Thanks for the kind comments, JR. I agree with your comment about human intuition, though at times I see something interesting in the neuroscience and bio-inspired ML literature. As we struggle to create better neural nets, we start learning more about human intuition. Sometimes the parallels are spooky.

But what machine intelligence today really lacks is cognition. None of the AI out there right now, even the state of the art stuff, gives the machine an awareness of what it is, why it is making decisions, and what the consequences of those decisions might be. A self-driving car would happily drive you straight into a truck, just as soon as it would stop the car at a red light, if its training never covered its current sensor inputs.

And yes indeed, machine learning algorithms can be tricked. There is a whole area of research devoted to this, especially in computer vision, coming up with ways to fool ML algorithms (e.g., adding "white noise" to an image that is imperceptible to humans, but causes bedlam for the algorithm).