3 ms·
They're still doing powerful research at DeepMind, training reinforcement learning agents in VR where progress can be much faster. I think the path to AI will b
by visarga 9y ago
They're still doing powerful research at DeepMind, training reinforcement learning agents in VR where progress can be much faster. I think the path to AI will be simulation - we need to simulate virtual worlds in order to master AI. AI agents need to experiment somewhere. Just make the simulation good enough, detailed enough and study the domain transfer problem to make training in virtual domain and applying in real domain work. We have games with amazing renditions of the world created for human pleasure, we need an AI training playground (of course, OpenAI Gym and DeepMind Lab are starting points).