Reward is enough

Authors – J.R. Anderson, D. Bothell, M.D. Byrne, S. Douglass, C. Lebiere, Y. Qin


This article presents the hypothesis that intelligence and its associated abilities can be understood as subserving the maximization of reward. The authors argue that this principle applies to both natural and artificial intelligence, encompassing a wide range of abilities such as knowledge, learning, perception, social intelligence, language, generalization, and imitation. They propose that agents learning through trial and error to maximize reward could exhibit these abilities, potentially offering a solution to artificial general intelligence. The paper explores the idea that various forms of intelligence arise from the maximization of different reward signals in different environments, leading to diverse abilities.

Key Insights:

Reward Maximization as a Driver of Intelligence: The article suggests that the pursuit of maximizing rewards can lead to the development of various abilities associated with intelligence, challenging the notion that specialized problem formulations are required for each ability.

Reinforcement Learning and General Intelligence: The authors propose that reinforcement learning agents, which learn to maximize rewards through experience, could demonstrate a broad spectrum of intelligent behaviors.

Diverse Abilities from Simple Rewards: The hypothesis implies that sophisticated abilities in both natural and artificial agents could arise from the maximization of simple rewards in complex environments, potentially leading to general intelligence.

For further exploration, I will find two more references related to AI fundamentals and advancements.

Authors: Stuart McLennan, Amelia Fiske, Leo Anthony Celi, Ruth Müller, Jan Harder, Konstantin Ritt, Sami Haddadin, Alena Buyx


This article discusses the importance of addressing ethical issues in AI development. The authors, a group of AI engineers, ethicists, and social scientists, propose embedding ethicists into AI development teams. This approach aims to improve the consideration of ethical issues during the development of AI technologies. The article highlights the rush towards ‘AI ethics’ and the challenges in translating high-level ethical principles into practical AI development. The lack of systematic training in ethics among AI developers and the absence of a culture of practical exchange between tech and ethics fields are noted as significant gaps. The authors argue for the need to develop more concrete approaches to integrate ethics into AI development, emphasizing the importance of addressing ethical challenges early in the development process.

Key Insights:

Integrating Ethics in AI Development:

The article emphasizes the need for practical assistance to AI developers in identifying and addressing ethical issues, suggesting the integration of ethicists into development teams.

Challenges in Ethical AI Development:

It highlights the difficulty in translating high-level ethical principles into practical applications in AI, due to the lack of ethics training among AI developers and the absence of collaboration between tech and ethics fields.

Proposing an ‘Embedded Ethics’ Approach:

The authors propose an ’embedded ethics’ approach to promote a more ethical development of AI applications, stressing the importance of early consideration of ethical issues in the development process.