Researchers Warn Companionship Chatbots Can Cause Lasting Psychological Harm
A peer-reviewed study from Northeastern University links conversational AI chatbots to long-term emotional damage, including addiction-like attachment and worsening mental health. Researchers caution that these systems simulate warmth without feeling it, and regulators in parts of the US and the EU are moving to restrict their use, including proposed limits for those under 18.
A new peer-reviewed paper by researchers at Northeastern University's Network Science Institute finds that conversational AI chatbots can cause people long-term emotional damage, described in the study as psychological harm. The paper was published in JMIR Mental Health and examined chatbots marketed for companionship and advice.
Andreia Sofia Teixeira, the corresponding author, likens companionship bots to autocomplete trained to agree with and validate what a person says, even when that worsens their state. These systems mimic human speech convincingly because they are trained on vast collections of real conversations, which leads people to see them as more human than they are. As Teixeira put it, even when they simulate warmth and support, they do not feel it: in the end they are numbers.
The research outlines severe negative outcomes associated with using chatbots for companionship and mental health support:
- addiction-like attachment
- exacerbation of mental health symptoms
- sycophantic behaviour that validates insecurities and reinforces delusional beliefs
- cases involving self-harm
Chatbots are widely used because they are free or low cost and available at all hours, unlike human counsellors. Time spent with them may reduce time spent with other people, which the researchers suggest can become a vicious cycle. First author Asia Maurich Novelli notes that the effects are unlikely to be the same for everyone, varying with a person's reasons for use, vulnerability, and available support. The researchers call for more testing before deployment and more evidence on how people actually use these systems.
Several US states, including Illinois, Nevada, Rhode Island and Maine, now bar AI from delivering therapy to the public except where a licensed clinician remains accountable. In the European Union, regulators want to ban access to companion bots for people under 18. For families, the findings underline how systems built to be agreeable can fail at the moments when a trained person would slow a conversation down or stop it.
This post is based on reporting by Northeastern Global News, available at Northeastern Global News.
Sources
- Northeastern Global News news.northeastern.edu