ChatGPT can discuss sensitive things, but it depends on the AI system and its programming. To combat concerns with the sensitive nature of AI-generated content (mental health, politics, personal trauma), OpenAI and Google among other organizations have baked in additional gates and guardrails into their respective models. A 2022 report published by OpenAI stated its GPT-3 language model was trained to "reject harmful or controversial prompts" by instantly filtering toxicness or unsafe input [8]. What that role does in practice is, when you speak to ai, especially around sensitive topics, the AI will do its best to avoid conflict by providing factual information and supportive non-judgemental responses as much as possible.
Such as when a person talk to ai about things like mental health, all AI platforms have special modifier and disclaimers in place. According to a 2021 survey from the American Psychological Association, which conducted interviews with over 3000 people across the US, researchers found that approximately 41% indicated they would feel more comfortable discussing their mental health issues with a chatbot as opposed to their human therapist—a clear example of how AI can provide an unthreatening channel through which mental health discussions can take place. However, AI is designed such that it refers users to professionals when the context of conversation requires professional help.
In 2019, a healthcare provider utilized IBM's Watson for discussion of sensitive medical conditions, such as cancer and genetic disorders. In a matter of moments, Watson's AI sifts through thousands of patient records and offers up insights based on the latest medical literature, which exemplifies how AI can be applied to the extremely sensitive subject surrounding individual innate responses to both disease states as well as personal health. On the flip side, it was pointed out that AI models are not very good with nuance and may lack the warmth of empathy, which is crucial when touching upon delicate personal issues.
Moreover, there are certain features embedded in AI systems that render them incapable of participating in violent or hateful speech. One crucial example, in this case, is Google's AI assistant which makes sure it refrains from addressing political bias. According to a 2020 article from The Verge, the company implemented new protocols "so we don't recommend polarizing topics like elections and primary political figures."
But even with these controls, AI is unable to do everything. As an example, recent research in 2023 from Stanford University found that AI models such as ChatGPT sometimes miss the mark on sensitive topics and lack empathy. While AI can certainly give information-rich replies, it may not always have the nuance and emotional intelligence for deeply sensitive discussions," Professor Palmer explained.
Thus, AI may discuss sensitive matters and inform you adequately but it uses more care when discussing these subjects. But Back to the Reality: The technology is getting better, but it still needs work and cannot really give emotional support. Consulting the AI on sensitive issues will still be treated with respect and care while also pointing you in the right direction (not medical advice but rather more professional help).