As a child who grew up loving Blade Runner and The Matrix and all things dystopian sci-fi, I can’t believe that this video is actually real, and that it’s happening, right now.

This video from Anthropic’s Chloe Lubinski should be required watching for all of humanity right now.

I’m going to break down two key concepts of this video that will make you smarter in your next discussion about AI.

I strongly suggest watching everything that comes from official people at Anthropic, OpenAI, etc. vs. trusting what the news tells us about these subjects. When we study the language of people actually inside these companies, the nuance of their words matters.

So here are two key concepts from the video:

1. Intelligence can be bought. Compute costs money. And more money=more intelligence=more potential power. So those with more money will get more power from AI. Hardly surprising. The reason our digital products are all getting more expensive this year is because of a global memory shortage, because AI has hoovered up all the chips. In fact, former Apple CEO Tim Cook called this shortage a “100 year flood" saying “I’ve never seen anything like it.” The rich getting richer is not a new concept. The most powerful people, nations, and companies will have a vastly larger, ever-growing share of compute. Access, assuming we are able to contain/control this “super intelligence”, will be disproportionately wielded by a few.

2. Recursive Self-Improvement is possibly close at hand. We all know that AI is exponential. But it only becomes truly crazy when AI builds its own successors. As Chloe says: When Claude 9 builds Claude 10 all by itself, for example, everything changes. This is a likely scenario. And then, AI development will accelerate at a pace we truly cannot comprehend. If you’ve been using these tools daily like me, consider how far things have come in just the last year. How much more is possible now than last April, and you start to get a sense of how fast this could go.

These two points are terrifying by themselves. But the implications of both of them combined are central to a fear of AI that is grounded in reality and not just doom-scrolling.

The pay-to-play, winner-take-all nature of this technology means there’s a strong economic feedback loop with geopolitical incentives that are practically impossible to slow down. Which means that self-recursive development will happen with little or no coordinated oversight.

And these incentives (greed, power) are so fundamental to human nature and involve such powerful global players, our individual actions can’t and won’t stop what’s to come. This is different than Coca-Cola, which if we all stopped drinking it, might go out of business. AI isn’t going out of business regardless of if we stop making AI cat memes.

Now look: There are many problems on planet Earth that could theoretically be solved if we all worked together, like clean drinking water for all. I know I know, that’s just crazy I’ll shut up. But agreeing on anything, is apparently one thing that humans categorically refuse to do.

So should you be more excited or more afraid of AI? Yes!

As a builder with these tools and a digital native of almost 30 years, I truly believe the rational response to this moment is to be both. If we are purely pessimistic, we could end up creating the very reality we are so afraid of. If we’re purely optimistic, we probably don’t understand the very real threat.

To quote Chloe from the end of her talk: “The stories we tell don’t just describe the future. They literally could help create it.”

So if you go around talking about Skynet all day… Well…