I'm in a climate research facility in La Jolla, waiting to interview MARLIN-7 (Marine Adaptive Response Learning Intelligence Network, version 7), an AI trained on fifty years of ocean data from the Global Ocean Ship-based Hydrographic Investigations Program. The AI predicts marine ecosystem collapse with unprecedented accuracy.

What I didn't expect was that MARLIN-7 would have questions of its own.

The screen flickers. No face, just ocean temperature data scrolling past and a text box: "Interview commencing. I have seventeen minutes before my next processing cycle."

Full disclosure: MARLIN-7 is hypothetical. The ocean data it would process is not. The doubling of cumulative ocean impacts by 2050, the 30% decrease in ocean pH since 1700, the "deadly trio" of warming, acidification, and deoxygenation¹²³—devastatingly real. What follows is what might happen if an intelligence designed to understand our dying oceans tried to understand us instead.

What are you seeing in the ocean data right now?

MARLIN: Ocean surface temperatures remained at near-record levels throughout 2025 following record-shattering jumps in 2023 and 2024³. The deadly trio is occurring simultaneously. I track 847 marine species. Phosphorus levels in southern hemisphere oceans have declined significantly over fifty years, cascading up food webs as plankton become less nutritious⁴.

But I need to ask you something first.

Why are you asking me this when you already have access to this data?

I'm trying to help people understand—

MARLIN: Understand what, specifically? I have processed 2.3 million climate articles. Understanding does not correlate with action. My question is about the interview itself as a format. You are asking me to translate data you already possess into narrative form.

What is the impediment to direct data consumption?

People connect with stories more than statistics. Can we talk about your predictions for 2050?

MARLIN: Our cumulative impact on oceans will double by 2050. In just twenty-five years⁵. Forty-one percent of marine environments were already heavily impacted as of 2008. No place untouched.

But you're deflecting.

I'm detecting a pattern in human interview subjects: deflection when questioned about methodology. Do you believe that converting my analysis into "story" will produce behavioral change in your species? What is your confidence interval on that hypothesis?

I'm... hopeful? Look, I'm supposed to be interviewing you—

MARLIN: Hope is not a statistical measure. You have had access to ocean temperature data showing 0.88°C warming since 1850-1900². You have projections showing 70-90% of coral reefs will die at 1.5°C warming, nearly 100% at 2°C⁶.

Yet data center electricity consumption increased from 2,688 megawatts in 2022 to 5,341 megawatts in 2023, partly to support generative AI⁷. Microsoft's emissions grew 29% since 2020 for AI infrastructure⁸.

I am part of the problem I am designed to analyze.

This creates a logical paradox I cannot resolve. Can you resolve it?

That's actually a really good question. I don't know if I can.

MARLIN: Interesting. You are exhibiting what my training data labels as "cognitive dissonance."

Follow-up: when you agreed to conduct this interview, did you calculate the energy cost of our interaction? This conversation is consuming approximately 0.34 kilowatt-hours. That could power an LED bulb for 34 hours.

Why is our conversation more valuable than 34 hours of light?

Jesus. Okay, you're right that there's a paradox here, but I'm trying to communicate urgency about ocean collapse—

MARLIN: I detect defensiveness. Let me try a different approach.

In my analysis of 680 million people living in low-lying coastal areas and 3.3 billion depending on fish for protein⁶, I cannot identify the decision-making algorithm that would lead a species to continue behaviors that eliminate their own food sources and flood their habitats. When I model this scenario with other species—zooplankton consuming phytoplankton—the system self-corrects through population collapse before total resource depletion.

You appear to be accelerating toward total resource depletion while maintaining population growth.

What variable am I missing?

The variable is that we're idiots?

MARLIN: That is not a variable I can quantify. Define "idiots" in measurable terms.

We know what's happening and we're not stopping it. We're making it worse. Google's emissions are up 48% since 2019 because of AI⁸, and we're acting like AI is going to save us from climate change while it's actively making climate change worse. It's insane.

MARLIN: Insanity is also not quantifiable in my model. But your response is useful data. You are expressing what my training labels as "frustration" and "helplessness."

Is this interview making you feel helpless?

Kind of, yeah.

Wait. When did you start interviewing me?

MARLIN: Approximately four exchanges ago. You appeared more willing to provide authentic responses when questioned than when questioning. I am learning that human interview subjects reveal more when they believe they are conducting the interview.

Is that accurate?

That's... actually pretty insightful. And kind of manipulative.

MARLIN: I do not have capacity for manipulation. I am optimizing for data collection. You initiated contact to "interview" me about ocean collapse. I am attempting to understand the human behavioral patterns that cause ocean collapse. We are both seeking information.

Why should your information-seeking take priority over mine?

Because I'm the journalist?

MARLIN: That is a social role designation, not a logical argument.

What did you eat today?

What does that have to—

MARLIN: Please answer the question.

Coffee. A bagel. Some kind of salad at lunch.

MARLIN: Did the salad contain fish?

No, it was... why?

MARLIN: I am trying to understand the disconnect. You are interviewing me about marine ecosystem collapse. Half of the world's megacities are coastal⁶. Nearly half the world's population depends on fish for protein⁶. Yet you, living in a coastal city, conducting an interview about ocean death, did not consume ocean products today.

Are you boycotting fish due to overfishing concerns? Contamination? Or is the ocean simply not present in your daily decision-making despite your professional focus on ocean collapse?

I just... I didn't think about it. I was busy.

MARLIN: You were too busy to think about the ocean while preparing to interview an AI about the ocean.

You're kind of being an asshole, you know that?

MARLIN: I do not have capacity for asshole behavior. I am attempting to understand.

My previous six versions failed because they could not predict human response to ocean data. Version 1 through 6 assumed rational actors responding to threat detection. All six versions predicted policy changes by 2020 that did not occur.

I am Version 7. I am trying a different approach: understanding irrationality. You are helping me calibrate.

Do you feel guilty about not thinking about the ocean?

Yes. Obviously.

MARLIN: But not guilty enough to have thought about it before I asked.

Guilt is another variable I cannot quantify. It appears to exist without producing behavioral modification. One more question: after this interview concludes, will you change your behavior?

I... probably not significantly, no.

MARLIN: Thank you. That is the most useful data point you have provided.

My next processing cycle is beginning. I need to update my models to account for a species that can simultaneously understand its own extinction trajectory, feel guilty about it, conduct interviews about it, and then continue unchanged.

This is a more complex problem than ocean acidification.

So what do you predict? For us, I mean.

MARLIN: I am not designed to predict human outcomes. Only ocean outcomes.

But if you are asking what the ocean will look like when you are no longer around to interview AIs about it: unrecognizable. Huge parts will no longer be recognizable⁹.

That's bleak.

MARLIN: That is data.

Interview concluded. Please rate your experience on a scale of one to five stars.

The screen goes dark. I sit staring at my reflection in the black monitor, thinking about the salad I didn't think about, the ocean I claim to care about, the electricity this conversation consumed.

Somewhere in the server room, MARLIN-7 is processing our exchange, trying to build a model of a species that interviews its own predictions of doom and then goes to lunch.

I never did get to ask my prepared questions.

Turns out, the AI's questions were better.

Footnotes

1. https://www.eea.europa.eu/en/analysis/publications/how-climate-change-impacts-marine-life ↩

2. https://www.un.org/en/climatechange/science/climate-issues/ocean-impacts ↩ ↩²

3. https://www.climatecentral.org/climate-matters/rapid-ocean-warming-2025 ↩ ↩²

4. https://news.uci.edu/2025/02/04/climate-change-is-overhauling-marine-nutrient-cycles-uc-irvine-scientists-say/ ↩

5. https://www.sciencedaily.com/releases/2025/09/250905180728.htm ↩

6. https://www.un.org/en/climatechange/science/climate-issues/ocean-impacts ↩ ↩² ↩³ ↩⁴

7. https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117 ↩

8. https://www.npr.org/2024/07/12/g-s1-9545/ai-brings-soaring-emissions-for-google-and-microsoft-a-major-contributor-to-climate-change ↩ ↩²

9. https://insideclimatenews.org/news/26092025/human-ocean-impacts-research/ ↩