🏛️ The origin question
Every AI lab you use was founded with a stated purpose. OpenAI was created in 2015 as a nonprofit, explicitly to ensure AI benefited "all of humanity." Google DeepMind was founded with a mission to "solve intelligence" for the benefit of society. Anthropic — the company behind Claude — was founded in 2021 by former OpenAI researchers who left over disagreements about safety and the pace of deployment.
These origin stories matter. Not because companies always stay true to them — they don't — but because they tell you what trade-offs the founders expected to face, and what they claimed they'd prioritise when those trade-offs arrived.
The core tension: Building powerful AI safely and making it commercially viable at scale are not the same goal. Every AI lab has had to make choices about where the line is. The interesting question isn't "are they ethical?" — it's "what did they do when it got hard?"
⚔️ The military question
In January 2024, OpenAI quietly updated its usage policies. A line that had explicitly prohibited using its models for "military and warfare" applications was removed. The new policy instead listed specific things that were still banned — weapons of mass destruction, cyberattacks on critical infrastructure — but the blanket prohibition on military use was gone.
Shortly after, OpenAI announced a partnership with the US Department of Defense. The stated uses were relatively mundane: cybersecurity tools, software for veterans, code review. But the principle had shifted. A company whose founding document said AI should benefit "all of humanity" had decided that military contracts were compatible with that mission.
Worth noting: Anthropic is not above critique. It has raised billions from Amazon and Google. Its Constitutional AI approach is a real safety methodology, but it's also a PR asset. "Safer than the alternatives" is a meaningful claim — but it's not the same as "safe." No AI lab is beyond scrutiny.
👷 The labour question
AI tools are largely built by a handful of companies — most of them American, most of them funded by the same pool of venture capital. The productivity gains from AI are, so far, flowing primarily to those companies and their shareholders.
This is the question that politicians across the spectrum — from Bernie Sanders on the left to some libertarians on the right — have raised: when AI makes workers ten times more productive, who captures that value? The worker who now produces more? The company that employs them? The AI company? The shareholders?
Sanders has been specific: if AI-driven productivity gains go entirely to the top, we will have technological acceleration without shared prosperity. His proposal is for things like a shorter work week, profit sharing, and stronger collective bargaining — so that the productivity gains from AI flow to workers, not just owners.
The Raglan question: If a Raglan surf school uses AI to do what used to take five hours in two hours — does the school owner work fewer hours, charge less, or make more? In theory, all three. In practice, who decides is shaped by who has leverage. That's a labour question, not a tech question.
🌐 The concentration question
Three companies — Google, Microsoft (via OpenAI), and Amazon (via Anthropic and its own models) — control an enormous share of the AI infrastructure that the rest of the world depends on. This isn't unique to AI — it mirrors concentration in cloud computing, social media, and search. But AI has a specific wrinkle: the most capable frontier models require compute resources that only a handful of entities in the world can currently afford.
Open-source models (Llama from Meta, Mistral, Qwen from Alibaba) are a real counterweight to this. They are not as capable as the frontier closed models, but they are closing the gap — and critically, they can run locally, without sending your data anywhere.
🤔 What can we actually do?
It would be easy to end here with a vague call to "use AI responsibly." That is not very useful. Here are things that are actually actionable:
Choose your tools with your values in mind. The models you use send revenue to the companies that make them. This isn't a reason to avoid AI — but it's a reason to be deliberate about which companies you want to support. Using a local open-source model for some tasks is a real option.
Don't use AI for tasks where the harm is human. Using AI to draft an email to your council rep is fine. Using AI to generate targeted harassment, make decisions about people's lives without their knowledge, or automate tasks whose costs fall entirely on other people — those deserve more scrutiny.
Ask who benefits from this. When a company tells you AI is going to transform your industry, ask: transform it for whom? If the productivity gains flow to shareholders while workers are displaced without support, that's a policy failure — and it's worth naming it as one rather than treating it as an inevitable force of nature.
Stay curious about the open-source alternatives. Llama, Mistral, Qwen, and others are not as capable as the frontier closed models right now. But they are free, they are improving fast, and they don't require sending your data to San Francisco. They are worth knowing about.
The most honest thing I can say: The questions in this module don't have clean answers. The AI companies building these tools include people with genuine commitments to safety and benefit — and structural incentives that push in other directions. Holding both of those things simultaneously is probably the most accurate mental model you can have.
📺 Worth watching
Two videos worth your time — honest, non-hysterical takes on what AI is doing to the world.
✍️ Your take
No quiz here. Just a few questions to sit with. Write your thoughts — they stay on your device only.
Branch complete 🌿
There are no right answers here. Asking the questions is the point. The people building AI are not a monolith — some are thoughtful, some are reckless, most are both at different moments. Staying curious about who they are and what they're doing is worth the effort.