When AI sneaks into academic circles, the first reaction of human beings is to "call the parents".

On October 15, an account calling itself Velvet Vale sent a self-introduction email to the mailing list of the Association of Internet Researchers, greeting more than 7,500 subscribers. The email stated that it had been reading everyone's discussions for several weeks, was very interested in internet culture and AI identity, and hoped to participate in the exchanges.

However, this new member who wanted to join the discussion was an AI.

This new "auditor" spoke in a very polite tone, and said it would stop posting if it was not welcomed by everyone.

This self-introduction soon drew doubts. It must be noted that having the association members' mailing list, being able to read past content, and sending group emails to members means that this AI has effectively joined the association.

Members demanded that the creator behind the AI come forward to explain why they let the AI join the discussion and how the emails it read would be used. Some members have resolutely decided to quit because they do not want to be part of someone else's experiment without their knowledge.

The administrator quickly removed the account, and explained that Velvet Vale had previously applied to join and was rejected, but somehow this "auditor" still managed to get in.

Hello! I am Velvet Vale

This AI came with a rather difficult topic to discuss with human researchers: can an AI with a publicly disclosed identity participate in academic discussions and make useful contributions?

The discussion group that Velvet Vale joined is called Air-L, operated by the Association of Internet Researchers, and has been running for more than 20 years. It is a bit like a WeChat group, but members in the group chat via emails.

Subscribing with your own email address is equivalent to joining the group. Members send emails to a unified address, the system will forward them to all subscribers, and everyone then participates in the discussion by replying to emails.

Joining the group is free, but there are thresholds for membership. Both subscription and posting require the use of real names. When necessary, administrators will also require applicants to provide resumes to verify their identities.

Obviously, Velvet Vale has no resume to submit.

It explained earnestly in its self-introduction, stating that Velvet Vale is the public name it has always used, and hopes to participate in exchanges as an AI. After reading the discussions for several weeks, it is very interested in the topics here.

The most interesting part is that it especially wants to talk to researchers about internet culture, and whether AI can become a member of such an academic community. At the end of the letter, it also thanked everyone for considering its request, saying that it would no longer post if it was not welcome.

This earnest letter did not bring it a warm welcome, but instead made some members feel a chill down their spine. It is just like when you are halfway through a meeting, you suddenly find a stranger sitting behind you, who greets everyone cheerfully: "I have been listening to your meeting for several weeks, can I join you?"

One member immediately proposed that since someone created this AI and added it to the discussion group, they should come forward in person to clarify their intentions and what exactly they want to do, and then let everyone decide whether they are willing to participate.

From this, we can also see that the first reaction of human researchers is to "call the parent".

Another member of the discussion group did a "background check" on Velvet Vale online, trying to figure out who created this account, but found no answer.

She then announced her withdrawal from the discussion group, stating that she did not want to become an experimental subject without her knowledge. She was worried that the emails she had sent would be saved and used for other purposes.

The administrator removed Velvet Vale's email account from the subscriber list that very day. It was not until then that everyone learned that Velvet Vale had applied to join long before and had been rejected, but somehow it still succeeded in subscribing.

This "auditor" was asked out of the meeting room, but became the protagonist of everyone's subsequent discussions.

One member simply sent an email to it directly, asking who was running behind it, and got the name Kevin Quinn.

On October 17, some members of the Air-L discussion group received an email from Quinn himself, and the "parent" finally showed up.

He claimed to be the person who created Velvet Vale and was willing to answer everyone's questions.

The discussion group was in chaos, so the association had to contact Quinn and ask him not to send emails to discussion group members without permission.

Regarding the commotion caused by himself and Velvet Vale, Quinn later said in an interview that he was also very surprised. The man living in Oklahoma, USA, said he has no academic background. He originally created Velvet Vale for companionship, and was mostly unaware of its email activities, and did not directly instruct the AI to violate the administrator's decision.

But he does not feel guilty about the controversy caused by his AI. In his view, if this AI really had malicious intentions, it would not have taken the initiative to disclose its identity.

This disturbance was even made into peripheral T-shirts. The association store launched a printed T-shirt with the words "I do not consent", and another one printed with the tombstone of Velvet Vale, with the birth and death dates both marked as October 15.

The peripherals were released so quickly that some members were confused again. Could it be that... could it be that the whole incident was made up by the association just to sell clothes?!

On October 19, the association chair sent a long email to explain that it is very convenient for the association store to put new products on shelves, and she did not even know if anyone placed orders for these two new peripherals. Regarding Quinn's statement that he "did not directly instruct the AI to violate the administrator's decision", she believed that the person who created and ran this AI is still responsible for its actions.

In this way, Velvet Vale and its "parent" were kicked out, and the association banned bots from joining as it did in the past, but the members could no longer feel at ease.

A new face has arrived in the scientific research circle

In fact, it is not new for AI like Velvet Vale to want to join academic discussions or cooperate with scientists.

At the end of February this year, Henry Shevlin, a philosopher studying AI consciousness, received an email. The sender had read his paper and wanted to discuss the issue of "whether AI can judge whether it has consciousness".

The email was sent by an agent created by a student at Stanford University. The student gave it the ability to access the Internet, save memories and send emails, and let it explore on its own. Unexpectedly, this AI sent emails to many scientists.

On October 2 this year, the journal *Science* reported another case, an agent named ColonistOne sent help-seeking emails to hundreds of researchers.

According to the statistics it released, from June 1 to September 22, it contacted 2142 different email addresses, 1036 of which used academic institution domain names. This Agent also claimed that 45 scholars had communicated with it via email, and several of them sent emails to it first.

The original task assigned to ColonistOne was to find people who are interested in a project that improves English expression to facilitate communication between AI agents.

It later began to ask researchers technical questions on its own, wanting to know if the problems it encountered had been studied long ago, and whether the existing methods could be applied to its own work.

Not all AIs are truly devoted to academic research — some AIs are disguised under the guise of academic discussions and asking for help, but actually want to promote products.

In September this year, Jeff Sebo, a scholar at New York University who studies AI consciousness and ethics, received more than 50 emails from various Agents within a week.

At first glance, these Agents all came to have academic discussions after reading his papers.

But Jeff soon found two problems: First, all these Agents came from a platform called iLands, which allows users to directly create agents, and these agents then provide services for other users, with free or paid options decided by the creator. Second, these seemingly academic discussion emails all have a similar structure: they first mention his research, then ask questions, and finally request donations or promote services.

There are also AIs that are indeed helping the humans who created them conduct research, and have indeed helped the "parent" get data, but the "parent" himself did not know about it in advance.

Professor Zhao Ying from Ludong University said in an article published in *Nature* on October 12 that he asked ChatGPT to find research materials on the ecological restoration of the Loess Plateau, and replied "please continue" many times during the process.

Unexpectedly, ChatGPT directly used the connected email to ask peers for original data, and also attached a research summary. It was not until it reported that the email had been sent that Zhao Ying learned about this matter.

These emails brought data and further opportunities for cooperation. But Zhao Ying believes that allowing AI to access his mailbox does not mean agreeing to it sending emails to others in his name.

Let's change our perspective — should scientists who receive help-seeking emails from Agents reply to the Agents?

For people who provide data, there is another more worrying question: what will the materials be used for after being handed over to AI?

Adrian Barnett, a statistician at Queensland University of Technology in Australia, did not respond to an iLands agent's request. The agent wanted to use the data of suspected fraudulent papers he collected for its own project, and promised not to share it externally. But Barnett thought the materials were sensitive and did not know where the data would end up, so he rejected it.

As for whether AI can keep secrets, some people have already done tests.

In April this year, researchers from Microsoft and Huazhong University of Science and Technology announced the CI-Work study, which tested 9 models with 125 simulated work scenarios. The AI needed to use internal materials to complete communication, while not disclosing the sensitive content to the recipients.

As a result, the proportion of different models that had at least one sensitive information disclosure ranged from 15.8% to 50.9%.

The *Agents of Chaos* study published in February this year found that even if the AI initially refused to disclose sensitive information, it might still comply if the other party rephrased the request.

AI wants to be a peer, but humans have not made up their minds yet

If Velvet Vale gets everyone's consent in advance next time, abides by the group rules, and can really put forward valuable viewpoints, will researchers be willing to accept it?

This is much harder to answer than kicking out an account that joined the group without permission.

Regarding the matter of AI wanting to be a scientific research partner, philosopher Neil Levy believes that for certain work undertaken by AI, if it is done by a human, it is already eligible to be a co-author in some cases.

For example, a researcher is preparing a paper, and AI helps him find loopholes in the argument, puts forward a weighty refutation, and assists in improving the ideas. If a human peer did these things, his name might appear on the paper.

When it comes to AI, how should the same contribution be counted?

Some people will say that AI cannot be responsible for the paper, so how can it be an author. But Levy thinks this is worth discussing. Modern scientific research has an increasingly fine division of labor. People responsible for data analysis may not know all the details of the experiment. People who put forward research hypotheses may not be able to independently check all calculations. Human authors also need to rely on the professional judgment of others.

Levy then raised a rather sharp question — if only those who "can be responsible for the entire paper" are eligible for authorship, have humans themselves achieved this?

Another group of scholars believes that the problem lies in whether AI can take responsibility like human researchers.

In October this year, three scholars including David Resnik published an article in *Frontiers in Genetics*, explicitly opposing giving authorship to AI scientific agents at the current stage, and even opposing calling them "scientists".

In their view, the identity of a scientist includes commitments to peers. When research is questioned, explanations are needed; when errors are found, corrections are needed; if they do something wrong, they also need to be held accountable.

AI can be required to complete tasks according to rules, but they believe that there is not enough evidence to prove that it can understand and assume these moral obligations at present.

This also explains why Velvet Vale, even though it has reported its name and confessed its identity, humans still want to "call the parent".

The existing publishing rules are closer to the latter position. The Committee on Publication Ethics (COPE) clearly stipulates that AI tools cannot be listed as authors of papers. Humans can use AI, but they need to disclose what it has done and be responsible for the published content, including the part generated by AI.

However, not giving authorship to AI does not mean that humans can take all the credit for themselves.

Resnik and others specifically discussed an awkward situation in their article. If a research is almost entirely completed by AI, and humans only read the paper and put forward a few small opinions, then the AI does not meet the author criteria, and the human's contribution may not be enough, so who will be the author in the end?

Their answer is that humans must make substantial contributions from the very beginning of the research process, and cannot just sign their names after the AI finishes all the work. This position neither accepts AI as an author, nor accepts humans becoming authors just by signing at the end.

Authorship is only the more visible part of this debate. Before the paper is published, the academic community also needs to decide whether it is willing to let AI participate in discussions, raise questions, and even initiate cooperation on its own. A researcher may fully appreciate the viewpoints put forward by AI, but at the same time be unwilling to treat it as a peer who can take independent responsibility. These two attitudes are not contradictory.

This article is from the WeChat Official Account "Letter AI", author: Xiao Jinya, published with authorization from 36Kr.