NeuroAI paper proposes "Embodied Turing Test" to evaluate AI (2023)

Neuroscience has historically played a crucial role in advancing artificial intelligence (AI), and a new perspective emphasizes the need for further investment in NeuroAI to accelerate AI development. The authors propose the "embodied Turing test," which evaluates AI models based on their ability to interact with the physical world in ways comparable to living animals, rather than focusing solely on human-like capabilities such as language and game playing. This approach aims to leverage insights from over 500 million years of evolution, emphasizing sensorimotor skills that are foundational to intelligence. Current AI systems, while proficient in specific tasks, struggle with general adaptability and sensorimotor interactions, which are innate to even simple animals. The authors argue that understanding biological intelligence can inform the design of AI systems that match human capabilities. They advocate for a large-scale research initiative to explore the principles of biological intelligence, which could lead to significant advancements in AI. The embodied Turing test serves as a benchmark for evaluating AI's progress in achieving these goals, focusing on the shared sensorimotor abilities across species.

- Neuroscience is essential for advancing AI, particularly through the emerging field of NeuroAI.

- The embodied Turing test evaluates AI based on its sensorimotor interactions with the environment.

- Current AI systems lack the adaptability and physical interaction skills of even simple animals.

- Understanding biological intelligence can guide the development of more capable AI systems.

- A large-scale research initiative is needed to explore the principles of biological intelligence for AI advancement.