Large technology providers are launching new artificial intelligence products at an astounding rate, even though their offerings often break down as soon as they hit the market. A study by researchers at Stanford and the University of California, Berkeley found that ChatGPT has become less accurate over time, for example. Earlier, two New York attorneys found themselves in hot water after they cited several cases ChatGPT had invented.
More recently, Google Bard was found to make up emails it was supposedly summarizing, while Amazon Alexa recommended users visit museums in the wrong part of the country.
One reason for this, technology executives say, is FOMO, or “the fear of missing out.” According to the Washington Post, tech firms are afraid they’ll miss opportunities to develop an early audience. They also need more people to use their products in order to generate the data that can improve their performance.
Vendors say they’re not hiding the fact that their AI products are still new, and far from bullet-proof. But, notes CIO, some companies exaggerate their capabilities when in fact their products use little, if any, AI at all.
According to Gartner, generative AI is at or near its peak of inflated expectations. (From there, it will slide into the “trough of disillusionment.”)
Such a hype dynamic isn’t new. Not so long ago, HR technology vendors incorporated the idea of “people data” into their messaging for almost any kind of product, and “AI” was used to describe products that bore little resemblance to today’s generative AI capabilities.
Now, the idea of “AI powered” is attached to everything from time clocks to performance reviews. Often, vendors’ messages focus on the technology’s ability to streamline processes so employees can devote more time to “strategic work.” Exactly what that means, and how it can benefit customers, is rarely defined.
Hiccups along the way aren’t impeding the momentum. In February, Microsoft’s AI-based Bing chatbot began referring to itself by an alter ego’s name and told users it could think and feel. Microsoft dialed down the chatbot’s creativity and moved on, the Post said. Now, seven months later, Microsoft plans to feature AI “copilots” to help users with Word and Excel, as well as troubleshooting chores and summarizing long web articles.
Down in Flames
The danger in all of this, analysts say, is that users may become skeptical about AI’s potential if they run into too many instances where it doesn’t work. “It may backfire,” Gartner Vice President Jim Hare told the Post.
Chatbots are built to sound authoritative, which can lull users into accepting inaccurate information at face value. And, said Mozilla Chief Product Officer Steve Teixeira, developers aren’t being transparent enough when discussing how they take advantage of user-generated data.
None of this is slowing the companies down. During its launch, Amazon’s conversational chatbot for its Alexa home speakers fell into long pauses during what was supposed to be a “near human” conversation. Said one executive: “It’s not the endgame.” That’s not stopping Amazon from making the chatbot available to anyone who uses Amazon speakers.