Visualizing the gap between what humans can do and what needs to be done

January 18, 2025

The overwhelming volume of work that needs to be done (click for full size)

Every minute, millions of security events flow through corporate networks. Thousands of telescopes capture asteroids that could threaten Earth. Medical researchers analyze countless genetic sequences looking for disease patterns. And millions of hours of video are captured—many of which include crimes being committed.

But nobody's paying attention.

Not because we don't care, but because there's just too many things to watch. To do. To monitor. To take action on.

Let's call these Work Tasks.

All the different work tasks that we could potentially be doing

This mock-up shows the size of the problem. The x-axis represents Work Task volume—how many tasks, how much data, how many decisions. The y-axis represents Work Task difficulty—the complexity, expertise required, and cognitive load.

The area under this curve? That's everything that should be monitored, analyzed, and acted upon to maintain and improve our civilization.

The Human Reality ​

Now let's overlay what humans can actually accomplish:

We're only doing a tiny amount of what needs to be done

That tiny blue area? That's us. That's the sum total of human capacity to process information, make decisions, and take action.

Think about it practically from a trained professional perspective:

• A security analyst can carefully review only a certain number of alerts per day, while their network generates millions
• A radiologist can thoroughly examine only so many scans per day, while thousands await review
• A fraud investigator can deep-dive into very few cases, while thousands of suspicious transactions flow by

And there are only so many of these people in the first place. They're not easy to train, hire, or keep available in a talent pipeline.

We're not failing, we're just finite. And the bigger our society the more impossible it is to cover all the tasks that require human-level analysis.

Also keep in mind that part of the answer here is automation, for sure. Some things can—and have—been done with tooling that can find basic patterns in data. But what we're talking about here is work that can't be done by automation because it requires intelligence.

Enter AI ​

So this is where AI comes in. Yes, of course, one way to look at AI is that it's going to disrupt millions of jobs and be a huge problem for society. And I do think that's the case.

But another way to think about this is comparing AI not to the work that a human would have done, but the work that a human never would have gotten to.

And not just any particular human, but any human.

AI taking on some of the work that never would've been done

On this view, AI isn't about replacing humans, but rather doing the work that was never getting done at all.

The two axis of augmentation ​

There are two dimensions that we're actually getting help with here: the amount of work, and the difficulty of the work.

• Volume expansion: AI can monitor millions of security events, read millions of logs, analyze thousands of medical images, review countless transactions—all at a scale that completely dwarfs human capability.
• Difficulty expansion: AI can also do work that requires extraordinary training for humans, and that is very limited in availability. Like there are only so many Ph.Ds, and only so many private investigators, and only so many threat hunters. And the smarter the AIs get, the more they can detect patterns too subtle for most humans to perceive—either due to training or raw human capability. AI can also correlate data across vast datasets, and solve optimization problems with thousands of variables.

And this isn't theoretical; it's already happening all over the world.

• AI systems monitoring network traffic catch breaches that would have gone undetected for months
• Machine learning algorithms identify potential drug compounds that human researchers would never have time to test
• Computer vision systems spot manufacturing defects in millions of products that would have shipped with flaws

This has been happening for decades already with Machine Learning. But now with modern AI we can do things that require more advanced human-level intelligence.

Real-World Impact ​

Let's make this concrete with actual examples:

The gap Between what humans can cover and what they need to cover is vast

Understanding the framework ​

The whole idea is based on this concept of Work Tasks, where each work task has a difficulty level. And there are simply n number of Work Tasks that exist in the world, n number that are being covered by humans, and n number that are not.

The area under the curve framework reveals three critical insights:

1. The work exists whether we do it or not: Those security breaches happen. Those diseases go undiagnosed. Those asteroids keep flying. The work is real, with real consequences.

2. Human capacity has hard limits: We can't just "try harder" our way out of this. The gap is too vast. It's not about effort—it's about fundamental biological constraints.

3. AI can help cover the gap: Look at the visualizations again and notice that AI doesn't necessarily have to shrink the human portion of the coverage. I think that will also happen, honestly, but for different reasons and the point is unrelated to this one.

A useful model ​

Understanding work as area under the curve—combining both volume and difficulty—gives us another frame to think about human-AI collaboration.

There are so many instances of fraud, corruption, subversion of democracy, malicious actors in the cyber world, criminals caught on camera, criminals recorded in audio, plainly obvious dangerous health signals, etc—that are just never acted upon.

Everyone can't hire a security team. Everyone can't hire a team of journalists to investigate a company.

This essential human work doesn't scale.

And AI can be part of that solution.

Summary ​

• The world needs exponentially more work done than humans can possibly handle
• Consider every piece of work as a Work Task, with each task having a difficulty level
• Human capacity is a tiny corner of total Work Tasks—we can only manage small volumes at relatively low difficulty
• Our capacity is also limited both in volume and in difficulty
• Most critical work simply never gets done: fraud uninvestigated, threats undetected, patterns unnoticed
• AI augmentation fills the gap by handling high-volume, high-difficulty work humans can't even address, let alone execute
• The focus here isn't on replacing human work, but using AI to help cover some percentage of the gap