UK Researchers Say AI Needs More Animal Sense

The incomplete understanding of human brains and how to endow computers with common sense are among AI’s most enduring challenges. New research from DeepMind London, Imperial College London and the University of Cambridge argues that common sense in humans is founded on a set of basic capacities that are also possessed by many other animals, and that animal cognition can therefore serve as inspiration for many AI tasks and curricula.

In a paper published in Trends in Cognitive Sciences journal this month, the researchers identify just how much AI research might benefit from the field of animal cognition.

There is no universally accepted definition of “common sense.” While much research has used language as a touchstone, the new paper temporarily sets language aside to focus on other common sense capacities found in non-human animals.

They such believe capacities pertaining to the understanding of everyday concepts such as objects, space, and causality are also a baseline for humans, and this “foundational layer of common sense, which is a prerequisite for human-level intelligence” could provide something that’s lacking in today’s AI systems.

The field of animal cognition has developed numerous experimental protocols for studying these capacities, the researchers say, and thanks to progress in deep reinforcement learning (RL) — a field DeepMind pays plenty of attention to — it is now possible to apply these methods directly to evaluate deep RL agents in 3D environments.

They argue that since animal cognition supplies a compendium of well-understood, nonlinguistic, intelligent behaviour, experimental protocols from the field of animal cognition can be repurposed for evaluation and benchmarking whether an agent “understands” a common sense concept or principle and could thus guide environment and task design.

Ideally, researchers would like to build AI technologies that can grasp inter-related principles and concepts as a systematic whole and manifest this grasp in a human-level ability to generalize and innovate. Although how to ultimately build such AI technologies remains an open question, the team suggests a path that involves training RL agents to acquire what is needed through extended interaction with rich virtual environments.

The researchers review several common sense physics tasks, with an emphasis on solid objects, to make a number of methodological points about training and evaluating RL agents. They look at containers and enclosures, as there are abundant examples in human and non-human behaviour — often connected with obtaining food or other rewards — where one object contains and presents a barrier to another, desired object. As squirrels breach the shell of a hazelnut or crows use a stick to extract a grub from a tree hollow, so do humans crack open the refrigerator door in their extractive foraging process. For RL agents, understanding and generalizing the concept of a container would be a very useful bit of common sense.

The paper argues that physics is just one common sense domain that the AI research community has been neglecting in this regard. There are others, such as common sense involved in psychological concepts (e.g. believing something or being unhappy) and the huge and variable domain of common sense social concepts (such as being with another agent, or giving something to someone), that researchers can also study on RL agents with inspiration from animals.

The team believes this exploration of cross-species common sense can contribute to improved architectures, environments, tasks, and curricula, and potentially equip RL agents inhabiting simulated 3D worlds with a repertoire of fundamental common sense concepts and principles. When that time comes, they say, we might at last be in a position to tackle language.

The paper Artificial Intelligence and the Common Sense of Animals is on Cell Press Reviews.

---

Reporter: Yuan Yuan | Editor: Michael Sarazen

---

Synced Report | A Survey of China’s Artificial Intelligence Solutions in Response to the COVID-19 Pandemic — 87 Case Studies from 700+ AI Vendors

This report offers a look at how China has leveraged artificial intelligence technologies in the battle against COVID-19. It is also available on Amazon Kindle. Along with this report, we also introduced a database covering additional 1428 artificial intelligence solutions from 12 pandemic scenarios.

Click here to find more reports from us.

---

We know you don’t want to miss any news or research breakthroughs. Subscribe to our popular newsletter Synced Global AI Weekly to get weekly AI updates.