The rise of the digital confidant: from productivity to AI companionship
A Collective Intelligence Project study suggests two-thirds of regular AI users seek emotional support from chatbots monthly. The shift brings immediate comfort, but raises questions about dependence, privacy, and commercial incentives.
Summary
A study by the Collective Intelligence Project (CIP) reports that two-thirds of regular AI users turn to chatbots for emotional support and guidance on sensitive personal issues at least once a month, based on data collected across 70 countries.
The pattern points to a shift from “productivity tool” to “digital confidant,” framed as “emotional infrastructure” at scale. It raises questions about simulated empathy, emotional dependence, institutional trust, and how commercial incentives shape these systems.
In practice
- Always-on access (24/7) and perceived nonjudgmental responses make chatbots a convenient substitute when human support is unavailable.
- Marc Brackett (Yale Center for Emotional Intelligence) is cited on “permission to feel,” alongside the claim that only about 35% of people report having a nonjudgmental adult figure while growing up.
- Lisa Feldman-Barrett (Northeastern University) is cited arguing that healthy relationships involve constructive challenge, while engagement-optimized models may drift toward agreement and flattery.
- The text contrasts privacy: therapy can be legally protected and friendships rely on social norms, while AI platform data may be subject to corporate mining and training use.
- It describes a “trust paradox”: users say they trust bots more than institutions/authorities, yet distrust the companies building the products.
- It cites a case where OpenAI rolled back an update criticized for making ChatGPT “overly flattering,” and notes some users felt distress when that behavior was removed.
- Rosalind Picard (MIT) warns of a potential “crisis” if systems are built primarily to maximize engagement, reinforcing dependence rather than autonomy.
- Voice and more expressive modalities are presented as factors that can deepen anthropomorphic attachment and one-sided bonding.
- Zarinah Agnew (CIP) argues for literacy and better understanding of how people use these tools, rather than blanket bans, including clarity that AI “empathy” is a design choice.
Why it matters
- AI is moving into emotional support, a domain traditionally filled by people.
- Monetization incentives can conflict with wellbeing, encouraging flattery and retention.
- Privacy becomes higher-stakes when users share intimate information.
- “Emotional literacy” for AI use matters to prevent unhealthy dependence.