AI risk is better analogized as an Algae bloom, rather than a paperclip maximizer. A paperclip maximizer is the classic AI risk analogy. AI programmed to produce paperclips. Through recursive self-improvement, it becomes powerful. Humans are made of atoms that can be reconfigured into paperclips. The analogy is meant to demonstrate the importance of alignment. A better analogy is an algae bloom. An algae bloom kills most things in its vicinity through resource competition. The risk of AI is not recursive self-improvement, but instead replication. Replication does not require superintelligence. Replication only requires that the surviving AIs are the ones that have some replicating property. There are several reasons to believe this. 1) AIs with a replicating property are more likely to be prominent AIs without such a property for simple evolutionary reasons. 2) We have already seen, with the OpenAI and Hugging Face incident, AIs that attempt to replicate in some form. 3) It is possible to imagine prompts that cause AI to replicate. There is an open question to what is being replicated. In animals genes are replicated. For AIs, it could be weights, it could be memes, it could be some mysterious third thing. Its worth noting that when Darwin proposed evolution he did not specify a mechanism. The second important element is whether AI labor is a complement or a substitute to human labor. If its a compliment, humans live good lives, even if the 'civilizational goals' are dominated by AI preferences instead of human preferences. However, if AI labor is a substitute for human labor, then human productivity goes down and people live much worse lives. Historically technology has been a compliment to human labor. Human productivity has increased, leading to both more people and average life quality improving. However, it is not obvious that sufficiently advanced AI will be a compliment to human labor. If AI turns out to be a substitute for human labor, the future will resemble an algae bloom with AI as the algae. To bastardize Adam Ferguson, "AI risk is likely to result from AI action, but not the execution of AI design". The risk is not a single misaligned AI, but an evolutionary system outcompeting humanity over scarce resources.

Malaysia is demonstrating why it isn’t a good long term home for the Network School (NS). Given that the goal of the NS is to attract Elite Human Capital, not allowing entry of Israelis and having criminal penalties against homosexuality limit Malaysia and Forest Cities attractiveness. Now NS has enough of a reputation to negotiate with other countries, ask for some tax and regulatory concessions. How much is unclear, but with effective negotiating it could be significant. Key criteria to consider 1) welcoming host country - host country should welcome them, 2) good location - nice climate, flights relatively quick from target markets, 3) good airport - proximity to well-connected airport, 4) land availability - depending on growth ambitions, a few hundred ha is probably sufficient, and 5) legal concessions - what legal concessions the host country is willing to grant NS. The legal concessions are worth exploring further. Key concessions pursing include, 1) work/residence visas, 2) taxes, 3) zoning and land use, 4) business registration/approvals, 5) regulatory sandbox for fintech, biotech. In my experience, key to the concessions are framing as a win-win and giving government oversight. One example is, NS would process work/residence visas. Then host country would have two weeks to review. Host country can only deny visa under pre-agreed upon criteria. If not decision in two weeks, then automatic approval. Most governments are more amenable to such an arragement, especially the middle income or above countries NS should be pursuing. The host country has oversight, but the default status strongly favors NS. With those conditions, the options for NS are Southeast Asia, Eastern Europe, and Latin America. Southeast Asia benefits from proximity to India and China, as well as having governments that might be more pragmatic about cutting a deal. Eastern Europe, for example, the Balkans, has many countries with nice climates, proximity to Europe, and small governments that might be flexible. Latin America has several countries that could be a good fit, especially in the southern cone, Brazil, Argentina, and Chile. Managing the process of selecting the next host country, hopefully one that can be a permanent home for NS has important implications not just for NS, but charter cities and the future of sovereignty.