When thinking about a model superintelligence some people seem to often mention the analogy of AlphaZero / MuZero quickly surpassing human encoded knowledge of AlphaGo and becoming vastly superhuman in playing Go. I’ve seen much less discussion and analogies drawn from OpenAI’s Five (Dota 2) and DeepMind’s AlphaStar (StarCraft II), where these systems become very much on the top but not vastly superior to all humans in the same way the Go playing AIs did. Do people have thoughts on that? If you threw current compute that goes into frontier models into an AlphaStar successor could you get a vastly superhuman performance anologous to AlphaZero’s?
I assume they could do it since all capabilities have increased by orders of magnitude since 2019 (2019!), but I also don’t see the economic incentive to do so.
Plus I personally think these real-time scenarios in some ways stack the field against the human players by giving the AIs not just superior reasoning ability, but also allowing them to use superior reflexes etc. Like, in some sense it’s entirely unremarkable if an AI with 3000 apm (= actions per minute) beats a human player with 400 apm; and IIRC even if the AI is restricted to 400 apm, it still gets to take much better actions than the inflated apm count of human players.
Turn-based games don’t have this problem, and nor do real-life real-time tasks with robots because there it’s clearer that you’re measuring and comparing both physical and intellectual aspects.
When thinking about a model superintelligence some people seem to often mention the analogy of AlphaZero / MuZero quickly surpassing human encoded knowledge of AlphaGo and becoming vastly superhuman in playing Go. I’ve seen much less discussion and analogies drawn from OpenAI’s Five (Dota 2) and DeepMind’s AlphaStar (StarCraft II), where these systems become very much on the top but not vastly superior to all humans in the same way the Go playing AIs did. Do people have thoughts on that? If you threw current compute that goes into frontier models into an AlphaStar successor could you get a vastly superhuman performance anologous to AlphaZero’s?
I assume they could do it since all capabilities have increased by orders of magnitude since 2019 (2019!), but I also don’t see the economic incentive to do so.
Plus I personally think these real-time scenarios in some ways stack the field against the human players by giving the AIs not just superior reasoning ability, but also allowing them to use superior reflexes etc. Like, in some sense it’s entirely unremarkable if an AI with 3000 apm (= actions per minute) beats a human player with 400 apm; and IIRC even if the AI is restricted to 400 apm, it still gets to take much better actions than the inflated apm count of human players.
Turn-based games don’t have this problem, and nor do real-life real-time tasks with robots because there it’s clearer that you’re measuring and comparing both physical and intellectual aspects.