Computer systems that memic the role of an expert in solving specific kinds of problems have been pretty successful. These expert systems have been used to diagnose medical conditions, trouble-shoot and resolve problems on the DS-1 space probe, determine the cause of operating system crashes (I worked on that one), etc. True AI as the term was meant to imply is a much more difficult problem DS-1 used a self-navigation capability that was pretty radical for the time. While under ion propulsion, you could give it the coordinates of the desired destination (such as a comet rendezvous) and DS-1 would automatically determine the best course to fly there. This was a non-trivial task because of the ion propulsion. While the force imparted by the ion engine was quite small, it had a big impact on the navigation calculations.

The cynics will tell you that artifical intelligence will never overcome natural stupidity. Perhaps the best hope for AI is Moore's Law about the power of computer systems doubling every 18-24 months.