The company behind AI apps ChatGPT and DALL-E is trying out a new audio feature for ChatGPT in its GPT-4o form. This will enable ChatGPT to “talk” in real-time to users. The downside, however, is that, according to OpenAI, ChatGPT’s voice mode has the potential to be emotionally addictive.
This exceptional new technology can respond to audio input in 232 milliseconds, which is close to human responses, according to experts.
In a demonstration carried out last May, Open AI showed that ChatGPT’s voice mode could also translate conversations in real-time. One might think that this tech is straight out of the plot of a Sci-Fi movie, and that’s because it is.
The revolutionary ChatGPT voice mode is quite similar to the AI in the 2013 movie Her, in which Joaquin Phoenix’s character falls in love with an AI device voiced by Scarlett Johansson. This is exactly what OpenAI wants to avoid.
OpenAI warns about the risks of anthropomorphization
In the GPT-4o System Card, OpenAI warns about the novel risks of ChatGPT’s voice mode on users. One of the main risks is anthropomorphization.
Essentially, anthropomorphization refers to people’s attribution of human behaviors onto non-human entities.
In the report, OpenAI warns, “This risk (of anthropomorphization) may be heightened by the audio capabilities of GPT-4o, which facilitate more human-like interactions with the model.”
OpenAI also explained that during tests with ChatGPT’s voice mode, users may be witnessed “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.’”
The company also claims that users of ChatGPT’s voice mode may form a relationship with the technology, which could benefit lonely individuals. Nonetheless, this could have a negative impact on healthy relationships over longer periods of time.
In the report, OpenAI claims that ChatGPT’s voice mode can also increase access to health-related information and improve health-related workflows.
OpenAI researchers assessed ChatGPT-4o’s clinical knowledge by formulating 22 text-based tests with 11 separate data sheets. The idea behind these tests is to assess whether ChatGPT actually comprehends the tests or not.
The results were remarkable. OpenAI’s team observed that GPT-4o improved over the previous model in 21 out of 22 evaluations. It also exceeded test results of medical-focused AI such as Med-Gemini L by 84 percent.
OpenAI is concerned with GPT-4o’s societal impacts
Researchers at OpenAI are also concerned with this innovative model’s societal impacts. They fear the technology could be used for representational harm, spreading misinformation and becoming entangled in other crimes such as fraud.
Overall, the company is expecting to debut the technology later this year for OpenAI’s premium users.
Additionally, OpenAI’s researchers found that text-based tests look good so far, but more experiments are needed to determine if satisfactory results are achieved when both text and audio are used.
