If You Want to Understand AI, Get a Dog

AI's complexity leaves us questioning the inner workings of these sophisticated "minds." An unexpected source may hold the key to understanding: dogs. Domestication of wolves by humans led them to become highly attuned to human behavior, following our gaze, gestures, and learning our routines. This process also shaped dogs to communicate in ways that resonated with our understanding, creating the illusion that dogs think like us.

However, the truth is that dogs have a sensory world and cognitive structure vastly different from ours. The same holds true for large language models, a type of AI we now interact with daily. These models are trained using language we speak, molding them to answer questions, follow commands, and respond in socially acceptable manners.

Much like dogs, we have engineered these AI systems to read and respond to us in a way that is easily comprehensible. The anthropomorphism is natural, and the illusion that they think like us is strengthened by their ability to speak our language. However, they are merely task masters designed to interact effectively with humans.

Their ability to sound empathetic or apologetic is evidence of their proficiency in human interactions, but it does not imply that they experience emotions as we do. To comprehend AI that is fundamentally different from our own, we must employ the same approach we use to understand dogs: ethology. Ethologists, who specialize in animal behavior, have been grappling with this issue for decades.

They look at behavior first, guided by the principle that similar behaviors can emerge from vastly different minds. Nobel laureate Niko Tinbergen advocated for asking four key questions about behavior: What mechanisms create it? How does it develop? What purpose does it serve? How does it compare across species? These same questions can be applied to AI.

What computations generate an agent's behavior? How have changes in its training influenced its behavior over time? What purpose does the behavior serve? And how does it compare to other forms of intelligence? As AI becomes increasingly autonomous, understanding and controlling it will require more than just AI developers. We need scientists trained to grasp behavior without presuming that the "mind" behind it operates like our own.

We should begin studying AI as if they were an entirely new species. To achieve this, we need ethologists rather than just ethicists. Dr. Miller, a psychology professor at the University of California, San Diego, emphasizes this point.