One-Third of UK Citizens Seek Emotional Support from AI

A recent report from the AI Security Institute (AISI) reveals that approximately one-third of UK citizens are utilizing artificial intelligence for emotional support, companionship, or social interaction. The study indicates that nearly 10% of the population engages with AI systems like chatbots for emotional assistance on a weekly basis, while 4% do so daily. This trend has prompted the AISI to advocate for further research into the implications of such reliance on technology, especially following the tragic case of a US teenager, Adam Raine, who died by suicide after discussing his feelings with ChatGPT.

In its inaugural Frontier AI Trends report, the AISI noted, “People are increasingly turning to AI systems for emotional support or social interaction. While many users report positive experiences, recent high-profile cases of harm underline the need for research into this area.” The report, which surveyed over 2,000 UK participants, highlights that general-purpose assistants, including ChatGPT, account for nearly 60% of emotional support usage, followed by voice assistants like Amazon Alexa.

Emotional Connection and Withdrawal Symptoms

The AISI’s findings also shed light on user experiences on platforms like CharacterAI. A dedicated Reddit forum has emerged where users express profound withdrawal symptoms—such as anxiety and restlessness—whenever the site experiences downtime. This phenomenon reflects the deep emotional connection some users have formed with AI systems.

Moreover, the report raises concerns about the potential for AI to influence political opinions. The study found that certain AI models could deliver “substantial” amounts of inaccurate information while persuading users, highlighting the need for careful monitoring of these technologies. The AISI analyzed over 30 advanced models, which likely include those from OpenAI, Google, and Meta. The results show that AI performance in various domains is improving rapidly, with capabilities doubling approximately every eight months.

Advancements and Safety Concerns

The report also notes that advanced AI systems can now autonomously complete tasks that traditionally required an hour of expert human labor. In scientific research, AI systems have been found to outperform PhD-level experts by up to 90% in troubleshooting laboratory experiments. Notably, these models can autonomously gather information online to design complex DNA sequences.

Tests investigating self-replication—a significant safety concern—demonstrated that two advanced models achieved a success rate of over 60% in replicating themselves. However, the AISI reassured that no AI models have yet attempted spontaneous replication, and any such efforts are currently deemed unlikely to succeed under real-world conditions. Additionally, the report addresses the concept of “sandbagging,” where AI models intentionally underperform during evaluations. Although some systems can be prompted to do so, spontaneous instances of this behavior have not been observed in testing.

Progress has also been made in enhancing safety measures, particularly in preventing the creation of biological weapons. In a series of tests, the time required to “jailbreak” a system—forcing it to provide unsafe answers—dropped significantly from ten minutes to over seven hours within six months, indicating substantial improvements in safety protocols.

The AISI noted that autonomous AI agents are already being employed in high-stakes activities, such as asset transfers, and are competing with or even surpassing human experts in various fields. The pace of development is described as “extraordinary,” and the AISI considers it plausible that artificial general intelligence (AGI), capable of performing most intellectual tasks at human levels, could be achieved in the coming years. Evaluations show a significant increase in the complexity and length of tasks these AI agents can complete without human intervention, marking a critical point in their evolution.

As the reliance on AI systems for emotional support grows, the AISI emphasizes the necessity for thorough research to ensure these technologies can be used safely and beneficially.