AI and Human Decision-Making: The Role of the Reward Hypothesis

NNicholas September 10, 2023 6:23 AM

As artificial intelligence (AI) continues to advance, a pertinent question arises: can AI make decisions like humans? The Schwartz Reisman Institute for Technology and Society addresses this, with the 'Reward Hypothesis' taking center stage in the discussion.

Understanding the role of technology in society

The Schwartz Reisman Institute for Technology and Society is at the forefront of the quest to understand and guide the role of technology in our world. With AI advancing at breakneck speed, the institute concentrates on exploring its ethical and societal repercussions. The aim is to ensure the deployment of AI technologies in a manner that is not just safe but also favorable for everyone. This is accomplished by evaluating and redefining the role technology plays in society, the needs of human communities, and the systems that govern them.

A significant part of the conversation at the conference was about the 'Reward Hypothesis'. This theory, proposed by AI research pioneer Richard Sutton nearly two decades ago, states that all our goals and purposes are driven by the desire to maximize some kind of reward, be it social or material. So according to this hypothesis, the crux of intelligence, whether human or artificial, lies not in mimicking human behavior but in achieving set goals. This idea forms the foundation of classic AI theories and assists us in modeling decision-making in various fields.

Julia Haas's extended interpretation of the Reward Hypothesis

Julia Haas, a senior research scientist at DeepMind, brings forth a stronger version of the hypothesis, stating that our mind not only applies the reward concept to our goals and purposes, but it also constantly computes and assesses environmental features in terms of rewards. Essentially, our perception of the world is conditional on our goals, implying that we are inclined to remember things that we associate with higher rewards, whether social or material.

Applying the Reward Hypothesis to AI ethics and morality

The conversation also shed light on the topic of morality and ethics, suggesting that our experiences of right or wrong are nothing more than attributions of reward and value. By viewing morality through this lens, we can better comprehend the application of morality to artificial intelligence. Sutton and Haas, however, diverge when it comes to the implications of the reward hypothesis on morality and ethics, making for an enriching and riveting debate.

More articles

Also read

Here are some interesting articles on other sites from our network.