There are three main reasons: reaction time, sensor inadequacy, and predictive intelligence. Taking them in order...

For certain types of stimulus-response activities a human can respond faster and has a better reaction time than a computer. For example, imagine it takes 5 milliseconds to sense something completely, 10 milliseconds to transmit that information to the computer, and 10 milliseconds for the computer to figure out the right action, and 8 milliseconds to transmit the response to an actuator. That's a total of 33 milliseconds. A human can respond to an event in 20 milliseconds. Note that for some types of actions the computer's sense-and-respond loop could be much slower than the numbers I have given. For example, most artificial intelligence algorithms are complicated and take a lot longer than 10 milliseconds to complete. Some types of optimization algorithms, like dynamic programming have unpredictable running time, so they might finish in say 5 milliseconds with one set of inputs and 100 milliseconds in another.

The next big problem is that the human eye, which is wired in parallel directly to the brain, is a far better sensor than anything available to an aircraft. It has a wider field of view, higher bandwidth and better focusing and movement ability than any man-made optical system. Currently the sensors on planes are things like inertial measurement units, GPS and flight instruments (altimeter etc). These sensors are not enough to reliably land a plane on flight deck except in very calm conditions.

Finally, landing on a moving flight deck requires predictive intelligence. You basically have to guess where the deck is going to be in the future, combine that with estimated wind gusts, and then calculate in the future motion of the aircraft given a range of factors. Putting all this together is currently way beyond the ability of artificial intelligence systems.

There are actually a lot of things MUCH simpler than landing on a flight deck that we cannot do. For example, if the Air Force could just land UAVs in heavy cross winds that would be a huge win for the military. Currently, if the cross wind component is above a certain amount we cannot even land UAVs because they would crash. This is because the UAV cannot react fast enough to a wind gust the way a real pilot can. We lose hundreds of millions of dollars in UAVs every year because of simple stuff like this, and remember these are manned system that are only semi-automated. Fully automated aircraft don't even exist, because they would crash so much as to be much more costly and dangerous than useful.

There are three main reasons: reaction time, sensor inadequacy, and predictive intelligence. Taking them in order...

For certain types of stimulus-response activities a human can respond faster and has a better reaction time than a computer. For example, imagine it takes 50 milliseconds to sense something completely, 100 milliseconds to transmit that information to the computer, and 100 milliseconds for the computer to figure out the right action, and 80 milliseconds to transmit the response to an actuator. That's a total of 330 milliseconds. A human can respond to an event in 200 milliseconds. Note that for some types of actions the computer's sense-and-respond loop could be much slower than the numbers I have given. For example, most artificial intelligence algorithms are complicated and take a lot longer than 100 milliseconds to complete. Some types of optimization algorithms, like dynamic programming have unpredictable running time, so they might finish in say 50 milliseconds with one set of inputs and 1000 milliseconds in another.

The next big problem is that the human eye, which is wired in parallel directly to the brain, is a far better sensor than anything available to an aircraft. It has a wider field of view, higher bandwidth and better focusing and movement ability than any man-made optical system. Currently the sensors on planes are things like inertial measurement units, GPS and flight instruments (altimeter etc). These sensors are not enough to reliably land a plane on flight deck except in very calm conditions.

Finally, landing on a moving flight deck requires predictive intelligence. You basically have to guess where the deck is going to be in the future, combine that with estimated wind gusts, and then calculate in the future motion of the aircraft given a range of factors. Putting all this together is currently way beyond the ability of artificial intelligence systems.

There are actually a lot of things MUCH simpler than landing on a flight deck that we cannot do. For example, if the Air Force could just land UAVs in heavy cross winds that would be a huge win for the military. Currently, if the cross wind component is above a certain amount we cannot even land UAVs because they would crash. This is because the UAV cannot react fast enough to a wind gust the way a real pilot can. We lose hundreds of millions of dollars in UAVs every year because of simple stuff like this, and remember these are manned system that are only semi-automated. Fully automated aircraft don't even exist, because they would crash so much as to be much more costly and dangerous than useful.