Artificial Intelligence (AI), despite its recent successes on several fronts, has still serious limitations when it comes to the problems related to the deep understanding the world. Law is an area, where mostly shallow AI probabilistic solutions are not useful. As a consequence, many of the standard routines, where humans are performing well, are still not solvable with the current AI techniques and systems. In the presentation we will touch what are the limits of the current AI and what is the reason for them. Recent successes and popularity of AI is often raising expectations what technology could do – in reality, many of the “impressive” technological solutions are based on shallow effects which machine learning (as a key today’s AI technology) can capture from the big data. But there is a serious problem on how the world and the context is perceived by the technology. We will present a thin line between what AI technology can do and what cannot do based on current developments and where this line can move on the future and how the law can benefit out of it.