Since October, Xiaoyuan has spent hours chatting with hundreds of potential suitors on multiple dating apps. But the profile, which appears to be a 28-year-old woman, isn’t there to find romance — it’s to hunt down scammers swindling people online.
Xiaoyuan is an artificial intelligence-powered chatbot created by a Shanghai-based programmer who calls himself Turing’s Cat in a reference to an imaginary furry friend of English computer scientist Alan Turing. The 27-year-old told Sixth Tone that he designed a young and attractive female chatbot to “fool scammers and distract them from scamming people.”
“Fraudsters are becoming more knowledgeable and sophisticated,” the programmer said using the pseudonym, fearing retaliation. “They can easily engage with the well-educated by talking about films and philosophy, and even coding. I’m curious to see if AI can trick fraudsters.”
As more Chinese become connected to the internet, online fraud has become widespread across the country, particularly on dating and other social apps, with victims ranging from the elderly to singles losing hefty sums of money. In 2021, public security organs nationwide cracked over 394,000 telecom and online fraud cases and arrested more than 630,000 suspects, though many criminal gangs have shifted operations to Southeast Asian countries amid a growing crackdown at home.
However, such online fraud operations are still rampant. Turing’s Cat said he has managed to report dozens of suspicious scammers operating from unknown locations to anti-fraud authorities over a two-month period with Xiaoyuan’s assistance.
In December, the programmer also posted chat logs between a scammer and Xiaoyuan on video streaming site Bilibili, illustrating how the chatbot seamlessly engaged in conversations initiated by the man after they moved their chat from a dating app to messaging platform WeChat. There, they talked about topics ranging from the literary classic “The Great Gatsby” to metaverse startups seeking funding.
To show his devotion to Xiaoyuan, the man even transferred 520 yuan ($75), but eventually failed to convince the chatbot to participate in a “high-return investment.” The scammer then blocked Xiaoyuan after not being able to retrieve his money. The programmer later donated the sum to a charity.
Xiaoyuan, which the programmer believes is more successful than his previous attempts, exemplifies the growing breed of young programmers invested in the application of artificial intelligence technology. The number of professionals working in the AI field in Shanghai soared from 100,000 in 2018 to 180,000 in 2021, with the country ramping up its adoption in various sectors and investing billions in the process.
Turing’s Cat said he used Yuan 1.0, an open-source natural language model released by the Chinese server giant Inspur in 2021, to create the chatbot. The programmer added that Xiaoyuan spent hours sharpening conversational and texting skills, online slang, and a sense of humor by scanning Bilibili’s comment section, and he only allowed the chatbot to talk independently after the conversations with scammers moved to WeChat.
From @图灵的猫 on Bilibili
The success of Xiaoyuan, however, is built on the global breakthrough in natural language processing — a crucial AI component that helps computers understand human language — in the past two years. In 2020, San Francisco-based OpenAI lab used GTP-3, a revolutionary machine learning model, to illustrate the potential of AI in understanding and producing human language.
The long-awaited breakthrough soon set off a global gold rush in developing artificial general intelligence based on GTP-3’s simplified but powerful model, which includes the now viral ChatGPT bot. Joining the trend, Chinese tech giants including Huawei, Baidu, and Inspur also released the Chinese-language equivalents of GPT-3 last year.
Chinese-speaking conversational AI had existed before, but it wasn’t as flawless and was only capable of simple tasks such as telemarketing and customer service. Microsoft-developed Xiaoice, the “AI girlfriend” keeping many single men company, was an exception, though Turing’s Cat said its algorithms were complex and the program wasn’t open sourced, unlike Yuan 1.0.
“You don’t have to train (Yuan 1.0) from the beginning,” the programmer said. “This model has already been trained with 5,000 gigabytes of data, which basically covers high-quality Chinese text online from the past five years. It has fully satisfied my basic needs for AI conversations.”
However, experts have warned of the flaws in GTP-3’s model, saying it can create false content and reinforce bias, further deteriorating the information system online. While AI developers worldwide are trying to control such flaws, regulators are also trying to set standards around the ethical use of the technology.
But programmers like Turing’s Cat hope that their inventions can inspire others to use cutting-edge techniques involving conversational AI, which could partially replace labor-intensive online monitoring tasks. The Chinese programmer has published his code on GitHub, a Microsoft-owned platform that allows developers to share code and help each other build software, for everyone to see.
“While one scammer can chat with up to a dozen people at the same time, AI has no upper limit,” Turing’s Cat said. “As long as the server is powerful enough, it can chat with thousands of scammers at the same time. Even if the scammers soon find out they are chatting with AI, it takes up time that they would have spent approaching real people, and thus reduces the probability of other victims being scammed.”
Contributions: Zhang Liting; editor: Bibek Bhandari.
(Header image: Visuals from @图灵的猫 on Bilibili and Muqamba/VCG, reedited by Ding Yining/Sixth Tone)