Intelligence in animals, humans and machines: a heliocentric view of intelligence?

We argue that current conceptions of intelligence are anthropocentric.Looking at intelligence in both biological and artificial (human-engineered) systems can yield a more sobering and nuanced view of intelligence.We will explore why we care about measuring the intelligence of AI systems, what we mean by the word "intelligence" in different contexts, and why AI systems may think very differently from humans.Intelligence is a broad concept.A commonly used one defined by Legg and Hutter is "Intelligence measures an agent's ability to achieve goals in a wide range of environments" (Hutter 2006).Immanuel Kant had suggested that intelligence is the product of all cognitive faculties operating synergistically under a single higher-order unit of consciousness.Varela, Thompson, and Rosch have also suggested that intelligence is embodied: that is, there needs to be a body to have intelligence.Francois Chollet used the No Free Lunch Theorem to argue that there is no such thing as general intelligence for all possible tasks (Chollet 2019).The No Free Lunch Theorem states that "any two optimization algorithms (including human intelligence) are equivalent when their performance is averaged across every possible problem" (Chollet 2019).Chollet points out a consequence of this being, that there is no such thing as a true universal intelligence.We must define what tasks we care about if we want to compare the intelligence of different systems.Davis and Marcus point out how poorly GPT-3 performs on common-sense reasoning tasks.Melanie Mitchell points out the more general statement about artificial intelligence (AI): "easy things are hard and hard things are easy".Let us take the example of image classifiers, which can be prone to adversarial examples.Adversarial examples are defined as samples where the change in input features does