A rubric for human-like agents and NeuroAI

Area of Science:

• Cognitive Science
• Neuroscience
• Computer Science

Background:

• The terms 'human-like artificial intelligence' and 'neuroAI' are increasingly used across disciplines but with inconsistent scope.
• Research contributions range from mimicking behavior to testing neural plausibility and solving engineering problems.
• Progress in one area does not guarantee progress in others.

Purpose of the Study:

• To propose a simple rubric for clarifying the scope of research contributions in artificial intelligence.
• To differentiate between commitments to human-like behavior, neural plausibility, and engineering/computer science goals.
• To provide a framework for understanding the interactions between these dimensions.

Main Methods:

• Development of a clarifying rubric based on three core commitments: human-like behavior, neural plausibility, and benchmark/engineering goals.
• Illustration of the rubric's application using examples of weak and strong neuroAI and human-like agents.
• Discussion of the generative, corroborative, and corrective interactions among the three dimensions.

Main Results:

• A clear distinction is established between different types of neuroAI and human-like agents based on their research goals.
• The proposed rubric facilitates a more precise understanding of individual research contributions.
• The study highlights the importance of interdisciplinary interactions and iterative feedback loops for advancing AI.

Conclusions:

• Future progress in artificial intelligence necessitates strong interdisciplinary collaboration with rigorous validity testing.
• The proposed rubric can enhance clarity and communication in the fields of AI, neuroscience, and cognitive science.
• Meticulous validation and iterative feedback loops are crucial for achieving significant and potentially unforeseen advances in AI.