The link: Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics (an NBER Working Paper)

The summary:

• Most of the impact of AI on productivity has yet to happen.
• As with other general-purpose technologies (e.g., electricity), this is partly due to slow capital accumulation, and partly to outdated organizational processes and culture.
• Successful organizational adaptation is rare, and there's little aggregate productivity growth on company cohorts older than five years.

Our takeaway for politics, government, and security: Even large, successful, technologically advanced States, groups, and security forces are at an structural disadvantage against competitors new or desperate enough to (re)build their processes and culture around new technologies, specially AI. Doubling down on size or technology, absent very rarely successful organizational change, won't help.

Practical tip: If you're running or facing an organization of any kind — political party, NGO, government function, security force, insurrection, etc. — consider creating a new one designed from scratch around contemporary and near-future technology; the potential upside is exponentially larger than just doing a technological upgrade. Not ending with the old one plus drones and a mobile app is the difficult part.

PS: The next issues of Reinforcement Learning are likely to be more focused on specific technologies and less drastic suggestions, but this one might very well end up being the most useful of all.