in these pursuits. The current target article represents an important iteration of this research paradigm in formally describing emergent dynamics rather than merely (quasi-)steady states. This affords more in-depth considerations of the spatio-temporal complexities of cross-scale causality - as we have encouraged and built towards in previous publications (e.g., [4-9]). In this spirit of constructive feedback, we submit a few technical comments on some of the matters that appear to require further attention, in order to improve the clarity, rigour, and applicability of this framework.

q-bio.NC↗
arXiv2023

The inner screen model of consciousness: applying the free energy principle directly to the study of conscious experience

This paper presents a model of consciousness that follows directly from the free-energy principle (FEP). We first rehearse the classical and quantum formulations of the FEP. In particular, we consider the inner screen hypothesis that follows from the quantum information theoretic version of the FEP. We then review applications of the FEP to the known sparse (nested and hierarchical) neuro-anatomy of the brain. We focus on the holographic structure of the brain, and how this structure supports (overt and covert) action.

q-bio.NC↗
arXiv2021

Dream to Explore: Adaptive Simulations for Autonomous Systems

One's ability to learn a generative model of the world without supervision depends on the extent to which one can construct abstract knowledge representations that generalize across experiences. To this end, capturing an accurate statistical structure from observational data provides useful inductive biases that can be transferred to novel environments. Here, we tackle the problem of learning to control dynamical systems by applying Bayesian nonparametric methods, which is applied to solve visual servoing tasks. This is accomplished by first learning a state space representation, then inferring environmental dynamics and improving the policies through imagined future trajectories. Bayesian nonparametric models provide automatic model adaptation, which not only combats underfitting and overfitting, but also allows the model's unbounded dimension to be both flexible and computationally tractable. By employing Gaussian processes to discover latent world dynamics, we mitigate common data efficiency issues observed in reinforcement learning and avoid introducing explicit model bias by describing the system's dynamics. Our algorithm jointly learns a world model and policy by optimizing a variational lower bound of a log-likelihood with respect to the expected free energy minimization objective function. Finally, we compare the performance of our model with the state-of-the-art alternatives for continuous control tasks in simulated environments.

cs.LG↗
arXiv2019

Bayesian Analogical Cybernetics

It has been argued that all of cognition can be understood in terms of Bayesian inference. It has also been argued that analogy is the core of cognition. Here I will propose that these perspectives are fully compatible, in that analogical reasoning can be described in terms of Bayesian inference and vice versa, and that both of these positions require a thorough cybernetic grounding in order to fulfill their promise as unifying frameworks for understanding minds. From the Bayesian perspective of the Free Energy Principle and Active Inference framework, thought is constituted by dynamics of cascading belief propagation through the nodes of probabilistic generative models specified by a cortical heterarchy "rooted" in action-perception cycles that ground the mind as an embodied control system for an autonomous agent. From the analogical structure mapping perspective, thought is constituted by the alignment and comparison of heterogeneous structural representations. Here I will propose that this core cognitive process for analogical reasoning is naturally implemented by predictive coding mechanisms. However, both Bayesian cognitive science and models of cognitive development via analogical reasoning require rich base domains and priors (or reliably learnable posteriors) from which they can commence the process of bootstrapping minds. Here in the spirit of the work of George Lakoff and Mark Johnson, I propose that embodiment provides many of the inductive biases that are usually described in terms of innate core knowledge. (Please note: this manuscript was written and finalized in 2012.)

q-bio.NC↗
arXiv2019

Rapid Anxiety Reduction (RAR): A unified theory of humor

Here I propose a novel theory in which humor is the feeling of Rapid Anxiety Reduction (RAR). According to RAR, humor can be expressed in a simple formula: -d(A)/dt. RAR has strong correspondences with False Alarm Theory, Benign Violation Theory, and Cognitive Debugging Theory, all of which represent either special cases or partial descriptions at alternative levels of analysis. Some evidence for RAR includes physiological similarities between hyperventilation and laughter and the fact that smiles often indicate negative affect in non-human primates (e.g. fear grimaces where teeth are exposed as a kind of inhibited threat display). In accordance with Benign Violation Theory, if humor reliably indicates both a) anxiety induction, b) anxiety reduction, and c) the time-course over which anxiety is reduced, then the intersection of these conditions productively constrains inference spaces over latent mental states with respect to the values and capacities of the persons experiencing humor. In this way, humor is a powerful cypher for understanding persons in both individual and social contexts, with far-reaching implications. Finally, if humor can be expressed in such a simple formula with clear ties to phenomenology, and yet this discovery regarding such an essential part of the human experience has remained undiscovered for this long, then this is an extremely surprising state of affairs worthy of further investigation. Towards this end, I propose an analogy can be found with consciousness studies, where in addition to the "Hard problem" of trying to explain humor, we would do well to consider a "Meta-Problem" of why humor seems so difficult to explain, and why relatively simple explanations may have eluded us for this long. (Please note: RAR was conceived in 2008, and last majorly updated in 2012.)

q-bio.NC↗
arXiv2019

Multilevel evolutionary developmental optimization (MEDO): A theoretical framework for understanding preferences and selection dynamics

What is motivation and how does it work? Where do goals come from and how do they vary within and between species and individuals? Why do we prefer some things over others? MEDO is a theoretical framework for understanding these questions in abstract terms, as well as for generating and evaluating specific hypotheses that seek to explain goal-oriented behavior. MEDO views preferences as selective pressures influencing the likelihood of particular outcomes. With respect to biological organisms, these patterns must compete and cooperate in shaping system evolution. To the extent that shaping processes are themselves altered by experience, this enables feedback relationships where histories of reward and punishment can impact future motivation. In this way, various biases can undergo either amplification or attenuation, resulting in preferences and behavioral orientations of varying degrees of inter-temporal and inter-situational stability. MEDO specifically models all shaping dynamics in terms of natural selection operating on multiple levels--genetic, neural, and cultural--and even considers aspects of development to themselves be evolutionary processes. Thus, MEDO reflects a kind of generalized Darwinism, in that it assumes that natural selection provides a common principle for understanding the emergence of complexity within all dynamical systems in which replication, variation, and selection occur. However, MEDO combines this evolutionary perspective with economic decision theory, which describes both the preferences underlying individual choices, as well as the preferences underlying choices made by engineers in designing optimized systems. In this way, MEDO uses economic decision theory to describe goal-oriented behaviors as well as the interacting evolutionary optimization processes from which they emerge. (Please note: this manuscript was written and finalized in 2012.)

q-bio.NC↗
arXiv2018

Integrative Biological Simulation, Neuropsychology, and AI Safety

We describe a biologically-inspired research agenda with parallel tracks aimed at AI and AI safety. The bottom-up component consists of building a sequence of biophysically realistic simulations of simple organisms such as the nematode $Caenorhabditis$ $elegans$, the fruit fly $Drosophila$ $melanogaster$, and the zebrafish $Danio$ $rerio$ to serve as platforms for research into AI algorithms and system architectures. The top-down component consists of an approach to value alignment that grounds AI goal structures in neuropsychology, broadly considered. Our belief is that parallel pursuit of these tracks will inform the development of value-aligned AI systems that have been inspired by embodied organisms with sensorimotor integration. An important set of side benefits is that the research trajectories we describe here are grounded in long-standing intellectual traditions within existing research communities and funding structures. In addition, these research programs overlap with significant contemporary themes in the biological and psychological sciences such as data/model integration and reproducibility.

cs.AI↗