Artificial Intelligence and the Common Sense of Animals

Abstract

The problem of common sense remains a major obstacle to progress in artificial intelligence. Here, we argue that common sense in humans is founded on a set of basic capacities that are possessed by many other animals, capacities pertaining to the understanding of objects, space, and causality. The field of animal cognition has developed numerous experimental protocols for studying these capacities and, thanks to progress in deep reinforcement learning (RL), it is now possible to apply these methods directly to evaluate RL agents in 3D environments. Besides evaluation, the animal cognition literature offers a rich source of behavioural data, which can serve as inspiration for RL tasks and curricula. Common Sense before LanguageThe challenge of endowing computers with common sense has been seen as a major obstacle to achieving the boldest aims of artificial intelligence (AI) since the field's earliest days [1] and it remains a significant problem today [2-6]. There is no universally accepted definition of common sense. However, most authors use language as a touchstone, following the example of [1], who stated that '[a] program has common sense if it automatically deduces for itself a sufficiently wide class of immediate consequences of anything it is told and what it already knows'. Consequently, tests for common sense are typically language based. For example, one such test uses so-called 'Winograd schemas' [7][8][9]. These are pairs of sentences that differ by a single word and contain an ambiguous pronoun whose resolution depends on understanding some aspect of common sense. Consider the sentences 'The falling rock smashed the bottle because it was heavy' and 'The falling rock smashed the bottle because it was fragile'. The pronoun 'it' refers to the rock in the first sentence, but to the bottle in the second. We are able to resolve the pronoun correctly in each case because of our common sense understanding of falling and fragility. In this paper, by contrast, we will set language temporarily to one side and focus on common sense capacities that are also found in non-human animals. Our rationale is that these capacities are also the foundation for human common sense. They are, so to speak, conceptually prior to language and human language rests on the foundation they provide [10]. HighlightsEndowing computers with common sense remains one of the biggest challenges in the field of artificial intelligence (AI).

Primary Cognitive Categories Are Determined by Their Invariances

“…In other words, it is what we can do with objects that matter, not how they look. Shanahan et al (2020) discuss this problem. As an example, they take the concept of a "container" that is central to much human interaction with the world.…”

“…The appearance of containers can vary widely, but it is their affordances that are crucial for how we interact with them. There seems to be no good model of how to capture the affordances of objects from, say, a video stream (Shanahan et al, 2020). As regards actions, they are understudied in robotics.…”

Social Robots as Intentional Agents

“…model on which most AI research is based, this precondition of intelligent behavior necessarily appears to AI researchers as the need to find "a formal representation in which all the knowledge and beliefs of an average adult human being can be made explicit and organized for flexible use". While it suffices for our purposes to note that endowing computational technology with common sense remains one of the biggest challenges in the field of AI (Shanahan et al, 2020;Mitchell, 2021), the arguments presented by Dreyfus (1992) cast doubt on whether a satisfactory formalization of common-sense knowledge can ever be achieved.…”

“…Endowing robots and other computer-based systems with folk psychology requires solving a more general problem that computer science has so far been unable to solve: the common-sense problem in artificial intelligence (AI) research (McCarthy, 1960;Davis & Marcus, 2015;Lake, Ullman, Tenenbaum, & Gershman, 2017;Shanahan, Crosby, Beyret, & Cheke, 2020;Levesque, 2017). As argued by Dreyfus in his 1992 book What Computers Still Can't Do, "Intelligence requires understanding, and understanding requires giving a computer the background of common sense that adult human beings have by virtue of having bodies, interacting skillfully with the material world, and being trained into a culture. "…”

Direct Human-AI Comparison in the Animal-AI Environment

“…6-29-2 was a variant of L1 basic navigation tasks but with colourings of the walls, roof, and floor altered. A suggestion for why the AIs performed considerably worse than the children in the L6 tasks, and compared to their performances in comparable levels (e.g., L1, L2, and L4) is that the AIs used perceptual cues of colour or shape to perform successfully, whereas children are capable of abstracting beyond those cues to more features pertaining to objecthood (see Dubey et al 2018;Shanahan et al 2020). However, post hoc analyses showed that performance on 6-29-2 and the analogous task 1-4-3 were significantly different, as were performances on 3-11-1 compared to 6-12-2.…”