This race to harness the power of AI algorithms and create the next generation of chatbots has the potential to completely change the way we search the web.
Bing, Google, and Baidu are all competing to develop their own versions of OpenAI’s ChatGPT, a chatbot that has taken the world by storm with its ability to answer questions in a surprisingly coherent and insightful manner. With the help of these chatbots, searching the web may become more of a conversation rather than a process of clicking links and exploring sites. The underlying language technology could also have the potential to transform other tasks such as email programs that write sales pitches or spreadsheets that dig up and summarize data for the user.
Microsoft has announced that it is rewiring Bing to use ChatGPT technology, which is already proving to be a hit among users. During demonstrations at Microsoft headquarters, the company showed how the new Bing can generate a vacation itinerary, summarize product reviews, and answer tricky questions such as whether an item of furniture will fit in a particular car. This is a far cry from Microsoft’s previous Office assistant Clippy, which was hapless and often frustrating for users.
Google is also joining the competition with the release of Bard, its own version of ChatGPT. The company showed how the AI behind Bard can answer web searches and plans to make the AI available to developers. It’s no secret that Google is unsettled by the potential for being upstaged in the search industry, which provides the majority of Alphabet’s revenue. And with AI researchers having developed the machine learning algorithm at the heart of ChatGPT, it’s understandable that Google would want to get in on the action.
Baidu, China’s largest search company, has also entered the race with its own ChatGPT competitor, Wenxin Yiyan, or “Ernie Bot” in English. Baidu plans to release the bot after completing internal testing in March.
These new search bots are examples of generative AI, a trend fueled by algorithms that can generate text, craft computer code, and dream up images in response to a prompt. Despite the tech industry experiencing widespread layoffs, interest in generative AI is booming, with VCs imagining entire industries being built around this new creative streak in AI.
While the text generated by these chatbots may look human, the AI models behind them do not work like a human brain. Their algorithms are designed to learn to predict what should come after a prompt by feeding on statistical patterns in huge amounts of text from the web and books. They have no understanding of what they are saying or whether an answer might be incorrect, biased, or representative of the real world. These AI tools generate text purely based on patterns they’ve previously seen, making them prone to “hallucinating” information. ChatGPT’s power also comes from a technique involving human feedback on questions, but this feedback only optimizes for answers that seem convincing, not for those that are accurate or true.
These issues may be a problem if we’re trying to use the technology to make web search more useful. Microsoft has apparently fixed some common flaws with ChatGPT in Bing, but the real test will come when it’s made widely available. Google has already shown that Bard can produce incorrect answers, such as claiming that the James Webb Space Telescope was the first to take a picture of a planet beyond our solar system.