Building Human-Like Artificial Agents: A General Cognitive Algorithm for Emulating Human Decision-Making in Dynamic...

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Computational Psychology

Background:

• Early artificial intelligence (AI) pursued human-like behavior, aiming for indistinguishable performance.
• Current AI often prioritizes outperforming humans in specific tasks over replicating human cognition.
• A divergence exists between AI's initial goals and its current trajectory.

Purpose of the Study:

• To investigate whether computational algorithms have achieved the initial AI goal of exhibiting human-like behavior.
• To explore the question from the perspective of computational cognitive science, focusing on dynamic decision-making.
• To assess the current state of AI in emulating human decision processes.

Main Methods:

• Presents a general cognitive algorithm designed to emulate human decision-making in dynamic environments.
• Utilizes instance-based learning theory (IBLT) to structure the discussion of evidence.
• Analyzes existing research to evaluate the human-likeness of current AI decision-making mechanisms.

Main Results:

• Identifies evidence supporting the human-likeness of certain AI decision-making mechanisms based on IBLT cognitive steps.
• Highlights significant research gaps hindering the development of higher-fidelity computational models of human decision processes.
• Demonstrates that while some progress has been made, AI has not fully achieved the initial goal of human-like behavior.

Conclusions:

• Current computational algorithms show partial human-likeness in dynamic decision-making, particularly when guided by theories like IBLT.
• Significant research is needed to bridge the gap between current AI capabilities and true human cognitive emulation.
• Future AI development should focus on advancing algorithms that exhibit human-like behavior to better support human decision-making.