The human brain has 80-100 billion brain neurons (even 2B+ in the stomach). The parallel way in which it handles sensory data and its ability to combine timed different inputs is amazing. The use of symbolic retrieval to speed up decisions, makes previously encountered situations instantly available. The link between body and maind is inherent and produces a usually synced organism. The decison making also involves history, logic emotions, hints and blind luck. It is also easily damaged, illogical and sometimes annoying.
Synthetic physical entities are durable, can be produced in great numbers quickly and start with their abilities immediately after manufacture. They are limited by power requirements and require external input for deeper analytical power. The bit barns have a much greater ability to handle data and access to power. They are of course not mobile. Basically "AI" as it stands is 2D and has to build up layers to simulate the 3D of the real world. The present working "neuron" maximum at the present is around the 1B mark but design and engineering move faster than evolution.
Google's Deepmind AlphaGo victory over traditional Go champions was a stunning exhibition of how far machine learning has come. It's use of unorthodox and persistent winning algorithms showed the advancement of mathmatical logic. What was downplayed was the massive use of background resources including computer, power and technical support to win this one form of board game. Mr. Fan Hui on the other hand had his own knowledge, a cup of coffee and various snacks to keep him ticking over. This shows the inefficent but dependable use of corporate brawn.