ChatGPT and Other AI Chatbots Raise Concerns Over Falsehoods
Artificial intelligence (AI) chatbots like ChatGPT have been found to spout falsehoods, causing concern in various industries. Whether it’s businesses, organizations, or high school students using generative AI systems for document composition and work tasks, the issue of chatbots generating inaccurate information has become a problem.
Daniela Amodei, co-founder and president of Anthropic, the company behind chatbot Claude 2, stated that hallucination or fabrication of facts is present in every model today. These chatbots are designed to predict the next word, but there will always be a level of inaccuracy. Both Anthropic and OpenAI, the creators of ChatGPT, acknowledge this issue and are actively working on making their AI systems more truthful.
However, experts like Emily Bender, a linguistics professor, believe that the problem of falsehoods in AI chatbots is inherent and cannot be fully fixed. The proposed use cases for generative AI technology, such as providing medical advice, highlight the crucial need for accuracy that chatbots currently struggle with.
The reliability of generative AI technology carries significant weight. McKinsey Global Institute predicts that it could add trillions of dollars to the global economy. Chatbots are just one piece of this technological advancement, which includes AI systems generating images, video, music, and code. Almost all these tools incorporate a language component.
Large companies like Google are already exploring the use of AI in news writing, where accuracy is paramount. The Associated Press is also partnering with OpenAI to improve their AI systems using text from their archive. The potential consequences of hallucinations in AI-generated text are evident in various sectors, such as recipe creation or legal brief writing.
During a visit to India in June, Sam Altman, the CEO of OpenAI, addressed concerns about hallucinations raised by Ganesh Bagler, a computer scientist. Altman expressed optimism about resolving the issue within a year and a half to two years, finding a balance between creativity and accuracy in AI models.
However, some experts, including University of Washington linguist Bender, argue that improvements won’t be enough. Language models, like ChatGPT, essentially generate text by repeatedly selecting the most plausible next word. While they can be tuned to be more accurate, they will still have failure modes that are harder for readers to detect.
Although some marketing firms find benefit in chatbot hallucinations for generating new ideas, accuracy remains a concern. Shane Orlick, president of Jasper AI, expects companies like Google to invest in resolving this problem, given their responsibility for delivering accurate search engine results. While perfection might be challenging to achieve, continuous improvement can be expected.
There are techno-optimists like Bill Gates who believe AI models can be taught to distinguish fact from fiction. Research institutions are exploring methods to detect and remove hallucinated content automatically. Altman himself admits to not fully trusting ChatGPT’s answers when seeking information.
In conclusion, the issue of falsehoods generated by AI chatbots raises significant concerns for businesses, organizations, and individuals relying on their capabilities. While efforts are being made to improve accuracy, the inherent nature of language models poses challenges. The balance between creativity and accuracy remains a goal for developers, but achieving perfect reliability may be an ongoing endeavor.