I read the GATO paper. I also saw the press hype. It is clear that the press has no idea about what GATO seems to be able to actually do.
I can highly recommend the book "Rebooting AI" by Marcus and Davis (http://rebooting.ai/). They lay out a clear case why the current emphasis on things like deep learning won't get very far toward more general AI.
In my research project the task was human/robot cooperation to change a car tire.
Deep learning was useful for recognizing lugnuts and recognizing mis-threading due to torque issues. It is useful for things that are hard to describe such as how to ride a bicycle.
GOFAI (good old fashioned AI) was useful for task modeling and planning such as setting the parking brake before attempting to work on the tire. It is useful for things that are easy to describe such as how to assemble a bicycle.
Some tasks are learned by "compiling actions", an area without a lot of research as far as I know. If you want to learn to play a guitar the early stages are repetition. Gradually the process is "compiled into the hands". Typing is similar. Emacs is my ultimate example. Often when I try to tell someone about a good key sequence I have to act it out with my fingers before I can explain it because the sequence is "compiled into the fingers".
People seem to think that Artificial General Intelligence is "in the near future". Musk, for example, is claiming that he will soon produce general purpose robots. He found out that self-driving was harder than he expected. Wait until he tries to make a robot enter a new house and make a cup of coffee. At best, both the coffee and the experience will be bitter.
As to the question of robots being conscious, consider that people who drink a lot can "black out", still capable of doing many things, but being unaware that they have done them. "Doing" and "awareness" seem to be unrelated.
The constant hype will eventually lead to a second "AI winter".
