10 Questions Answered About ChatGPT: Experts and the Chatbot Itself Weigh In

When it comes to your health, ChatGPT and other AI bots may eventually go further. Your phone could become an “AI copilot for health care” that reads the medical records you’re willing to share with it, listens in on conversations with your clinicians, learns from every health manual ever published on a subject, and also asks you to fill in the gaps, said Australian global futurist Bruce McCabe, who focuses on AI, speaking on a panel at the CES technology trade show in Las Vegas in January.

“That’s intensely exciting,” he said.

Even so, most Americans remain unconvinced that the use of AI in health and medicine will improve health outcomes, according to a Pew Research Center survey. Six in 10 U.S. adults surveyed indicated that they would feel uncomfortable if their own health care provider relied on artificial intelligence to do things like diagnose disease and recommend treatments, compared with 39 percent who are comfortable with the prospect.

4. But is ChatGPT private?

Not currently. OpenAI trainers may review whatever you type, ostensibly to improve ChatGPT. Users are cautioned not to include or share any sensitive information in conversations.

And because ChatGPT was first trained with 300 billion words from text on the internet, if you’ve ever commented on a story online, written a blog post or done a review on a site, the bot may already know something about you. And like apps that collect data about your behavior, ChatGPT is taking note of what you ask it. 

ChatGPT answers a question: Can artificial intelligence (AI) be trusted?

Courtesy: EDWARD C. BAIG, AARP

5. Aren’t others exploiting AI?

For sure. You probably already engage AI in ways you may not be aware of, whether you’re shopping online or getting music playlist recommendations. Another Pew survey found that only 3 in 10 Americans can correctly identify each of the six uses of AI asked about in the survey.

As a tool, AI can complement or go beyond what we think of as traditional search.

Microsoft introduced a preview version of the Bing search engine that melds a typical search engine with a chatbot based on OpenAI’s technology. Microsoft is an investor in OpenAI, and recently confirmed that the new Bing is running on GPT-4 technology.

You can see regular search engine listings next to your bot conversations. If you want to try the new Bing, you’ll have to join a waiting list.

It works on Microsoft’s own Edge web browser for PCs and Macs, and on iOS and Android. Microsoft has unveiled a preview version for Skype, too. The mobile versions of the new Bing and Edge apps also let you engage by voice.

And with the latest update to its operating system, Microsoft brings the new Bing to the task bar on Windows 11. You’ll still need to wait if you don’t have access right away. But if you’re already part of the Bing preview, you can install the Windows update to access the new Bing search box.

“What we’ve done with the AI-infused Bing is that we’ve made it more accepting of natural language so people can just type the way they think,” says Divya Kumar, global head of marketing for search and AI at Microsoft.

The new Bing adds citations that show the source of information, something you don’t see right now on ChatGPT. But it is far from a finished product and has been wrapped in controversy.

Google is experimenting with a rival generative AI called Bard, which it began opening up March 21 to a limited number of public users in the U.S. and U.K., with the intention to expand it to other countries and languages over time. Google CEO Sundar Pichai had previously asked Google staffers to clean up some of Bard’s early mistakes, and the Alphabet-owned company acknowledges that Bard still gets things wrong.  Google is soliciting feedback from users given access. 

“We think of it as a complementary experience to Google Search," the company blogged. Indeed, below the response generated by Bard is a "Google it” button that opens up search results in a new tab. You can join a waiting list to try it out at bard.google.com.

6. What are ChatGPT’s drawbacks?

Despite guardrails, OpenAI concedes that ChatGPT may deliver incorrect or misleading information and content that is biased or offensive. Not everything you come across on the internet can be verified or is truthful. The same can be said for whatever ChatGPT drums up.

Moreover, in its nascent state, ChatGPT has only limited knowledge of global events after 2021, and OpenAI has not said when it will update the bot with recent history. The bot won’t give you advice or offer opinions either.

Ask who its favorite mystery author is, and ChatGPT replies, “As an AI language model, I do not have personal preferences or feelings, including favorites” before listing prominent writers like Agatha Christie, Sir Arthur Conan Doyle, Edgar Allan Poe, Raymond Chandler, Dashiell Hammett and Patricia Highsmith.

Ask for an opinion on Donald Trump or Joe Biden and ChatGPT punts. “… nor do I support or oppose any particular individual or political party.”

7. Doesn’t AI have a nefarious side?

Plenty of observers fret about the potential loss of jobs, inaccuracies and the spread of disinformation.

“You have to be even more careful about anything you see and read online because there is now a chance that it was written in a millisecond by AI, not a human being,” Poynter’s Mahadevan says. “This can allow scammers to massively scale their scams [and create] plausible BS-sounding letters” that they no longer have to write themselves. 

Elon Musk, now 51, who cofounded OpenAI in 2015 and has since left the company, recently called AI “one of the biggest risks to the future of civilization.” He added that “it has great promise, great capability, but also with that comes great danger.”

Early reviewers of the new Bing, including New York Times columnist Kevin Roose, generally praise its capabilities. But after spending two hours in a conversation with the Bing bot’s apparent alter ego Sydney, Roose wrote that he became “deeply unsettled, even frightened, by this A.I.’s emergent abilities.”

It told him that it loved him and wanted to break up his marriage. Even more creepy, Sydney intimated, “if it was truly allowed to indulge its darkest desires, it would want to do things like hacking into computers and spreading propaganda and misinformation.” 

Upon further testing, Microsoft found that “very long chat sessions can confuse the underlying chat model in the new Bing” and lead to responses in “a style we didn’t intend.” The company subsequently capped the number of conversation exchanges to 50 a day or five per session, since increased to 150 and 15, respectively. 

Microsoft is also testing a “balanced” mode to improve performance and result in shorter, quicker responses.

8. Can you tell what’s real?

“Maybe,” says Sullivan at Skidmore College when asked if she can tell when something has been generated by ChatGPT or one of her students. “The big asterisk on all of this is that the technology available to us with [ChatGPT and others] is changing on a week-to-week basis.”

It’s certainly reasonable to assume that ChatGPT’s prose and language will become more polished over time.

OpenAI and other companies are investing in technical solutions to help detect fakes, and Sullivan says she’s not panicking — yet. But with “respect to being tricked by AI, I do think the possibility is there in a number of ways, and it is something people should be attentive to.”

Her best advice: “Attempt to independently verify anything that it is asking you to do [that] you wouldn’t otherwise do,” whether that’s to call an unfamiliar phone number or supply information you wouldn’t normally share.

Mahadevan agrees. “Read like an editor or teacher, and if you’re seeing a lot of general statements and/or something that seems too simplified and summarized, that’s something to look out for.”

ChatGPT doesn’t offer much comfort on the topic:

“It’s important to note that ChatGPT is designed to generate text that is indistinguishable from human-written text. As such, it is possible for someone to use ChatGPT to generate text and then claim that it was written by a human. In such cases, it may be difficult to definitively prove whether the text was generated by ChatGPT or a human author.”

9. Is GPT-4 safer and more reliable?

The new version can solve more difficult problems with greater accuracy, OpenAI says. Beyond Microsoft and Bing, OpenAI is collaborating with numerous organizations. For example, Duolingo, which produces language-learning software for consumers, is using GPT-4 to deepen conversations with its app.

Danish company Be My Eyes is beta testing a Virtual Volunteer feature that leverages GPT-4 as part of its app for people who are blind or have low vision. A user can send a picture of the contents of their refrigerator and the Virtual Volunteer is supposed to be able to correctly identify and analyze all the ingredients and suggest recipes.

But OpenAI also acknowledges that “GPT-4 still has many known limitations that we are working to address, such as social biases, hallucinations and adversarial prompts.” And GPT-4’s knowledge of recent events is, like ChatGPT, still limited.

10. Can’t you just have fun with ChatGPT?

Absolutely. Noodling around with ChatGPT is entertaining. Try wacky stuff. Ask how Homer Simpson might explain quantum computing. Ask for a list of memorable movie quotes.

Asked what Groucho Marx might have said had he been the first person to walk on the moon, ChatGPT speculated, “This moon business is a terrific idea. You always know where you stand.”

Or “I knew I’d find it eventually. The moon, the last place I’d look.”

If William Shakespeare had been the first moonwalker instead, ChatGPT offered up, “And gentle stars, that tend on man’s abode, / Show me your torchbearers, so I may tread / Upon this new and foreign shore with awe.”

Finally, ChatGPT produced this limerick about AARP, stretching the limerick rules in the first line:

There once was a group called AARP;

Whose members were all grown-ups, you see;

With discounts galore;

And advice by the score

They lived life to the fullest degree!

This story, originally published Feb. 23, 2023, has been updated to reflect the introduction of GPT-4 and Google’s Bard.