BLOG
Nooriam: Solving AI Coordination Problems
Legal infrastructure is sovereign infrastructure because it solves a major coordination problem that exists within a nation state: rapid propagation of AI systems which compel usage but with no ready means of coordination and machine speed governance.
There’s a valuable series of essays describing coordination problems which abound frequently in the world, and which arise with the rapid distribution of AI across digital, social and institutional systems.
Coordination problems already arise in the world, with common examples being arms races and the tragedy of the commons. But there are many other scenarios.
For example, in an arms race, nations will compete to increase their military spending. Every increase in spending by a perceived external threat, necessitates retaliatory spending at the expense of other spending. The relative position remains the same but the absolute position is worse. The solution is to recognise the trap and coordinate: ceasefires, peace talks or non-proliferations treaties.
So we’re talking about systems or games where most people if not all people don’t really don’t like the system but are seemingly forced to participate in the system. However, the process of participating results in formally, a negative sum game. It’s a race to the bottom which can’t seemingly be exited.
Coordination problems are by their nature complicated and hard to correct. They only get solved, when people become aware that they’re in a coordination problem and then coordinate: but by their nature, these systems tend to limit cooperation. Compounding this is that often these systems fall on an existing social topography, and exacerbate pre-existing inequalities and deficiencies.
So what does it mean for AI? AI may be such a system which gives rise to a coordination problem. The world may potentially benefit from its productivity gains but where a system is pervasive across so many sectors, it also can produce negative externalities such as destruction of industries, loss of human attention and input, and governance failures, which can’t be escaped because we still have to use the AI systems to avoid falling behind.
James Stephen Brown, mentions what we shouldn’t do:
- Focus on one variable.
- Implement one measure.
- If it fails for any reason, scrap the program.
Education that these problems exist is a starting point. Problems of these nature require multiple variables to be considered together, transparency and coordination by parties.
Nooriam is fundamentally a permissioning engine with a few special features, which fulfils coordination within and between organisations by allowing them to share permissions, and allocate rights and obligations dynamically at scale. It resolves the Legal Attribution Gap (another formal name for the AI coordination problem) utilising perhaps the most ubiquitous and millennia old coordination mechanism: the legal contract.
Links to essays:
- Scott Alexander: (https://slatestarcodex.com/2014/07/30/meditations-on-moloch/ (https://slatestarcodex.com/2014/07/30/meditations-on-moloch/))
- James Stephen Brown (https://www.lesswrong.com/s/wDX7LJSytg8dLKnxp (https://www.lesswrong.com/s/wDX7LJSytg8dLKnxp))