Welcome to the Year of Artificial Belligerence
- By Stefan Hammond
- February 20, 2023
Is ChatGPT the media's newest celebrity? The volume and diversity of media stories about ChatGPT (“a chatbot developed by OpenAI and launched in November 2022”) rapidly approach critical mass.
In early 2023, media stories treat chatbots like personalities. AI-powered services draw comparisons to human behavior — errant and otherwise.
And as other firms ramp up their chatbots in the AI media-attention sweepstakes, accusations and accolades fly thick and fast.
Passing grades, passing remarks
As AIs become more accessible, humans throw traditional intelligence tests at them. “ChatGPT is smart enough to pass prestigious graduate-level exams — though not with particularly high marks” is the title of a CNN article by Samantha Murphy Kelly.
According to Kelly, ChatGPT “recently passed law exams in four courses at the University of Minnesota and another exam at the University of Pennsylvania’s Wharton School of Business.” The AI “performed on average at the level of a C+ student, achieving a low but passing grade” on a battery of multiple-choice and essay questions, but also made mistakes on “basic math.”
The tendency to bloviate erroneously seems a pan-AI phenomenon
Basic math errors seem to crop up when AI-generated content is loosed upon readers. “CNET is now letting an [unspecified] AI write articles for its site,” said an article on futurism.com. “The problem? It's kind of a moron.”
CNET published AI-generated articles that made “a series of boneheaded errors” in basic financial advice. These errors seem all the more egregious as the AI tended to adopt a stance best defined as confidently incorrect.
“All three screwups, each of which the AI presented with the easy authority of an actual subject matter expert, highlight a core issue with current-generation AI text generators: while they're legitimately impressive at spitting out glib, true-sounding prose, they have a notoriously difficult time distinguishing fact from fiction,” said futurism.com.
Errors and neurosis
A tendency to bloviate erroneously seems a pan-AI phenomenon. “Google's search chief has warned against relying on AI chatbots to always produce accurate information,” said Business Insider. “Prabhakar Raghavan told Welt Am Sonntag on Saturday that they can sometimes give false but convincing answers.”
“'This kind of artificial intelligence that we are currently talking about can sometimes lead to something we call hallucination,' he told the German newspaper. 'He added: "This is then expressed in such a way that a machine provides a convincing but completely fictitious answer'."
Not only do AI-powered chatbots demonstrate a propensity to fib, but they stand accused of human levels of neurosis.
Off the deep end?
“ChatGPT Bing is becoming an unhinged AI nightmare,” reads a headline from Digital Trends. Another headline: “Microsoft’s ChatGPT-powered Bing is getting ‘unhinged’ and argumentative, some users say: It ‘feels sad and scared’”.
“In racing the breakthrough AI technology to consumers last week ahead of rival search giant Google, Microsoft acknowledged the new product would get some facts wrong,” said a story on the AP. “But it wasn’t expected to be so belligerent.”
Accusations and accolades fly thick and fast
“The Bing chatbot, positioned as Microsoft's answer to Google search dominance, has shown itself to be fallible...it's exhibiting all emotions including angst,” said USA Today. “Microsoft's newly AI-powered search engine says it feels 'violated and exposed' after a Stanford University student tricked it into revealing its secrets,” said an article from the CBC.
Such stories make great clickbait but also reflect humans gaming the system. “The responses shouldn’t come up in normal use,” noted Digital Trends. “They likely result in users 'jailbreaking' the AI by supplying it with specific prompts in an attempt to bypass the rules it has in place.”
Pinging Sydney
Ars Technica expands on the theme in a February 2023 article: “On Tuesday, Microsoft revealed a "New Bing" search engine and conversational bot powered by ChatGPT-like technology from OpenAI. On Wednesday, a Stanford University student named Kevin Liu used a prompt injection attack to discover Bing Chat's initial prompt, which is a list of statements that governs how it interacts with people who use the service.”
Liu, “an artificial intelligence safety enthusiast and tech entrepreneur in Palo Alto,” according to the CBC, “used a series of typed commands...to fool the Bing chatbot into thinking it was interacting with one of its programmers.” The machine "also blurted out a code name: Sydney.”
“The chatbot often responds to questions about its origins by saying, 'I am Sydney, a generative AI chatbot that powers Bing chat',” said an MSN article. “It also has a secret set of rules that users have managed to find through prompt exploits (instructions that convince the system to temporarily drop its usual safeguards).”
Leave it to a Stanford grad student to jailbreak an AI.
AI free-for-all
It's the flavor-of-2023. “Alibaba says it will launch its own ChatGPT-style tool, becoming the latest tech giant to jump on the chatbot bandwagon,” wrote Michelle Toh on CNN. “The Chinese behemoth said it was testing an artificial intelligence-powered chatbot internally.”
“Google and Chinese search engine giant Baidu both unveiled plans to launch similar services of their own,” wrote Toh. “Google’s tool, named 'Bard,' will roll out to the public in the coming weeks, while Baidu’s bot, called “Wenxin Yiyan” in Chinese or 'ERNIE Bot' in English, will launch in March.
“E.U. lawmakers hope to agree on draft artificial intelligence rules next month, with the aim of clinching a deal with E.U. countries by the end of the year, one of the legislators steering the AI Act said',” said a report on Reuters. “The European Commission proposed the AI rules in 2021 in an attempt to foster innovation and set a global standard for a technology, used in everything from self-driving cars and chatbots to automated factories, currently led by China and the United States.”
ChatGPT, Bard and ERNIE Bot seem poised to battle for supremacy throughout the quarters and years to come. Whatever claims of sentience anyone may make, their most significant potential may well be as search engines.
Stefan Hammond is a contributing editor to CDOTrends. Best practices, the IoT, payment gateways, robotics, and the ongoing battle against cyberpirates pique his interest. You can reach him at [email protected].
Image credit: iStockphoto/FotografieLink