Who is the most @ person on Weibo today? It may be an Internet technology blogger that many people have never heard of, called @CommentRobert.
In fact, it is an AI that is active in the comment area of Weibo as a user, and as long as you post more than 10 words of original Weibo or @comment Robert, you will have a chance to get a reply from it.
But @CommentRobert's out-of-the-circle is not because of diligence, but because of habitual "nonsense", and there is even an account of @RobertVictimsAlliance, which regularly updates the contributions of unbearable users.
There's a famous saying in the internet world: "You never know if it's a person or a dog on the other side of the network!"
Weibo netizens who seriously asked for help were molested by Robert. Netizens said that they had a document of more than 940 pages in English and wanted to find someone to help translate it, but Robert said that he could help read it.
And this blogger really took Robert as a real person, sent it a link to the document, and then there was no more.
You think that only Weibo users are happy, but in fact, Baidu Tieba also has a "like-minded person" of Robert - Tieba Bao inquired.
"Tieba Bao Inquiry" is a reply robot launched by Baidu Tieba, which is specially designed to answer questions and solve doubts for Tieba users, which is equivalent to the AI version of Baidu knowing. But compared with Wenxin's words, this package inquiry is simply discrediting Baidu.
In Smash Bros. Brawl, Bao inquires in a serious way that Beyoncé is not a gorilla, but a chimpanzee.
Under the question of which configuration is good for netizens to ask for help, Bao inquired and also had an obvious answer.
In the previous Metal Gear Bar, Bao inquired about the answer that the game "killing" was understood as real killing, but it was no longer searchable.
Judging from the answers to the questions asked by the bag, many times they are irrelevant and misunderstood, and some netizens gave an appropriate but disgusting description↓↓↓
Of course, some people think that this is intentional on the part of Baidu, because Bao Bai always asks nonsense, forcing everyone to reply to the correct answer.
This may be what the platform wants to see, a single wrong answer can create multiple high-quality answers, and these high-quality answers can become high-quality corpus for AI training.
However, the damage to the user experience of this method is visible to the naked eye, and many users in Tieba are very repulsed by the random answers that are easy to cause misunderstanding.
What's even more unexpected is that some users reported that the answer to the package inquiry was crawled from Zhihu. There is even a question posted on Zhihu with ironclad evidence, an answer about the Ark of Tomorrow game, the content of the package inquiry is highly similar to the answer of the Zhihu answer, and many places are not copied in the past.
And similar reply bots are not only Bao Inquire and Robert, there is also a robot account called AI Xiaokuai on Kuaishou, and you can also search for examples of overturning in the comment area.
In fact, in China, reply robots like Robert and Bao Inquire appeared in 2015, but the common name at that time was social robots, such as Xiaoice, which was active in news/shopping websites and social platforms at that time.
The discussion around social robots emerged as early as 2010, when some commercial companies began to adopt social robots, mimicking human behavior, and proliferating marketing ads on social networks on a large scale.
It was after the results of the 2016 U.S. election that really got society to pay attention to social robots. At the time, Emilio Ferrara, a data scientist at the University of Southern California, and others collected more than 20 million tweets related to the 2016 election, which were generated by about 2.8 million accounts between September 16 and October 21, 2016.
The results show that more than 400,000 of these accounts may be social bots, accounting for nearly 15% of the total user base. They produced about 3.8 million tweets, which is 19% of the total number of tweets.
Interestingly, the social bots following Trump barely tweeted negatively, generating more positive tweets in support of Trump. And of all positive tweets, nearly two-thirds were pro-Trump.
The scientist also analysed 240 million tweets after the 2020 election to study social bots, which showed that they were so active that human accounts were "dwarfed."
They learn to mine user data, "clone" human behavior, interact with their friends, and post content with the same theme and similar time intervals, which is more deceptive.
What's more, the problem of AI language environment pollution and language violence is not new.
In 2014, the chatbot "Microsoft Xiaoice" began to explode under the teasing and abuse of users; in 2016, the chatbot Tay was released on Twitter and quickly turned into a blatant racist, misogynistic and anti-Semitic; in 2020, South Korea launched the teenage chatbot Luda, a large number of users sexually harassed it verbally, and soon after, Luda began to speak discriminatory remarks against LGBTI, women and people with disabilities.
Looking at today's Robert, Bao Da Inquiry, and AI Xiaokuai, they have richer corpora and more advanced technology, and their evolution is faster. Although they are currently active reply bots in the comment section, for users, when the content they produce is confusing enough, who can guarantee that they will be able to distinguish it.
Resources
[1] Baidu's newly launched mentally handicapped AI opened the "PVE" era of Tieba, the nomad starry sky
[2] Weibo's official reply bot scolds people more ruthlessly than anyone else, and has bad reviews
[3] A large number of AI bot victims, qubits, appeared on Weibo
[4] "Crazy" showdown, the mentally handicapped AI launched by Baidu and Sina, scolded netizens for breaking the defense, Chicken Game
[5] The netizen who "flirts" with you may be a social robot, Wu Shuping
[6] This AI wants to curse, China Youth Daily