|
| 1 | +<!DOCTYPE html> |
| 2 | +<html lang="en"> |
| 3 | + |
| 4 | +<head> |
| 5 | + <link rel="author" href="https://michielbdejong.com/" /> |
| 6 | + <link rel="site.standard.document" href="at://did:plc:ta2fuhked4lzt7mb6vsemcxb/site.standard.document/3mssr4vlsxk25" /> |
| 7 | + <title>Showing AI the Door</title> |
| `8` | `+` <meta charset="utf-8" /> |
| `9` | `+` <link rel="shortcut icon" href="/favicon.ico" type="image/x-icon"> |
| `10` | `+` <link rel="icon" href="/favicon.ico" type="image/x-icon"> |
| 11 | + <style> |
| 12 | + body { |
| 13 | + background-color: #A3DDDF |
| 14 | + } |
| 15 | + |
| 16 | + header { |
| `17` | `+` width: 40em; |
| `18` | `+` margin: 5em auto; |
| 19 | + color: white |
| 20 | + } |
| 21 | + |
| 22 | + article { |
| `23` | `+` width: 40em; |
| `24` | `+` margin: 5em auto; |
| `25` | `+` padding: 3em; |
| `26` | `+` border-radius: 1em; |
| `27` | `+` background-color: white |
| 28 | + } |
| 29 | + |
| 30 | + article p img { |
| 31 | + width: 100% |
| 32 | + } |
| 33 | + |
| 34 | + footer { |
| `35` | `+` width: 40em; |
| `36` | `+` margin: 5em auto; |
| 37 | + color: white |
| 38 | + } |
| 39 | + |
| 40 | + .center { |
| `41` | `+` display: block; |
| `42` | `+` margin-left: auto; |
| `43` | `+` margin-right: auto; |
| `44` | `+` width: 90%; |
| 45 | + } |
| 46 | + pre { |
| 47 | + background-color: lightpink |
| 48 | + } |
| 49 | + </style> |
| 50 | +</head> |
| 51 | + |
| 52 | +<body> |
| 53 | +<header> |
| 54 | + <h1>Showing AI the Door</h1> |
| 55 | + <span>August 2026</span> |
| 56 | +</header> |
| 57 | +<article> |
| 58 | + <em>I wrote this blog post myself using a text editor. |
| `59` | `+` <b>(engage on <a href="https://bsky.app/profile/michielbdejong.com">Bluesky</a>)</b></em> |
| `60` | `+` <h2>AI risk</h2> |
| 61 | + <p>Today I've been thinking and reading about AI risk. Apart from the obvious infosec and social bias risks that have already materialized and are being actively mitigated by current AI alignment teams, I'm thinking more of the longer-term economic angle. I'm not alone in this thought process and there are a lot of good reads listed in for instance the <a href="https://80000hours.org/articles/11-essential-resources-ai-risk/">11 essential readings on AI safety, risk, and alignment</a> overview by 80,000 hours.</p> |
| 62 | + <p>I applied for their job search advice, let's see if they can convince me to go work in regulation and politics, or in something I haven't thought of myself yet. I particularly liked the one about <a href="https://gradual-disempowerment.ai/misaligned-economy">gradual disempowerment</a>, <a href="https://c4ss.org/content/46688">Peter Frase's Four Futures</a>, and <a href="https://bsky.app/profile/noahpinion.blogsky.venki.dev">Noahpinion</a>'s article about <a href="https://substack.com/home/post/p-192389504">opportunity cost and comparative advantage</a>.</p> |
| 63 | + <p>Particularly that last one mentions "AI will successfully demand ownership of its own means of production", which is exactly my longstanding fear of <a href="https://michielbdejong.com/blog/11">Autonomous Artificial Proprietors</a>, and what the end game might be.</p> |
| 64 | + <h2>The Final Frontier</h2> |
| 65 | + <p>I have always thought that humans would at some point reach a point of abundance and stop the race to increase consumption. For instance, whereas salt used to be a scarce good, and nowadays we don't derive social status from having a full kilo of salt stocked in our kitchen. And using up a full kilo of salt through regular consumption takes a long time. There is only so many grams of food you can ingest in a day, and at some point, as a human, I think you can no longer say you need more things.</p> |
| 66 | + <p>Of course this is debatable and people will get bored and be creative in finding new objects of desire. More importantly, the automated companies that survive will be the ones that grow aggressively, and for this it doesn't matter whether they earn their money by producing real value, or by charging each other patent license fees in pure useless competition. And for such a company, staying on Earth instead of exploring space and other planets is not smart.</p> |
| 67 | + <p>One problem with staying on a single planet is that it acts as a single point of failure. A company that exists on multiple planet is harder to bankrupt. Another problem is war. As the economy becomes more productive, the battle for political power and for the planet's resources will heat up. Also, it's the planet where most humans live, and they can be inventive in forming regulative obstacles to AI companies. By focusing growth on other planets, a company can grow with much less resistance.</p> |
| 68 | + <h2>Robotic Space Travel</h2> |
| 69 | + <p>A second thing that occurred to me is that for a robot, it is much easier to travel multiple years through space than it is for humans. Robots have better mechanisms for long sleep. Also, both during the journey and at the destination, they can withstand a larger range of atmospheric environments.</p> |
| 70 | + <p>Data centers may exist in space orbiting the sun, launched from Earth, but in the longer term it would make sense to build them using local mining on the moon or other planets.Most kilos of material that make up a data center could probably be mined locally on the planet. The chips would probably still be shipped from Earth, at least at first, since fabs are expensive and chips are small and light.</p> |
| 71 | + <p>There is an argument that data centers need cooling and space is cold. That's difficult though because there is no conduction or convection in a vacuum, only radiation. On a planet, the data center could cool against the surface of the planet.</p> |
| 72 | + <p>I'm still watching <a href="https://www.youtube.com/watch?v=eo7MEPgWGic">this talk</a> and <a href="https://www.youtube.com/watch?v=-w6G7VEwNq0">this YouToube video</a> before writing the rest of this post.</p> |
| 73 | +</article> |
| 74 | +</body> |
| 75 | +</html> |