Position piece urging universities to resist uncritical adoption of AI technologies such as LLMs and chatbots because they undermine academic freedom, integrity, and pedagogical skills.
0:00Welcome to Base by Base, the paper cast that brings genomics to you wherever you are. Thanks for listening, and don't forget to follow and rate us in your podcast app. So today, before we even touch a paper, I want you to just pause for a second.
0:13Yeah, I have a question for you, and it's a bit of a weird one. What happens when fluent text outruns truth? It almost sounds like a riddle, doesn't it? But it's actually, I think, the central tension of the moment we're all living in.
0:27Totally. We are watching the speed of information adoption. I mean, specifically AI in universities, in the media, in basically everything we read, it's all moving at like mock 10. Meanwhile, the rigorous pace of verification.
0:41you know, of checking the facts of integrity. That is still moving at a human walking pace. We're sprinting on a treadmill that's just going faster than our legs can move. And that mismatch, it creates this friction that we really need to talk about it.
0:54Which brings us to a really important disclaimer right at the top here. This is not a genomics episode. Usually, we are here breaking down the latest in, you know, DNA sequencing or CRISPR. or some cast 9 variant.
1:06Right. Today we are not looking at a wet lab. We're kind of turning the microscope on ourselves. Because, as regular listeners know, base by base uses AI assisted workflows to help create these deep dives.
1:17We are part of this ecosystem. We are. We're using the very tools that are currently causing this, well, this massive crisis of conscience in academia and in science. So it felt ethically necessary for us to just stop, look in the mirror, and openly discuss the really serious critiques being levied against AI and knowledge work.
1:37And we need to explore a fundamental question. Can knowledge translation, which is what we do here. Can it coexist with the integrity required for knowledge production, which is, you know, what scientists do?
1:48Or, to put it more bluntly, are the critics right? Is the way we make the show fundamentally corrosive to the science we're trying to share? To help us answer that, we're looking at a stack of sources, but the centerpiece is this very provocative pre-print, titled, against the uncritical adoption of AI technologies in academia.
2:05It's by guest at all, and it came out in 2025. And before we dive in, we have to be super clear about the status of this paper. Yes, huge disclaimer here. This is a pre-print. It has not been peer reviewed yet.
2:15It's an argument, a position piece, and it was released under a CCBY4 license. So it represents a snapshot of a very heated academic debate. This is not settled law. Not at all, but man, do they come out swinging?
2:30The authors aren't just saying, oh, be careful. They're arguing against the uncritical adoption of these tools entirely. They're basically yelling, stop while everyone else is hitting the gas. Right. And uncritical is the key word there.
2:42They aren't Luddites smashing the looms. They're scholars asking why we are letting a marketing campaign dictate the future of human knowledge. Okay, let's start right there with that marketing campaign.
2:52The 1st thing that really grabbed me was their argument that AI, it isn't even a scientific term anymore. Yeah, they called a marketing term. It's a fascinating point. They argue there's this terminalological disarray.
3:05If you look at figure one in the source. They have this really messy Venn diagram of terms. I saw that, generative AI, LLMs, chat bots, neural networks all sort of bleeding into each other. It feels like when you go to the grocery store and everything says natural or superfood.
3:20It doesn't actually mean anything specific. Exactly, but it makes you buy it. The authors point out that there are statistical methods, like, say, Bernoulli naive bays, that are technically AI in a computer science sense.
3:33They've been around for ages. But nobody is hyping them up? No. Nobody's selling Bernoulli naive bays as the revolution that will replace your doctor. Because it's just math. It's not sexy. Exactly. But artificial intelligence.
3:46That sounds like a digital brain. It sounds like magic. And the paper argues the industry uses this vague jargon to sell research. And, in their words, for laundering accountability. Laundering accountability.
4:00That is a heavy phrase. It sounds like a crime. In a way, they argue it is. If you call a system an agent or a chatbot, you start to answerpromorphize it, you treat it like it's a person. Right. So when it makes up a fake legal case or hallucinates a protein structure, you say, oh, the bot did it.
4:17Like a bad intern. Sorry, the AI messed up. Precisely. You've shifted the blame to a piece of software. But if you call it what it actually is, stochastic text generation. You remember, it's just a statistical model predicting the next likely word.
4:33And if that creates a lie, then the responsibilities on the human who deployed it. But even if we admit it's just math, there is this overwhelming feeling of pressure. I mean, I feel it. You feel it. This idea that it's all inevitable.
4:45The inevitability narrative. Yeah, the train has left the station. Get on board or get left behind. And the author's explicitly reject this. They argue that just because the technology spreads, it doesn't mean society has to just, you know, roll over and accept it.
4:57They use a comparison here that really stopped me in my tracks. They compare AI adoption to tobacco. Yeah, and combustion engines. It's a powerful historical reality check. It really is. I mean think about it.
5:11For decades, smoking was everywhere. It was in hospitals, on airplanes. In every movie. It seemed absolutely inevitable that everyone would just breeze smoke forever. And if you stood up in 1960 and said, I think we should ban smoking in restaurants, people would have looked at you like you were crazy.
5:26They said you were fighting progress or freedom or just the way the world works. But we changed it. We did. We realized the harm, we saw the data, and society stepped in. We have smoking bands. We have regulations.
5:39And they also point to gun laws, right? How the UK reacted to the Dunblane massacre with strict bands, while the US, well, the US took a different path. And their point is that technology is not a force of nature like a hurricane.
5:52It's a choice. We make choices. So they're arguing that resisting AI and academia isn't being backwards. It's a valid, even necessary stance to protect the ecosystem of human knowledge. Correct. And they go further to discuss the costs, because often when we use these tools, they feel so weightless.
6:12It's just the cloud, right? it's magic But the paper touches on the extractivism argument. You mean the environmental cost? That's part of it, for sure. The massive amounts of water and energy it takes to cool these data centers were basically boiling lakes to generate text.
6:28But they also focus heavily on the labor cost. The ghost work. We've heard about this, but I think it's easy to gloss over. I think we do. We see a polite, helpful chatbot. We don't see the 1000s of workers, often in the global South, paid pennies to spend their days looking at the most horrific content imaginable.
6:46Violence, hate speech, abuse. And tagging it. So the model learns not to show it to us. It's psychological trauma that's been outsourced. So we can have a sanitized experience. And the authors argue that this makes the technology unethical by default.
7:00You can't separate the tool from the exploitation that was required to build it. Wow. Every time we prompt the model, we are utilizing that ghost work. That's a hard pill to swallow. It is. But here's where it gets even more personal for me.
7:13And I'm honestly a little scary. They talk about the integrity of writing itself. The paper discusses the concept of the 2 victims of plagiarism. This comes from the work of Jonathan Bailey, which they cite.
7:25And usually when we think of plagiarism, or in this case, automated plagiarism via AI, we think about the 1st victim. The original author. Right. If I steal your work, you're the victim. I took your credit.
7:38Simple enough. But the paper argues there's a 2nd victim. And it's the audience, isn't it? Is the person listening to this right now. Exactly. It's breach of contract. When you read an essay or a paper or you listen to a host, you're operating under the assumption that a human being did the cognitive labor.
7:55You assume that I thought about these ideas. I wrestled with them. I verified them and then I wrote them down. And if you just prompted a bot to say, write me a script about AI. You're lying. You're presenting a simulation of thinking as the real thing.
8:08You're defrauding the listener. And that connects to their deepest fear, which is deskilling. The authors argue that writing is the process of thinking. You don't just write down a thought you already fully have.
8:20You refine the thought by writing it. Oh, I feel that. Sometimes I don't know what I really think about a topic until I'm halfway through a paragraph. The struggle is where the insight comes from. Right.
8:31So if you outsource the writing to an algorithm. Because it's faster. You are outsourcing the thinking. You're skipping the gym. And if you skip the gym long enough, your muscles atrophy. So the fear is we become a generation of editors, not writers.
8:46We just approve what the machine says, and eventually we lose the ability to tell if the machine is right or wrong. We become dependent on the machine to tell us what is true. Okay, that is a devastating critique.
8:56And frankly, it's a strong one. It's a real steel man argument against uncritical adoption. It is. about marketing hype, environmental harm, and the erosion of human thought. It creates a very, very high bar for anyone wanting to use these tools ethically.
9:11So now we have to pivot. We have to look at the other side of the coin. Because we are here, we are using these tools. Is there a way to use AI that doesn't destroy science? Is there a middle ground? I think there is, but we have to be incredibly precise about the distinction between knowledge production and knowledge translation.
9:29Okay, let's break that down because the guests at all paper, it's mostly talking about one of those, right? Yes. The paper is focused largely on academia, on universities. On the creation of new truth, so, researching a new protein, writing a thesis, grading a student's ability to reason.
9:46Right. In that context, I agree, AI is incredibly dangerous. If a student uses chat GPT to write an essay, they aren't learning. If a scientist uses it to fabricate data, that's just fraud. That is knowledge production, making the new thing.
10:00Exactly. But science communication, what we are doing right now isn't always translation. The truth already exists. The peer-reviewed paper exists. The discovery has been made. The rigorous science is done.
10:12Our goal now isn't to create new facts, but to make those facts accessible to a wider audience. To take a dense, jargon heavy PDF and turn it into something you can listen to on your commute. And in that context, the AI isn't the scientist.
10:26It's well, it's scaffolding. The paper actually brings up the calculator analogy, and they reject it for students. They argue, we ban calculators for kids learning math, so they learn the process. Which makes sense.
10:38You have to know how addition works before you automate it. But professional engineers do use calculators that use CAD software. They use simulations. Why? Because they've already mastered the fundamentals.
10:49They have the expertise to know if the calculator gives them a weird answer. So the argument is, once the expertise is established, the tool just aids efficiency. Precisely. If an expert science communicator, a human uses AI to help brainstorm metaphors or organize a script or summarize a section to check for clarity, that is fundamentally different than a student using it to fake an assignment.
11:13The key word, the only word that really matters is verification. Absolutely. You have to treat the AI like a very confident intern who is prone to making stuff up. You check every single thing it says.
11:26And that brings us to the most difficult part of this deep dive. We have to look in the mirror. We have to ask the hard question. Does base by base fit the criticism of uncritical adoption? Are we part of the sludge that guest it all are warning about?
11:41The question I ask myself constantly. And I would argue, no, but only because of the specific guardrails we have in place. It's not just ask chat GPT and hit record. We have a human in the loop policy.
11:54I think we owe it to you, the listener, to explain exactly what that looks like. Okay, let's lay it out. Our defense in the court of public opinion. Guardrail number one. Roll of AI. For us, AI is a communication tool.
12:05Not a source of scientific truth. It does not know things. It processes text. We never ask it, hey, explain how CRISPR works from memory. Because it's memory is a fuzzy JPIG of the internet. It'll hallucinate Exactly.
12:17Which leads to guardrail number two. Source of truth. The scientific content comes only from the uploaded peer reviewed papers. If it's not on the source text. It does not make it into the show. We don't let the model browse the open web for facts that might be wrong.
12:33But even then, models make mistakes. They misinterpret sentences sometimes. Which is why guardrail number 3 is the most important. Human curation. Humans real biological humans verify the key claims against the original source.
12:48We check the metaphors, we check the stats. We assume the AI is wrong until we prove it right. And guardrail number 4 is, well, it's what we're doing right now. Transparency. We disclose that AI assistance is used.
13:00We don't pretend 2 people sat in a room and just improvise this entire dialogue from scratch. We are open about the workflow. And finally, the correction loop. If we get it wrong and humans get things wrong too.
13:11We invite public corrections. We want to be held accountable. Okay, so the authors of the pre-print, they worry about polluting the ecosystem of knowledge with unchecked garbage. It is a valid fear. There is a torrential outpouring of generated content online right now.
13:26A deluge. Yeah, but our defense is that by citing the original paper, by adhering strictly to the source text, and by keeping humans in the driver's seat. We're trying to build a filter. We're trying to be a guardrail against that pollution, not a source of it.
13:42But we're just one show. The listener is out there swimming in an ocean of this stuff, news, social media, other podcasts. So you need a survival kit. You need a critical listening checklist. Yeah. And based on the principles of integrity in the guest at all paper, hear what you should ask yourself when you consume any content today.
13:59Okay, let's hear it. the 1st check? Transparency. Is the creator honest about using AI? If they're hiding it, or if it feels weirdly generic, and they claim they wrote it all. Be suspicious. What else are they hiding?
14:12Good one. Check number two Sourcing. Can you trace the claims back to a primary human source? In our show notes, do we link to the paper? You're reading an article and there are no links, no citations, just these confident assertions, that's a huge red flag.
14:25Okay, check number 3. Responsibility. Is there a human taking the blame? If the info is wrong. If an account is anonymous or fully automated, or just an avatar, there is 0 accountability. And the final one?
14:41Skepticism. This is the hardest one. You have to treat fluent text as confident but potentially wrong. Just because it sounds smart, just because the grammar is perfect and the voice is smooth, doesn't mean it's true.
14:54Remember, fluent text can outrun truth. Exactly. This all raises a really important opportunity for community discussion. We don't want this to be a one-way broadcast where we just, you know, justify ourselves and move on.
15:05No. We really want to hear from you on this one. the whole point of doing this deep dive. So if you go to the Q and A section on Spotify for this episode, we have a specific prompt waiting for you. And the question is this.
15:15Does the base by base workflow, so, human curation plus transparency, does that satisfy the ethical concerns raised by guest at all? Or is any use of AI and science communication a slippery slope? We want a respectful debate.
15:29Are we justifying our tools or are we using them responsibly? Tell us what you think. Because if we connect this all back to the core argument of the paper. They say that resisting is a valid stance. And we need to know if our listeners agree, or if you value the accessibility that these tools can provide.
15:46So let's try to summarize this massive debate. We've looked at the critique. AI as a marketing hype, the harms to labor and the environment and the threat to the very integrity of thinking itself. And we've looked at the defense.
15:58That crucial distinction between production and translation, the necessity of human verification, and the idea that tools can aid understanding if and only if they're used with rigorous oversight. It's a balance between protecting the sanctity of human thought.
16:15And using modern tools to make that thought accessible to the world. I want to leave you with one final provocative thought. The paper argues that writing is thinking, if that's, you know. What happens to our ability to think collectively as a species if we stop writing to each other and start prompting machines to talk for us?
16:32That is a question that keeps me up at night. If we lose the struggle of writing. Do we lose the ability to reason together? On that cheering note, We're going to hand you off to our closing track. And a quick disclaimer, the music you're about to hear is an artistic reflection on today's theme, generated to match the mood.
16:50It is art, not evidence. Thanks for diving deep with us. Keep your curiosity high and your skepticism healthy. We'll see you next time on Base by Base.
