If the computer piloting the plane was designed by the same criteria -- economics -- as the plane was, it'd have pretty much the same crash rate per passenger * mile. It has everything to do with managing costs and acceptable risks, and not some inherent insafety of flight. An airplane can be made just about as reliable and reliable against pilot errors as you want. The technology is there, has been for decades, it's only matter of costs and performance.

The crashed plane was a two-engine model, designed so in case of one engine failure it could still fly safely. However, the margin of safety was small enough that once the wrong engine was throttled back, the plane became uncontrollable before the engine regained enough power. Spinning up engines takes certain, well-known, time. It'd be entirely possible to design and build a plane that would handle such accidental throttling gracefully. All you need is one more engine, or larger wings giving more lift at slow speed and thus allowing it to remain in the air long enough to recover.

The crux is the airlines choose not to go for either option simply to cut costs, accepting lower safety margin.

Now consider fully automatic plane flown by computer -- designing and debugging a robust hardware/software solution is very costly. Especially when it comes to handling tricky, rarely-occuring cases, like unexpected failures. Throw in both cost of handling of multiple versions of planes -- different engines, different avionics, and handling of flightworthiness reduced by wear and tear. Also attach the risk of mismatched configuration, with no experienced pilot inside to spot the problem.

In the end, the amount of money spent on development, and the level of redundancy of software/hardware autopilot would be dicated by cost-cutting and going for what the airline deems acceptable risks.

The computer pilot would be just as (un-)safe as human pilot, for the reasons of economics. Safety is calculated and kept per passenger * mile, not dictated by current technology.