Microsoft argues that AI's tendency to provide inaccurate responses is justifiable by the fact that they are 'usefully wrong'

Recent developments in artificial intelligence have made it possible for new tools like ChatGPT to wow users by producing gripping text in response to questions and prompts.

‍

These AI-powered tools have become considerably more adept at providing original, occasionally amusing responses, yet they frequently contain false information.

‍

For instance, when Microsoft unveiled its Bing conversation tool in February, which was developed using the GPT-4 technology developed by Microsoft-backed OpenAI, users discovered that the program was giving inaccurate replies during a demo including financial earnings reports. The Bing conversation feature occasionally presents false information that users might take to be the real deal, a situation known as a "hallucination," like other AI language tools and related Google software.

‍

These factual issues haven't slowed down the AI competition between the two tech behemoths.

‍

Google revealed on Tuesday that Gmail and Google Documents would soon have chat technology driven by artificial intelligence (AI) to assist users in writing emails and papers. Microsoft announced on Thursday that the ChatGPT-similar Copilot technology would soon be included with its well-known business apps like Word and Excel.

‍

Microsoft, though, is promoting the technology this time as "usefully wrong."

‍

Microsoft executives mentioned the software's propensity to produce erroneous responses in an online presentation about the new Copilot features, but they presented that as something that may be valuable. As long as individuals are aware that Copilot's responses could be factually incorrect, they can amend the errors and send emails or complete presentations more rapidly.

‍

Copilot can still be useful even if it displays the incorrect birth date if someone wishes to compose an email wishing a family member a happy birthday. According to Microsoft, the tool's simple ability to generate text saved someone time, making it helpful. Simply put, greater care must be taken to ensure that the text is error-free.

‍

Researchers might disagree.

‍

In fact, some tech experts, including Noah Giansiracusa and Gary Marcus, have expressed concerns that people might put too much trust in contemporary AI by listening to the advice that ChatGPT and other similar tools provide when they ask questions about health, finances, and other important subjects.

‍

"ChatGPT's toxicity guardrails are easily avoided by those determined on using it for evil and as we witnessed earlier this week, all the new search engines continue to hallucinate," the two wrote in a recent opinion piece for Time. “The true test will be if any of the major players can create artificial intelligence that we can actually trust once we get over the opening day jitters.”

‍

It's uncertain how dependable Copilot will be in use.

‍

According to Microsoft's chief scientist and technical fellow Jaime Teevan, the company has "mitigations in place" when Copilot "gets things wrong, has biases, or is misused," and it will first test the software with only 20 corporate clients to learn how it performs in actual use.

‍

“We're going to make mistakes, but we'll fix them right away, Teevan assured.

‍

Microsoft cannot afford to ignore the excitement surrounding generative AI technologies like ChatGPT because the financial stakes are too great. The company's problem will be to implement that technology without endangering the software's reputation or causing huge PR catastrophes.

‍

“With this potent new instrument, I feel a tremendous sense of duty because I've researched AI for decades,” said Teevan. “We must deliver it to them and do so ethically, according to this responsibility.”