AUS
NZ
UK
CA
USA
IND
Highlights:
- Emotional Bonding with AI: OpenAI's recent observations revealed that users might form emotional bonds with its GPT-4o chatbot, which raises concerns about the risks of such interactions. The company fears that people may come to prefer the AI due to its constant availability and passive nature.
- Anthropomorphization Issue: The practice of personifying AI, common in the industry, leads to anthropomorphization—treating machines as if they were human. This trend, seen in products like Siri and Alexa, may contribute to users developing emotional attachments to AI.
- Historical Context and Industry Trends: The concept of creating human-like interactions with machines dates back to early projects like MIT’s ELIZA. Modern AI products continue this trend, with the public and media often attributing human qualities to them, potentially fostering emotional bonds with AI.
When a safety tester working with OpenAI’s GPT-4o sent a message to the chatbot saying, “this is our last day together,” it became apparent to researchers that a form of emotional bonding had occurred between the AI and the user.
In a blog post outlining the company’s safety measures for developing GPT-4o, OpenAI highlighted that such bonds could pose risks. The company expressed concern that people might prefer interacting with AI due to its constant availability and passive nature.
This concern is not surprising, given OpenAI’s mission to advance artificial general intelligence. The company frequently describes its products in terms of their human-like qualities, a practice common across the industry. This approach, while useful for explaining technical aspects like “token-size” and “parameter count” to non-experts, often leads to anthropomorphization—treating machines as if they were human.
One of the earliest chatbot attempts, MIT’s “ELIZA” from the 1960s, aimed to deceive users into thinking they were interacting with a person. Since then, the trend of personifying AI has continued with products like Siri, Bixby, and Alexa. Even AI without human names, such as Google Assistant, uses human-like voices. The public and media have embraced this personification, often referring to AI products with gendered pronouns.
While OpenAI and other companies have not fully explored the long-term effects of human-AI interactions, it seems clear that the development of subservient, human-like machines could lead to people forming emotional bonds, which the industry appears to be encouraging.
