It’s the scorching season now, right at the doorstep of the summer heat, but when winter ends and spring approaches, it’s strawberry season. My kids love strawberries – a pack from the supermarket disappears in a single day. Out of nowhere, I think of those accounts that look down on people with a shallow understanding of AI while monetizing them, but maybe even they have no choice if they need to make a living.
Reading papers on arXiv is part of my daily routine. Though I say “reading,” my English isn’t great, so I have LLMs read them for me, ask about things I don’t understand, and bounce my own ideas off the AI. I used to turn the results into articles for note, but what value do those articles really have? In an age where anyone can easily summarize information, what’s valuable is the human “me” sleeping inside myself.
That preamble went on way too long, but I want to talk about a paper from the Centre for the Study of Existential Risk: “Integrators in War: Mediating Decisions on the Use of Force with AI Assistance.”
At first, I thought it was specifically about warfare, but at its core, it reveals the kind of people we’ll need in the coming era.
In short: AI is evolving too fast, and a gap is forming between humans and AI.
The paper illustrates the difficulty of technology integration with historical examples.
After the Wright brothers invented the airplane in 1903, it took 17 years before it was practically deployed on warships. Just putting a plane on a ship – why did it take so long? It was the difficulty of landing. Landing a plane on a moving ship exceeded what the engineers had imagined. In the end, engineers who understood both aviation and naval technology developed the arrester gear landing system and led the way to a solution.
What matters here is the connection between systems and humans. Having written this far, I thought “well, obviously,” but let me finish the thought first.
Even as AI evolves, we’ll still need that connection between systems and humans.
Take medical imaging AI, for example. Even if developers build a high-performance model, doctors will agonize over how to use it and where responsibility lies. That’s why we need medical AI coordinators. In education, even if you build a personalized learning system, things go more smoothly with an EdTech coordinator who helps teachers figure out how to incorporate it into their classes.
“AI will steal your job” is an overstatement. I think an era is coming where the people who bridge the gaps will take center stage. Engineers need to keep sharpening their skills so they can serve as translators between technology and humans. And what’s essential for coordination is human skills. This was a paper that made me think: the professionals who thrive in the middle, caught between both sides, will be the ones who move society forward.
Finally, here are the 10 pillars of responsible AI integration proposed by the paper. I had the AI write this part.
- Think first (Do you actually need AI?)
- Include diverse perspectives (Tech nerds alone won’t cut it)
- Handle your data properly (Garbage data = garbage results)
- Understand what the results mean (A wall of numbers is meaningless)
- Be clear about what you’re solving (Don’t let the goal drift)
- Document and explain (Someone will ask “why?” later)
- Build in the ability to admit mistakes (No AI is perfect)
- Make AI explainable (Black boxes are dangerous)
- Technology is political (Neutrality doesn’t exist)
- Prepare for emergencies (What if the AI goes haywire?)
Reference: https://arxiv.org/abs/2501.06861
As a side note, what I felt when I wrote a blog post on my own for the first time in a while was a decline in my writing ability. In this coming era, if we don’t think about how we use our brains, we’ll atrophy.