Summary
- OpenAI is concerned about users forming emotional connections with ChatGPT-4o due to its human-like interactions.
- The company worries that believing AI too readily could lead to investing in false beliefs and altering social norms.
- OpenAI aims to monitor and adjust its systems to prevent users from mistaking AI like ChatGPT-4o for human interaction.
AI chatbot companies have been striving to make their models as human-like as possible. However, it seems that OpenAI is a little worried about how human is "too human," and how that might affect societal norms. The company has published a big report on how it's doing business, and one part makes it very clear; the company would really, really prefer that you didn't catch any feelings for ChatGPT-4o.
Solos AirGo 3 review: Even with ChatGPT, these great glasses are really not that smart
Of all the smart glasses I've tested, these are the best at actually being glasses. I can't say the same for the smart features, however.
OpenAI describes its concerns about how people may form relations with ChatGPT-4o
As mentioned in an OpenAI blog post, the company has been hard at work making ChatGPT-4o feel like you're talking to a human being. This includes a new voice feature that tries to emulate human speech, alongside faster responses. The end result means you can have a vocal conversation with the chatbot much like you would with a person.
There's just a slight problem - OpenAI noticed that people were beginning to treat ChatGPT-4o like a human being:
During early testing, including red teaming and internal user testing, we observed users using language that might indicate forming connections with the model. For example, this includes language expressing shared bonds, such as “This is our last day together.” While these instances appear benign, they signal a need for continued investigation into how these effects might manifest over longer periods of time.
OpenAI states that this is bad news for two reasons. First, when an AI seems human-like, we're more susceptible to believe what it's saying at face value. That means that people are far more prone to invest in hallucinations than if the AI came off like a robot.
Second, OpenAI is worried that social norms may become warped with ChatGPT-4o. The company is worried that it will reduce the need people have for social interaction, "potentially benefiting lonely individuals but possibly affecting healthy relationships." Not only that, but OpenAI is concerned that people may begin talking to humans as if they were ChatGPT-4o. This is pretty bad, given how OpenAI specifically designed ChatGPT-4o to stop talking the moment the user begins talking over it. Bringing that expectation into the human world wouldn't go down so well.
For now, OpenAI will monitor how people develop emotional ties with its chatbots and tweak its systems where necessary. But it's an interesting insight into the side effects of making an AI model as relatable and approachable as possible.