Dear friends, I believe AGI follows the two steps, describe the world and interact with the world. Most machine learnings are describing the world, although in a not very desirable way and with not uniform models. Some other AGI models are studying how to interact with the world, which seems to reflect part of intelligence. The reinforcement learning tries to unify the two and works well for now. But the good performance is based on the simple simulation to the world and we don't know what if the description is more similar to the real world and more complex. By the way, I believe many real problems that we human and AGI are interested in are very difficult to describe in a mathematical way or computational way (uncomputable problems?) and many problems are not optimization problems. Therefore, RL seems incomplete. I think for human, computation is difficult and other things are relatively easy, however, for computer, computation is the easiest thing but how to define and solve other real world problems are quite tricky.
I'm not not sure these ideas are completely right but just for discussion. Best regards, Sun ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T713c0382d1723414-Me0cfda63da1b3f06013d536d Delivery options: https://agi.topicbox.com/groups/agi/subscription
