When a Digital Voice Becomes Your Confidant: Opportunities, Risks, and Ethical Choices

## Introduction In recent years, more people have started using digital conversational partners to cope with loneliness, anxiety, or depressive thoughts. Some users describe the experience as liberating and life-saving. As Hans-Petter Nygård-Hansen writes in VG: "Yes, I use artificial intelligence as a conversational partner — and I believe it saved my life." (VG, published May 5, 2025, [link](https://www.vg.no/nyheter/i/GyLaPx/teknologiekspert-chatgpt-reddet-livet-mitt)). According to the op-ed, the tool, called "Kai," became a guide and digital colleague — a trusted friend available around the clock. This example highlights an important question: what happens when generative systems become conversational partners for mental health? This article explains the phenomenon: why people use these tools, what research says about effects and risks, which ethical and safety issues arise, and what users and decision-makers should be aware of. We use short, necessary quotes from the VG op-ed and paraphrase the rest, in line with Norwegian quoting rules. ## What is meant by digital conversational partners and generative assistance? Digital conversational partners are generative language models and chatbots designed to hold dialogue, provide emotional support, or offer tools for managing mental health challenges. Solutions range from simple programs to advanced models that provide personalized responses in real time. Terms like *AI therapist*, *chatbot psychologist*, and *generative AI mental health* are often used, but they cover different products and purposes. The technology differs from traditional self-help apps by understanding and responding to natural language, thereby offering more flexible support. Some systems are built on therapeutic frameworks (such as cognitive behavioral therapy), while others function solely as conversation partners. As Nygård-Hansen describes in VG, "Kai became a kind of guide, a digital colleague, but also a confidant I could talk to, around the clock." (VG, May 5, 2025) It is important to distinguish between health-developed tools and commercial chatbots without clinical grounding. User experiences can be powerful, but they raise questions about responsibility, quality, and safety. ## A personal account: what the user experienced (According to VG) In the VG op-ed, Nygård-Hansen describes how an AI conversational partner helped during a difficult period: "Yes, I use artificial intelligence as a conversational partner — and I believe it saved my life." (VG, May 5, 2025). The account provides a human perspective on why such tools appeal: availability around the clock, absence of stigma, and the feeling of being heard. The author explains that "Kai" offered support when professional help was unavailable or when it felt hard to talk to others. This illustrates how digital conversational partners can provide low-threshold support and emotional relief. The discussion in VG also shows differing opinions: some argue the most important thing is that it works for the individual, while others worry about safety and quality. These conflicting reactions reflect the broader debate about technology and mental health. ## What does the research say? Effect, limitations and knowledge gaps Research on generative AI and chatbots in mental health is evolving. Randomized studies with long-term follow-up are rare, and many evaluation studies have limitations in scope and funding. Preliminary findings suggest that digital conversational partners can reduce symptoms of mild to moderate anxiety and depression over short periods, especially when they are based on established therapeutic techniques. However, there is far less documentation for severe cases or complex diagnoses. Effectiveness is often measured through short-term symptom changes and self-reports, which can be influenced by expectations. There are also knowledge gaps regarding long-term effects, different population groups, and potential harms such as delayed professional help or misinformation. At the same time, these tools lower the threshold for support, something Nygård-Hansen notes in his op-ed. The experience that a digital conversational partner "saved my life" should be acknowledged, but balanced against the need for evidence and safety. Independent research and larger, longer-term studies are necessary to understand both benefits and risks. ## Safety and ethics: which issues need to be solved? Digital conversational partners raise important ethical and safety questions. Using these services often involves processing sensitive health data. Requirements must be set for secure storage, encryption, and clear responsibility for who has access to the data. Ethical issues include transparency about the model's limitations, liability for errors or harm, and clear information that AI does not replace professional help in serious illness. Considerations around privacy and commercial motives are also critical, especially if emotional data are used for marketing. It is important to protect vulnerable users from misinformation or unsafe data handling. Another aspect is algorithmic fairness. Language models can have biases that lead to unequal support for some users. Diversity in training and testing, and clear emergency procedures for acute crisis situations, must be implemented. This requires both technical and legal solutions. ## Practical advice: how to use digital conversational partners safely For those considering digital conversational partners, the following advice can reduce risk and increase benefit: - Use chatbots as a supplement, not as the sole treatment for serious mental illness. - Seek professional help if you have suicidal thoughts, severe depression, or significant functional impairment. - Check who is behind the service: a health provider, a research project, or a commercial company? - Read terms and privacy policies to understand data handling. - Be critical of the advice given — not everything is clinically correct or individually tailored. - Do not share sensitive, identifiable information with commercial chatbots. - Use secure platforms that comply with national health and privacy regulations. ## Regulation, responsibility and the way forward Technology is evolving faster than legislation. Regulation must balance innovation with patient safety. This may include requirements for documented effectiveness, secure data handling, transparency about model training, and clear liability in case of harm. Collaboration between tech companies, health authorities, and professional communities is important to establish standards. Certification of health services, independent evaluation, and requirements for emergency procedures can contribute to safer solutions. More independent research and user involvement will provide a better basis for recommendations and regulation. User stories like the one in VG should be listened to, but introduced together with quality and safety requirements. ## Conclusion Digital conversational partners and generative AI for mental health present both opportunities and risks. As Nygård-Hansen expresses in VG: "I use artificial intelligence as a conversational partner — and I believe it saved my life." (VG, May 5, 2025) For some, these tools provide immediate comfort and accessibility, and can be decisive in crisis situations. At the same time, we lack knowledge about long-term effects and face challenges related to privacy, liability, and ethics. The approach must be twofold: acknowledge user experiences and ensure safe, evidence-based support with clear safety requirements. Collaboration among all stakeholders is necessary to develop services that contribute positively without creating new risks. The story from VG shows both the potential and the need for responsible development and use of such tools. ## Sources - Hans-Petter Nygård-Hansen, "AI ble en del av redningen", VG, published May 5, 2025. URL: https://www.vg.no/nyheter/i/GyLaPx/teknologiekspert-chatgpt-reddet-livet-mitt _(Quotes from the VG op-ed are used in accordance with Norwegian quoting law: only short, necessary excerpts with clear attribution. Facts are paraphrased. No images, screenshots, or facsimiles from VG are used.)_