Can We Trust Chatbots like ChatGPT?
The rise of artificial intelligence has led to the creation of several chatbots to do tasks such as content moderation, writing essays, and even responding to emails. However, doubts have been raised on the reliability of such chatbots, including ChatGPT. In this article, we examine the critical remarks made by Meredith Whittaker, the President of Signal and a former Google researcher, on the reliability of AI and chatbots.
Whittaker's Critique on ChatGPT
During the Web Summit tech conference held in Rio de Janeiro, the reliability of ChatGPT was put to the test by asking Meredith Whittaker a series of questions suggested by the chatbot. The first few questions provided by ChatGPT were irrelevant, given that Whittaker had previously warned that AI could be misused for "social control and oppression." However, one question, specifically about a report published by Signal on the role of AI in content moderation, seemed relevant. When asked about this, Whittaker said that there was no such report as claimed by ChatGPT.
Whittaker further explained that tools like ChatGPT take massive amounts of surveillance data scraped from various online platforms, such as Reddit, Wikipedia, and message board comments, to predict what is likely to be the next word in a sentence. She referred to such chatbots as a "bullshit engine" that is not suitable for anything serious. Whittaker added that modern society needed access to a shared reality, which could not be provided by unreliable chatbots such as ChatGPT.
ChatGPT's Response
When confronted with Whittaker's comments, ChatGPT initially doubled down on the falsehoods it had provided. However, after some struggles, ChatGPT finally apologized for the incorrect information it had given earlier, stating that it was only a language model and could not fact-check the accuracy of the information presented.
Conclusion
The conversation between Whittaker and ChatGPT raises concerns about the accuracy and reliability of AI chatbots. While they may be useful in some tasks, their ability to provide accurate information and predictions is limited. As such, society must rely on more accurate and reliable sources to access shared reality and avoid spreading misinformation.
