Can We Trust Generative AI for Search?
The rise of generative AI has led to speculation about how it will change the landscape of knowledge, research, and content creation. One of the areas that stands to be most affected by generative AI is search, with the potential for AI tools like ChatGPT to deliver answers instead of just websites. However, there are several practical, legal and technical challenges that need to be addressed before generative AI can reach the robustness, scale, and reliability of established search engines like Google.
One of the major challenges facing generative AI is the issue of real-time information. In its current form, ChatGPT doesn’t have access to real-time information in the way that web-crawling search engines do. ChatGPT was trained on a massive dataset with an October 2021 cut-off, which means it doesn’t “know” anything beyond that date. This makes it difficult to rely on ChatGPT for anything important right now.
Another challenge is the issue of continuously retraining an LLM as information on the internet evolves. The tremendous amount of processing power and financial cost associated with these resources make it difficult to train an LLM in real-time, particularly if the aim is to process queries at the rate Google does.
Even if these challenges are overcome, there is still the issue of the actual information that AI tools like ChatGPT will deliver. The primary concern lies in the accuracy and validity of the information that such AI systems will provide. Language models like ChatGPT are akin to reflective surfaces that mirror the societal trends and patterns they observe. When trained on unfiltered data from the internet, such models could potentially produce harmful and offensive content. Even with carefully curated training data sets, it is not guaranteed that all the information in vast online datasets is free from bias and factual inaccuracies. The resulting AI-generated content will reflect the beliefs and perspectives of the dominant group within the training data.
These challenges make it difficult to rely solely on generative AI for search. While it has the potential to deliver deeper insights and understanding, there are still significant obstacles to overcome before it can be used as a reliable source of information.