We have seen how this Agent deals with a digital representation of an imaginary very simple environment where he can repeat the same routines over and over again until it manages to create an internal model of the game leading to the so called learning about how to play it.
I would like to see how this Agent performs in dealing with the real world where you cannot run the same event 1000 times changing your actions to see what happens. It would be nice in life to have a magic “wrong try again” button. The real world cannot be sampled in the way presented by this video, the real world has to be understood. Showing how the agent learns by dealing with games tell us nothing about the Agent ability to deal with the complex, non repetitive and often contradictory reality our brain has to master. I don’t think the speaker has been honest in saying that they have chosen to use games instead of the real world as a learning platform just because the team didn’t want to be distracted by solving electromechanical problems with the physical interface. The reality is that this Agent is still very far from what is needed to understand and operate in the real world. But I don’t think the Agent needs consciousness, as someone said in a previous email. Intelligence is not a synonymous of consciousness, you can have unconscious AGI, not that this makes me feel safer, quite the opposite, I would be terrified to deal with a super-intelligent unconscious Agent gone out of hand. I believe that when we will reach the goal to create a machine that can learn and understand, that can think and ask questions in order to improve its understanding we will be closer to AGI. We will see if and when consciousness will arise. I have a last consideration that I think is very important about the hard or soft take off. Once an AGI with the minimum characteristics above mentioned will be created, it will have to be trained in dealing with the real world and it will take a long time because the feedbacks form our world are very slow and not repetitive. The Agent will have to gain experience as we do and it will not matter how fast the Agent will process the information, it will still have to wait for feedbacks. It may take years, it may be like rising a child, nothing like an overnight full immersion. How close are we to this? ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
