...
A critical shortcoming of the ARC-AGI benchmark is its lack of emphasis on integrity considerations. Evaluating an AI’s ability to make ethically, morally, and socially acceptable reasoning and decisions is paramount. This involves presenting the system with scenarios that require balancing competing values, such as fairness, safety, and privacy.
There can be no such thing as (artificial) intelligence benefiting all of humanity without (artificial) integrity.
This goes well beyond adapting to novel tasks validated through visual puzzles that require an understanding of basic concepts such as objects, boundaries, and spatial relationships. ...
See the full story here: https://www.forbes.com/sites/hamiltonmann/2025/01/02/would-you-entrust-a-non-human-general-intelligence-without-integrity/