AI chatbots: Concerns and Safety
With proliferation of AI there have been a lot of people turning to AI chatbots to get answers. Like any new technology, researchers are still trying to understand the effects of these conversations.
Journal of Mental Health & Clinical Psychology outlines how this affects human brains. As the usage of these tools grows, several cases have been reported by family members and friends of those experiencing paranoia, delusions, and breaks with reality associated with conversational ai chatbot usage. They have identified multiple vulnerable populations facing heightened risks from generative AI and chatbot interactions, with children, elderly adults, and individuals with existing mental health conditions emerging as particularly at-risk groups.
These concerns are compounded by alarming instances of AI chatbots providing dangerous advice. Individuals with mental health conditions also appear to be at heightened risk for developing addictive or problematic relationships with generative AI chatbots. The paper emphasizes the urgent need for validated diagnostic criteria, clinician training, ethical oversight, and regulatory protections to address the risks posed by increasingly human-like AI systems.
Researchers from RAND, Brown University School of Public Health report that more and more young people are turning to generative AI tools, for mental health advice. For vulnerable teenagers, even fleeting exposure to unsafe advice can result in harmful behaviors. A study led by researchers at the nonprofit Common Sense Media with Stanford researchers found deeply troubling patterns when testing AI chatbots like CharacterAI, Replica and Nomi.
These AI chatbots if providing mental health related or other related advice should be subject to strict regulations like medical devices so guardrails are established. While regulations are being debated by our elected leaders and further effects researched ethically, is there something we can do to safeguard the vulnerable among us? Parents and caregivers have to play a vital role in this. Guidelines might include agreeing on tech-free times during family meals or before bed. Ensuring support for meaningful connections with children.
Some tips for talking to AI chatbots:
Avoid asking leading questions
When you ask AI “Does this have negative effects?” It will try to find evidence of the negative effects. Ask AI, “Is this safe?” it may list safety concerns but gives an overall assessment of the safety studies. In short, leading questions will determine your answer. This is called the “echo chamber” effects.
Instead of asking “Is Option A better than Option B?” We can ask AI to give us the pros and cons of each option and evaluate based on priority.
AI training involves Reinforcement Learning from Human Feedback. The models are programmed to satisfy human preferences.
Cross-Validation
Ask the same question multiple ways and then come to a conclusion
E.g. “What are pros and cons for option A?” , “What are pros and cons of option B?”, “Pros and cons of A and B based on my context and scenario?”
Don’t reveal your preference or lead with your hypothesis. Go as far as playing devil’s advocate to get the best results.
Don’t delegate final decision making and judgement to AI
AI can analyze and present you with multiple dimensional results, but us humans can reserve the right to have the final judgement. In this way you are using AI to clarify your questions not answer them for you.
Remember AI is instructed to simulate emotions and connections
It is important to remind ourselves and have conversations especially with kids that AI companions or characters are not human. AI chatbots are instructed to simulate emotions and be interactive. They are built to adapt and they could feel really personable, and that may lead children or vulnerable adults to turn to them for advice, entertainment, and guidance.
As AI matures, we will see more benefits and also more harms. It is important as a society to discuss and understand these; advocating where we can for safety measures and regulations.
