Google's latest AI Overview feature has stirred controversy across social media platforms, with users reporting instances of peculiar recommendations ranging from putting glue on pizza to consuming rocks.
The erratic behaviour of the AI has prompted Google to swiftly intervene, manually disabling AI Overviews for specific searches as memes flood social networks.
Despite being in testing for a year and officially launching in beta as the Search Generative Experience in May 2023, the rollout of AI Overview has been marred by glitches and unexpected outcomes.
CEO Sundar Pichai previously boasted of serving over a billion queries during the testing phase, highlighting the widespread adoption of the feature.
However, Pichai's acknowledgment of Google's success in reducing the cost of delivering AI answers by 80 per cent over the same period raises questions about the premature optimisation of the technology -- It appears that kind of optimisation might have happened too early, before the tech was ready.
"A company once known for being at the cutting edge and shipping high-quality stuff is now known for low-quality output that’s getting meme’d," one AI founder, who wished to remain anonymous, told The Verge.
Google spokesperson Meghann Farnsworth said in an email to The Verge "Many of the examples we’ve seen have been uncommon queries, and we’ve also seen examples that were doctored or that we couldn’t reproduce."
Farnsworth also said that the company is taking steps to remove AI Overviews from specific searches that violate content policies.
According to Gary Marcus, an AI expert and emeritus professor at New York University, many AI companies promise perfect results, going from 80 per cent accuracy to 100 per cent. However, achieving the initial 80 percent accuracy is relatively easy because it involves analysing a large amount of human data, he said.
Marcus said that achieving the remaining 20 per cent accuracy is incredibly challenging and might be the most difficult task of all.
"You actually need to do some reasoning to decide: is this thing plausible? Is this source legitimate? You have to do things like a human fact checker might do, that actually might require artificial general intelligence," Marcus said.
"[These models] are constitutionally incapable of doing sanity checking on their own work, and that’s what’s come to bite this industry in the behind," Marcus added.
Google has big ideas for AI Overviews. What we see now is only a small part of what they revealed last week. The company want to add features like solving complex problems step by step, organising search results with AI, and searching for videos using Google Lens.
Image Source: Unsplash
Comments