Interview, Fireside Chat
Garry Kasparov: IBM Deep Blue, AlphaZero, and the Limits of AI in Open Systems | AI Podcast Clips
- Humans are expected to prefer experienced practitioners over top experts in fields like radiology within a foreseeable future to avoid conflicting with machines' superior knowledge in approximately 95% to 96% of areas where machines outperform humans.
- Machines will maintain absolute dominance in closed systems such as chess, Go, Shogi, and video games by minimizing errors, with a consistency gap described as comparable to the difference between Ferrari and Usain Bolt that is expected to remain "insane."
- Specific milestones in machine superiority include mobile phone chess apps being stronger than the 1997 Deep Blue system, engines like Stockfish or Houdini identifying significant mistakes from that match within 30 seconds, and the historical precedence of IBM's machine-generated knowledge in backgammon dating to the early 1990s.
- While machines will correct specific weaknesses only after hundreds of thousands of games regardless of losses, humans will retain unique roles involving flexibility, the ability to make minor tweaks, and influencing subtle outcomes like altering a bullet's trajectory by 0.1 degrees to achieve a 10-meter difference at a mile.
- Critical risks involve humans attempting to interfere with machine knowledge in the 95% to 96% of territory where machines are superior, which is identified as a disservice and the greatest danger in human-machine relations.
- Future human-machine cooperation requires humans to understand their specific contributions, as machines remain unable to determine relevant questions, set directions, or reproduce unique human qualities despite humans eventually representing only the last few decimal points of total intelligence compared to machines.