All LCA models are wrong. Are some of them useful? Towards open computational LCA in ICT
This paper argues that current Life Cycle Assessment (LCA) practices for ICT lack the necessary rigor due to opaque and inconsistent modeling, and it proposes an open computational framework featuring explicit dependency graphs and versioned repositories to ensure model credibility through traceability, clear scope definition, and managed non-obsolescence.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Idea: "All Maps Are Wrong, But Some Are Useful"
Imagine you are trying to navigate a massive, complex city (the world of Information Technology) to find out how much "pollution" a specific building (like a smartphone or a data center) creates.
The authors of this paper start with a famous quote by statistician George Box: "All models are wrong, but some are useful."
Think of a model like a map.
- A map is never the actual city. It's a simplified drawing. It leaves out every single tree, pothole, and cat.
- Because it's a simplification, the map is technically "wrong."
- However, if the map is drawn well, it is useful for getting you from Point A to Point B.
The problem, the authors say, is that in the world of ICT (computers, phones, internet), we are using maps that are so messy, outdated, or poorly drawn that we can't trust them to tell us if we are causing too much environmental damage.
The Problem: Why Can't We Just Measure It?
You might ask: "Why don't we just go to the factory, weigh the smoke coming out of the chimney, and count the water used?"
The authors explain that this is nearly impossible for three reasons:
- It's too hidden: The pollution happens in thousands of tiny steps (mining the metal, making the glass, shipping the parts) that are hard to see.
- It's too expensive: You can't put a sensor on every single screw in a laptop factory.
- It's too complex: A laptop has millions of parts. If you tried to measure every single one, you'd need a team of scientists working for a century.
So, instead of measuring, we guess using models. We say, "A laptop with 16GB of RAM and a 500GB hard drive usually uses about this much energy to make."
The Two "Curses" of Computer Modeling
The paper says that doing this kind of modeling in ICT is cursed with two major problems:
Curse #1: The "Black Box" Validation Problem
- The Analogy: Imagine a chef claims to have invented a new soup. You ask, "How do you know it tastes good?" The chef says, "I just made it up based on a recipe I heard once."
- The Reality: In ICT, we often can't go back and check the "ingredients" (the raw data) because factories won't share their secrets, or the data is too old. We are building complex models on top of other models, but we rarely check if the bottom layer is actually true.
Curse #2: The "Lego Tower" Problem
- The Analogy: Imagine building a giant tower out of Legos. You have a red block, a blue block, and a green block.
- The red block says: "I am 1 inch tall."
- The blue block says: "I am 1 centimeter tall."
- The green block says: "I am 1 foot tall."
- If you stack them without noticing the difference in units, your tower will collapse.
- The Reality: ICT models are built by combining many smaller models (like Legos). But often, one model assumes a "year" is 365 days, while the one below it assumes a "year" is 360 days. Or one model is for 2010 technology, and the one above it is for 2024. When you mix them, the result is garbage, but it looks like a solid number.
The Real-World Mess (Examples)
The paper gives examples of how this goes wrong in real life:
- The "Email" Myth: For years, people thought sending an email created a huge amount of CO2 (like driving a car). It turned out that was a misunderstanding of how data centers work. The "map" was wrong, but people used it to make policy decisions.
- The "Outdated Database": Imagine a study from 2020 used a database about how much energy a specific chip uses. In 2023, that chip was updated to be 50% more efficient. But the study from 2020 is still being cited today as if nothing changed. The "map" is obsolete, but we are still driving by it.
The Solution: Building a "Digital GPS" for Science
The authors propose a new way to handle these models so they become useful again. They suggest treating environmental models like software code.
Here are their four rules for a better system:
1. The "Family Tree" (Model Lineage)
- The Idea: Every model should come with a family tree. You need to know: "Who made this? What data did they use? Is that data from a real factory or a guess?"
- The Benefit: If a model is built on a shaky guess, we know immediately. We don't pretend it's a solid fact.
2. The "User Manual" (Model Scope)
- The Idea: Every model needs a clear label saying exactly where it works. "This model works for laptops made in 2022 in Europe." It should not be used for "phones made in 2010 in Asia."
- The Benefit: This stops people from using the wrong map for the wrong city.
3. The "Breadcrumb Trail" (Traceability)
- The Idea: If you see a number like "This data center uses 100 tons of CO2," you should be able to click a button and see the exact math, the raw data, and the code used to get that number.
- The Benefit: Anyone can check the work. No more "trust me, I'm a scientist."
4. The "Software Update" (Non-Obsolescence)
- The Idea: Models need to be versioned, just like software (e.g., v1.0, v2.0). If new data comes out, the model gets an update. Old versions are archived so we know what was used in the past, but we stop using the old ones for new decisions.
- The Benefit: We stop making decisions based on yesterday's news.
The Proposed Framework: The "Open Source" Approach
To make this happen, the authors suggest building a central, open library for these models.
- Think of it like GitHub (where programmers share code) but for environmental data.
- Instead of hiding data in secret spreadsheets, everyone shares their "recipes."
- The system automatically checks if the "ingredients" match (e.g., making sure you aren't mixing metric and imperial units).
- If a new study proves an old model is wrong, the system flags it immediately, just like a virus scan flags bad software.
The Bottom Line
The paper isn't saying we should stop trying to measure the environmental impact of technology. It's saying: "We are currently using broken rulers to measure the world."
To fix this, we need to stop treating these models as magic numbers and start treating them as engineering tools that need strict rules, clear labels, and constant updates. Only then can we trust the numbers enough to make real decisions about saving the planet.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.