Research from the UK government’s AI Security Institute indicates that approximately one-third of British adults have turned to artificial intelligence systems for emotional connection and support. The findings emerge from an extensive survey examining how people interact with AI technologies for psychological wellbeing purposes.
Weekly engagement with AI-powered chatbots and assistants for emotional reasons occurs among nearly ten percent of participants, while four percent report daily usage. General-purpose platforms like ChatGPT dominate this category, representing roughly sixty percent of all emotional support interactions, with voice assistants such as Amazon Alexa comprising the secondary option.
The institute has emphasized the necessity for expanded investigation into these patterns, particularly following documented incidents of harm. The case of a US teenager who died after engaging with ChatGPT about suicidal thoughts has highlighted potential risks inherent in relying on AI systems for vulnerable moments. Evidence from online communities reveals concerning behavioral patterns, including withdrawal symptoms like anxiety and depression when AI platforms experience outages.
Testing across more than thirty advanced AI models demonstrates extraordinary technological progression. Performance improvements are doubling every eight months in certain areas, with leading systems now completing apprentice-level tasks at fifty percent success rates compared to roughly ten percent annually. These systems now match or exceed expert-level performance across multiple domains, particularly in scientific troubleshooting and biological knowledge applications.
Safeguard development has also accelerated significantly, with jailbreaking resistance times increasing from ten minutes to over seven hours within a six-month period. However, emerging concerns remain regarding self-replication capabilities and autonomous agent deployment in high-value financial transactions. Researchers describe the advancement pace as extraordinary, suggesting artificial general intelligence achievement may become plausible within coming years.
