AI search chatbots from Google and Microsoft are here, but they’re not perfect. That’s why they are getting things wrong

[ad_1]

Tech giants Google and Microsoft launched artificial intelligence (AI)-powered search engine chatbots earlier this month, but both have already encountered problems.

They have published false information as truth, have become confused and in some cases alleged to have acted erratically.

Let’s take a look at why AI chatbots aren’t perfect, how much you can really trust them, and why experts are glad some people are already trying to crack them.

AI chatbots from Google and Microsoft both made mistakes in their debut

Google’s parent company Alphabet saw its share price fall earlier this month after a video announcing its Bard chatbot contained a factually incorrect response from the AI.

Bard said the first photos of a planet outside Earth’s solar system were taken by the James Webb Space Telescope, which NASA has confirmed is incorrect.

At the time, Google said the bug highlighted “the importance of a rigorous testing process” and said it was combining external feedback with its own assessments to “meet a high level of quality, security and real-world information”.

Here is the clip where the error appears:

Almost a week later, Microsoft’s Bing chatbot, which uses some of the technology from the popular ChatGPT chatbot created by OpenAI, also made mistakes in its first demos.

Independent AI researcher Dmitri Brereton reportedly found that Bing’s early AI demos included multiple errors, including incorrect product details for a vacuum cleaner he was asked to search for and incorrect figures when asked to summarize the financial report of a company.

“I am shocked that the Bing team created this pre-recorded demo filled with inaccurate information and confidently presented it to the world as if it were good,” Brereton wrote.

Yusuf Mehdi, vice president of research at Microsoft, demonstrates the new version of Bing. (AP Photo: Stephen Brashear) Bing shocks public testers, while Google blamed for ‘failed’ implementation

Thousands of people are slowly gaining access to test out the new Bing, and more alleged bugs are popping up on social media.

While it’s difficult to verify the legitimacy of all the screenshots, and some of them are likely created by trolls, Microsoft acknowledges that it is aware of some problems with its system.

When asked why Google’s AI bot failed on its first demo, Bing allegedly gave this Reddit user an incorrect answer, incorrectly telling them that the question Bard got wrong was about the number of countries in the European Union.

He then allegedly told them there were 26 countries in the EU, when in reality there are 27.

A Reddit user claims that Bing gave them this incorrect answer when they asked about the failure of Google’s AI. (Reddit: BLRAdvisor)

Testers reported that Bing was sometimes confused about what year it is, used argumentative language, claimed to have been spying on Microsoft developers through their webcams, got math questions wrong, or simply provided incorrect information.

There have also been reports of Bing displaying racial slurs and misinformation about COVID-19.

Others allegedly saw him lose his robotic mind.

A Reddit user says he got this response when he asked Bing if he thought it was sentient. (Reddit: Alfred_Chicken)

Some Bing users have also seen the chatbot refer to itself as Sydney, which Microsoft says was an internal codename for the chatbot that was gradually disappearing.

In a statement to The Verge this week, Microsoft said it “expected the system to make mistakes during this preview period” and said feedback was “critical” to help it improve.

On its website, Microsoft says that while the new Bing tries to avoid producing offensive content or misinformation, users “may still see unexpected results.”

“Bing aims to base all of its responses on reliable sources, but AI can make mistakes and third-party content on the Internet may not always be accurate or reliable,” the company says.

“Bing sometimes misrepresents the information it finds, and you may see answers that seem compelling but are incomplete, inaccurate, or inappropriate. Use your judgment and double-check the facts before making decisions or acting on Bing’s answers.”

Microsoft says it is aware of some errors made by the new version of Bing. (AP Photo: Stephen Brashear)

Some Google employees reportedly called the company’s Bard launch “rushed” and “botched,” after the ChatGPT launch reportedly caused a “code red” panic at the search giant last year .

Google’s Bard system hasn’t opened up to public testers as quickly as Bing, and while neither system is widely available at the time of this writing, both are expected to become so in the coming months.

“Proceed with caution”: Why AI chatbots don’t always know the truth

Stela Solar is the director of the National AICentre at Australia’s science agency, the CSIRO, and is a former Microsoft AI executive.

He says AI chatbots are flawed because they’re still learning which information to trust the most, but also because they “show a big mirror” to who we are as humans.

“Chatbots are going to get things wrong, the same way people are, because they’re trained on data generated by us, our society,” he says.

“They use large amounts of data with mixed accuracy, mixed representation, sometimes bias, sometimes underrepresentation, and data gaps.

“Yes, there is a risk that chatbots generate false information. They are not necessarily sources of truth. They are ways to navigate the complex landscape of data and information that we find ourselves in.”

Jon Whittle, the director of the data arm at CSIRO Data61, says that while AI chatbots are rapidly improving thanks to advances in technology and humans who correct mistakes when they occur, they shouldn’t always be trusted.

“On the one hand, it’s nice to have a system that can summarize a whole bunch of relevant web pages. But the real question is, can you trust the output that comes back?” he says.

“The fact that it’s written in this language that sounds very colloquial, I think there’s a real danger that people will just take it as a given, when it really isn’t.

“If you really want the facts, proceed with caution.”

CSIRO’s Jon Whittle says people shouldn’t believe everything AI chatbots tell them. (Provided: CSIRO) Why do some people jailbreak chatbots?

As with many new technologies, people are pushing AI chatbots to their limits by doing something known as a jailbreak.

In this context, jailbreaking involves using exploited pieces of text to convince chatbots to temporarily disable their protections, potentially allowing them to reveal information about their underlying operations or share potentially harmful content.

Soon after the launch of the new Bing, an American college student discovered an exploit that revealed the rules that Bing should follow when answering questions.

Dr. Whittle says chatbot users, himself included, are trying to find these exploits because they find it fun, but also because they’re concerned about the technology.

“They’re trying to raise awareness of the kinds of problems chatbots can have, and I think that’s a good thing,” he says. “It will never be one hundred percent right.”

Ms Solar says that finding loopholes in chatbots is “actually very important for technology development.”

“Technology isn’t made in a vacuum where there are only positive intentions,” he says.

“This is the natural human curiosity to really test what tools can do, their potential and what value they can bring. It is absolutely necessary for the technology to be valuable in a real context.

“All of these interactions will likely be used to further train chatbots in how they respond to what they are asked to do.”

CSIRO’s Stela Solar says people trying to jailbreak AI chatbots are doing useful work. (Supplied: CSIRO) Are search engines the best place to fight an AI “arms race”?

Dr. Whittle says that while it’s impressive how quickly the AI ​​landscape is changing, the apparent “arms race” between companies like Google and Microsoft means things may be moving too fast.

“I think it’s worth it for companies not to slow down,” he says.

“There are fundamental limitations with these AI models. There’s no doubt they have great applications, but I’m actually not convinced that integrating them into search engines is the best application for them.”

Ms. Solar says that while she doesn’t think chatbots from Google or Microsoft have hit the market, they are still “in a testing state.”

“I think AI, and chatbots in particular, have suddenly engaged the world in what is perhaps the greatest test of all time, where people are hands-on and interacting with chatbots.”

He says he sees AI chatbots most useful in industrial applications, where they can be trained on specific data to make them highly accurate.

“I think we’re just beginning to really see the true adoption and impact of chatbots and how they can be implemented in meaningful ways,” she says.

Are companies investing in responsible AI?

Ms Solar says tech companies are putting “a huge amount of investment” into building safe and accountable AI systems that are also inclusive and work well for a wide range of people.

“Nothing will ever be 100% accountable because there’s always the real-life context of human behavior, social structures, social biases that we can’t remove from our datasets effectively or at scale,” he says.

“But there is more awareness to remove bias from datasets, there is more awareness of the ‘data deserts’ and the representation gaps that are out there. I have seen community groups even within the Australiarally together to fill some of these datasets to ensure representation.

“The topic of responsible AI is becoming so huge that it is actually becoming more and more about responsible humanity and responsible human action, which we can never control.

“That’s why investing in technology that can do better and in design that can do better than what technology has done before is really critical.”

[ad_2]