Intelligence Offloading and Neurodevelopmental Embodiment
For this year’s Embodied Intelligence workshop, Bradly Alicea, Jesse Parent, and Angela R. Oidas are presenting on “Intelligence Offloading and Neurodevelopment Embodiment”. Below is a summary of our presentation.
Cognitive Offloading is defined by both Andy Clark [1] and Risko and Gilbert [2]. The former defines offloading as the use of objects in the environment to augment cognitive performance. The Bluefin Dolphin utilize vortices in their aquatic environment to augment their muscle power during swimming.
Risko and Gilbert take a different approach. Like Clark, they emphasize the role of embodiment and physical action. In their definition, tasks are both altered and capacity-dependent. In our agent-based approach, alteration of information processing requirements come from externalization during development. Reduction of cognitive demand result from tradeoffs due to developmental constraints.
But what are developmental constraints? These arise from various pathways defined by innate features and interactions with the environment. This approach is inspired by developmental systems theory [3] and so-called genomic encodings (something our lab is working on that represents innateness in a developing agent). It is also more indirectly inspired by Gibsonian Information [4], a mathematical model of environmental complexity relative to internal processing capacity.
As development unfolds, agent explore their worlds, during which available energy is either invested into neuronal tissue (internalization) or used to encode information in the environmental affordances (externalization). This tradeoff is due to the relative scarcity of energy across development. In cases where there are great fluctuations or scarcity, more energy is invested in externalization.
We can characterize the effects of environment through both rate and sequence heterochrony. Rate heterochrony characterizes the growth rate as a curvilinear function that starts either early or late and continues in either an accelerated or decelerated fashion over developmental time. Sequence heterochrony allows for development of the neural substrate to proceed as a temporal sequence of acquisition: the acquisition of skills and knowledge is contingent on certain substrate components being present at a certain period of time.
In cases where those components are not present, the ability to utilize incoming information is either offloaded or becomes a missed opportunity. Some of our lab’s work on critical periods fits into this model as well: agents with a short critical period are much more highly constrained, and either engage in a lot of offloading, or do not develop significant brains relative to their body size. In humans and robotic models, offloading has been shown to both become more prominent over developmental time [5] as well as enabling further cognitive development in evolution [6].
We can also test this growth and acquisition hypothesis in the context of a real-world cognitive activity: celestial navigation. In this task, our agent’s goal is to learn how to navigate their environment using only the spatial cues of a starfield. This starfield is much like the nighttime sky as viewed from specific locations on earth: some stars rise and move across the sky, while others are only visible in certain parts of the sky. This requires the agent to engage in a mix of internal spatial representations and external landmarks. The sky is mapped during development and has an effect on the size and behavior of mature agents.
References:
[1] Clark, A. (1997). Being There: Putting Brain, Body, and World Together Again. MIT Press.
[2] Risko, E.F. and Gilbert, S.J. (2016). Cognitive Offloading. Trends in Cognitive Sciences, 20, 676–88.
[3] Oyama, S. (2000). The Ontogeny of Information. Duke University Press.
[4] Alicea, B., Cialfi, D., Lim, A., and Parent, J. (2022). Gibsonian Information: a new approach to quantitative information. In “Biologically Inspired Cognitive Architectures”. Studies in Computational Intelligence, 1032.
For a more expansive treatment, please see: Gibsonian Information: an agent-based paradigm for quantitative information. Psyarxiv, doi:10.31234/osf.io/ b8kzw
[5] Armitage, K.L., Bulley, A., and Redshaw, J. (2020). Developmental origins of cognitive offloading. Proceedings of the Royal Society B, 287, 20192927.
[6] Carvalho, J.T. and Nolfi, S. (2016). Cognitive offloading does not prevent but rather promotes cognitive development. PLoS One, 11(8), e0160679.
Other references on nervous system tradeoffs and energetics:
Reséndiz-Benhumea et.al (2021). Shrunken Social Brains? A minimal model of the role of social interaction in neural complexity. Frontiers in Neurorobotics, 15:634085.
Gonzalez-Lagos et.al (2010). Large-brained mammals live longer. Journal of Evolutionary Biology, 2(3), 1064–1074.
Huang et.al (2012). Reduction of metabolic cost during motor learning of arm reaching dynamics. Journal of Neuroscience, 32, 2182–2190.
Hsu et.al (2000). Cost of cone coupling to trichromacy in primate fovea. Journal of the Optical Society of America A, 17(3), 635–640.
