What’s Wrong with Being a Meat Proxy?

People online have recently been commiserating over a new kind of shared experience: you ask someone a question (typically a coworker), but instead of answering it themselves, they copy-paste a response from a chatbot. Sometimes they attribute the response to the chatbot, but other times it’s just obvious given the tone and verbosity that it hasn’t been written by a human. The internet has dubbed this kind of person a “meat proxy”: the meat proxy doesn’t do any of the work of answering a question themselves, but merely serves as the physical conduit (i.e., the “meat”) that can manipulate a keyboard and mouse well enough to move text from one place to another.

It is not a term that’s used positively. But why not? What’s wrong with being a meat proxy?

We should note that meat-proxydom is not strictly a phenomenon associated with AI use. For example, we sometimes act as messengers where our role is simply to repeat information and not do any of the cognitive work of figuring things out or evaluating information. In these cases, we serve a role that could have been served by a text message, letter, carrier pigeon, or really any other medium that can move information from one person to another; it just so happens that a person conveyed the message and a person is made of meat. There’s nothing inherently problematic about taking on this kind of role when the situation requires it: sometimes information needs to be conveyed verbatim, and doing so requires some kind of proxy.

The cases being discussed online, however, are different. When you ask someone for information, you invite them to participate in an act of information-sharing, and such acts are guided by both ethical and epistemic norms. The problem is that acting as a meat proxy for AI output violates norms of both kinds.

When you provide someone with information, you assume responsibility for it: if I ask you for some information and it turns out that what you told me was false, then you are deserving of blame and probably owe me an apology. Of course, if you put in a lot of work trying to figure something out and happened to be wrong by accident, then I likely won’t hold you accountable for the mistake. But if you gave me information that you didn’t really take the time to check, and then it turns out you told me something false, you have failed to uphold the epistemic responsibilities we have to one another as seekers and providers of information.

When someone acts as a meat proxy, then you risk violating an important epistemic norm: that you should provide information only if you have good reason to believe it’s true. This requires some amount of work on the part of information-providers, and because meat proxies don’t do any of this work themselves, they fail to meet this requirement.

There are, however, lots of cases where we provide information to someone but don’t put in much of the work in figuring things out ourselves. Say that you ask your coworker for some information, but they don’t know the answer offhand. So, they do a quick bit of research online and find some information that someone else has figured out, and relay that information back to you. This happens all the time: we occupy different positions in chains of knowledge-sharing, where sometimes we act as an intermediary between someone seeking information and someone else who has done the work of figuring things out. It doesn’t seem like anything has gone wrong in these kinds of cases, so what’s so bad about being an intermediary between someone and a chatbot?

There is a difference, though, between finding information someone else has already figured out and deferring to a chatbot so completely that you abandon any human judgment. This is because even when we are serving as an intermediary between someone seeking information and someone else providing it, we are still typically putting in some amount of work to try to ensure that we are meeting our epistemic responsibilities.

For example, let’s say that we’re coworkers, and you ask me for a solution to a problem you’re having. I don’t know the answer, so I go looking for information online. I come across three posts on an online forum claiming to have a solution: one initially looks promising, but upon closer inspection isn’t really relevant to the problem you’re having; the second addresses the problem head-on and provides an elegant solution; and the third tells me that the solution is to take a hammer and beat your keyboard into a fine powder. After not much thought, I choose to pass on the information in the second post.

I may not have done the hard work of figuring the information out that I’m passing on to you, but the fact that I chose the second post as opposed to the first or third shows that I did something to ensure that the information you’re getting is correct. Maybe in rejecting the first post I relied on my background knowledge to identify it as irrelevant, which is perhaps not something that a layperson would have been able to do. And in the case of the third post, I was able to use my nonsense filter to rule it out immediately, something that probably anyone else could have done but still required that I at least pay attention to what I was reading.

A problem with being a meat proxy, then, is that by eschewing the minimal amount of work involved in vetting information or passing it through your nonsense filters, you are failing to uphold your duties as a provider of information. But the meat proxy fails not just their epistemic duties, but some of their moral duties, as well.

Since the process of sharing information involves people interacting with one another, these acts have moral dimensions. One obvious way that an act of providing information can be morally wrong is if I lie to you. But there are other ways we can cause moral harm in our information-exchanging behaviors, as well. For example, instead of outright lying to you, I might just not treat your request for information with any sort of care, and blurt out the first thing that comes to mind. Or maybe we’re part of a group project where we are all reliant on each other for contributing to a body of research, but I slack off and don’t contribute anything, leaving the remaining group members to pick up the slack. In both of these cases, I am certainly acting like a lousy coworker or group member, but it also seems that I didn’t treat you as something worthy of respect.

By acting like a meat proxy, one fails to meet a moral requirement in the same way: taking someone’s request for information seriously requires that you do at least some cognitive work as the provider of that information, and blatantly pawning it off to a chatbot fails to treat that person with an adequate amount of respect.

So, here’s what’s wrong with being a meat proxy: you shirk your epistemic responsibilities as a provider of information, and fail in your moral duty to treat other people with respect. Thinking about these problems should deter us from being meat proxies. But they can also guide the way that we use AI in other cases, especially when we use AI to communicate with other people on our behalf. For instance, while it might not be so bad to get a chatbot to help you express yourself, you still need to do enough of the work to ensure that the information you’re conveying is true, and to show that you have taken the other person seriously enough to have put in the effort of responding. If you don’t, you’re not acting like a person; you’re just a pile of meat.