A head-to-head comparison of Bing AI and Google Chatbot

Ask Microsoft Corp. or Google's artificial intelligence what it would do if a dog got separated from its owner for some instance that illustrates just how far chatbots still need to go before they can be relied upon to answer even basic questions.

‍

Microsoft's Bing offered some sound advice, but on a completely different topic than what we were discussing. As it turned out, the program assumed that we were asking about separation anxiety and gave us a detailed response, including possible signs a dog is suffering from the condition, common causes, and some suggestions on how to comfort a lonely pet.

‍

During the same time period, Google suggested that if a dog were separated from its owner, it would remember the owner's telephone number and contact them when it was separated. Is it possible that the chatbot has been trained on too many viewings of the movie ET in order to do its job?

‍

There was an exciting potential revealed with this month's introduction of Bing with its new ChatGPT-like feature, but it also revealed some deep flaws that need to be addressed. There is no doubt that companies are aware of the imperfections of their machines. In order to limit the type of conversations that the chatbot would engage in, Microsoft has taken steps to limit these types of conversations and has deliberately taken a slow approach to welcome people off the waitlist. There is no public release date for Google's Bard bot, which hasn't even been released to the public yet.

‍

Bloomberg, however, was granted access by Google to a tool called LaMDA Test Kitchen, which serves as a demonstration tool for Bard's AI, which is based on the same technology that's used in Google's artificial intelligence. It won't be able to settle the question of whether Bing's or Google's chatbot is better than the other. As far as the demo of Google's service goes, it only does three things: it generates lists, it "imagines" scenarios, and it has conversations about dogs.

‍

In Google's Test Kitchen, the list function starts every query with "I want to do this." We filled in 'learn more about the new Bing search', but it wasn't taking it. It generated a list that included items such as "search for a new recipe," "search for a new hobby," and "try a new search engine."

‍

In response to a question about its chief rival, Bing gave a coherent, accurate, and objective answer, though parts of it appeared to be lifted straight from Google's corporate blog, bordering on plagiarism.

‍

One of the features of the Google demo is that users can enter a phrase that begins with the phrase “Imagine I am at...” (for instance, I typed “my computer chatting with the new AI-powered chatbot Bard.”). Among Google’s suggestions was to use Bard to find a therapist and to get advice based on its “decision-making algorithms” — which we would not rely solely on a chatbot for and certainly not in the state in which they are currently.

‍

In terms of freewheeling conversations with the user, there is only one scenario in which the Test Kitchen engages with the user in a manner similar to what Bing does. During this scenario, the chatbot pretends it's a tennis ball that is extremely fond of talking about dogs. Unless you're considering the hypothetical example of the exceptionally intelligent lost dog as an example, it's hard to make much sense of it.

‍

Microsoft and Google have both admitted that they need to do a lot more work if they want their AIs to be more reliable. As opposed to Microsoft, Google at least has the luxury of being able to do this in a largely unnoticed manner, whereas Microsoft is forced to soak up all of the attention and criticism. There is no way for us to be sure which strategy will prove to be the most successful in the end, and neither did the chatbots seem to know either.