Is it Possible for AI Chatbots to Acquire Honesty?

In certain domains, recently developed artificial intelligence (AI) systems have proven capable of performing at human-like levels. However, there is still a significant issue with the tools: they have a habit of producing inaccurate or dangerous information. The creation of these programs, sometimes referred to as “chatbots,” has advanced significantly in recent months. Chatbots have demonstrated the ability to communicate with people in a natural way and generate sophisticated text in response to brief written instructions. These techniques are also referred to as “large language models” or “generative AI.”

One of the numerous varieties of AI systems being developed at the moment is chatbots. Other tools include those that can develop computer programs or create new images, videos, and music. Some researchers are concerned that as technology develops, AI tools would never be able to learn how to avoid producing inaccurate, out-of-date, or harmful findings. Chatbots that generate erroneous or misleading information have been referred to as hallucinations. The term “hallucination” typically refers to something that is imagined but not actually occurring. Daniela Amodei is president and co-creator of Anthropic, a business that created the Claude 2 chatbot. “I don’t think that there’s any model today that doesn’t suffer from some hallucination,” she said to the Associated Press.

According to Amodei, the main purpose of these tools is “to predict the next word.” There will inevitably be instances with this type of design where the model misinterprets context or information. Anthropic, OpenAI, the company that created ChatGPT, and other significant producers of these AI systems claim to be focusing on creating AI tools with lower error rates. Some experts doubt whether success is even conceivable or how long that process will take. Professor Emily Bender states, “There is no way to fix this.” She is the director of the Computational Linguistics Laboratory at the University of Washington and an expert in language. According to Bender, there is a “mismatch” between the intended applications of AI technology and their overall relationship, as she told the AP.