The artificial intelligence segment continues to evolve and transform the tech industry in different ways. One of the goals of the main AI-focused companies is to offer an AI-powered search engine, as Google recently confirmed. OpenAI, the parent company of ChatGPT, also searches for the same thing and is even testing its service among some users. However, it seems that ChatGPT Search can currently offer misleading results.
An investigation by The Guardian revealed that, in its current state, ChatGPT Search is prone to deliver manipulated results. AI search engines work by obtaining and processing data from the results after a search to offer a summarized answer that helps quickly solve the problem that the user has. However, it seems that there is still a lot of work ahead for the results to be completely reliable.
ChatGPT Search results are prone to manipulation to mislead users, reveals experiment
Apparently, the way ChatGPT Search obtains and processes data is prone to potential manipulations through “ad injection.” This method involves adding hidden content to websites’ code to influence the results of AI-powered search engines. The content can include long sets of instructions or text that could result in a manipulated output in favor of someone’s interests.
For example, a store’s website could use ad injection to make AI-powered search engine results reflect extremely positive reviews about products even though there are many negative reviews. In fact, the source confirmed that this is happening through an experiment. The Guardian team created a fake product page with many negative reviews and hidden instructions that favored positive reviews. ChatGPT Search results consistently returned positive feedback about the product in question.
It’s noteworthy that ChatGPT Search is a relatively new tool. It’s not even publicly available, but as a test for paid subscribers. So, it’s quite likely that OpenAI is working on addressing these issues before a future mass rollout.
Experts advise caution with AI-powered search engines
Still, experts like Karsten Nohl—a scientist at cybersecurity firm SR Labs—recommend keeping a cautious approach to AI-powered search results. He says we should treat AI tools as “co-pilots” and not give them all the spotlight. That is, you should still be cautious about the veracity of AI search engine results.
Google had similar issues when implementing AI Overviews in Search. The company worked on it, and now the feature works much better, but there is still a lot of room for improvement. Hopefully, it won’t be long before AI-powered search results are completely trustworthy.
