After reading the report, the part that confuse me most is definition of AI. 
Actually the report doesnt give a precise definition of AI, but kind of gives some describe about it. In brief, report says AI should be able to enables an entity to function appropriately and with foresight in its environment and the difference between non-intellgent program or device and human brain is of scale, speed, degree of automomy and generality. The first one seems to easy to realize, a rubust internet browser that can load links on current webpage in advance can satisfy the requirements. The second just tells the difference but doesnt say what AI should achieve.The following is just what I think an AI should achieve and how that come into my mind.
We know that all programs, from the most essential aspect, are doing calculation, so what makes a difference to make the computing intellgent?
First I want to know why we think our usual programs are thought to be not intelligent. I think it is because its function is somehow computing for people. In other words, its just a tool like a hammer or a screwdriver just with different utility. However, if you have a pet, you train your pet with instructions like sit down, stand up, come here and so on or forbid them from entering a specific room or something else. When your pet can follow your instructions, you probobaly think the pet is clever and intelligent in some degree.
So I think the important thing should be that they can learn something and apply what they learn to their behavior, and the something should not be so specific that is limited to a certain job. Same example, if your pet can only take orders and go somewhere, you may feel its just a voice-driven toy. Pets are certainly not as intelligent as human, but they certainly can learn a lot of things. However, in my point of view, the most applications of AI mentioned in the report is for very specific usage, which is not AI but preparation for AI. Techniche like NLP, computer vision plays an important role in a real AI.
Another thing is the way of train or teach. One can say that the computers are the best students who can follow any program you teach them. It is true, but as people, or users, we certainly dont hope that we need to train AI by modifying the code or inserting a new program when they have come into use. We want to teach them in a way more natural for us: by language, guesture behavior and so on.
At last I want to say that the way AI is designed may not have to imitate the way human intelligence is formed. I think there should be some basic system generates our intelligence, maybe it relies on physical laws with complicated structure designed by evolution. So it is not because we designed the program such that the program is taking order then its not intelligent. On the contrary, we designed the program to be not intelligent. I think computing is another form of basic system that is able to realize intelligence with proper structure , input from all kinds of sensors and a more opening programing. However, the different basic structure may need different way to generate intelligence. Letting computer simulate our way of recognition can be costy and low efficiency.
