Recommenders for Humans

Typical recommender systems treat humans in a way that lacks humanity.

We’ve all experienced the trap of a news site with a modern recommender system: one moment you're clicking through articles, and the next you realize an hour has vanished. Despite these systems being engineered by experienced UX experts with deep-learning algorithms that maximize engagement, something still goes awry. So, what might we be missing

pixabay.com

1. Whom Should We Serve?
   Should recommendations cater to the person in the moment, to the long-term self, or to a future version of the same person? User objectives vary with time. For example, when I recommend a book to a friend, I’m not just considering its immediate appeal, but also its capacity to leave a lasting, meaningful impression. We need “system 2” recommenders—those that balance short-term gratification with long-term value.

2. Is Utility Hacking Acceptable?
   Recommendation systems can be designed in ways that subtly alter users’ values—steering them, perhaps inadvertently, toward materialism or other specific goals. Given our duty of care to respect what people truly stand for, it’s critical that these systems do not “hack” a user’s utility. In other words, recommendations should align with, rather than undermine, a user’s genuine aspirations.

3. The Value of Empathy and Social Connection
   Human recommendations carry an inherent depth because they are embedded in social relationships. A recommendation from a friend is not just a suggestion; it’s an expression of care and a link that strengthens social bonds. Recommender systems should be designed not to undermine human to human recommendations.

4. Supporting Executive Function
   People sometimes invest in tools to help them resist short-term temptations—like setting aside money in a retirement account to avoid impulsive spending. In a similar way, recommendation systems should support our ability to exercise self-control rather than simply gratifying immediate desires. Unlike “sugar-coated” recommendations that might yield regret later, thoughtfully curated suggestions can enhance our long-term well-being.

Key Insight:
All these factors—serving multiple temporal selves, preventing manipulative utility hacking, nurturing empathy and social connection, and bolstering executive function—should be integral to the design of recommender engines. And if current systems fall short, users should have the right to implement their own filtering mechanisms to ensure that what they see aligns with their true values and long-term interests.