How Long Does It Take To Aim a Baseball?
A meditation on Artificial Intelligence
A.I. detractors point at actions that machines cannot do - A.I. proponents say "just you wait". Both argue with simplified models, and are merely proving the inadequacy of their models, not anything useful about reality.
As far as I know, no artificial intelligence has played in a major league baseball game. One of many activities is throwing a baseball - to the right place at the right time in order to win the game. I am no ball player - I cannot hit the broad side of a steradian - worse, I don't care much for baseball. But many of the ball players who really care about baseball are really really good at throwing. Which makes them the best ball throwers in the galaxy.
A machine can propel a baseball fast. No doubt there are ball-throwing machines that can be automatically aimed. To date, none can do much of what a major league pitcher can do. Arguably a machine will be able to go through the motions in the near future. Perhaps a few decades later on, a machine will be able to do so strategically. But it will be quite a while before an artificial pitcher can "read a batter" and judge how to aim the ball in the best way to strike the batter out. Today, a machine can accomplish this with very high speed motors, sending the ball plateward so fast that the batter can't see it. But so could a mechanically augmented human pitcher. In either case, the high speed ball will kill the catcher, which means she will drop the ball, and the batter walks to first base (?). There is probably an official rule against killing catchers with ultrafast balls. This is not strategic.
So, lets focus on the title question. Since no machine baseball game pitchers exist today, we need not focus on what a machine might do, but on what humans - the only existing pitchers/catchers/base(persons)/shortstops in 2015 - can do. How long does it take to aim a baseball? There are (at least) 8 time periods involved.
100 microseconds
The shortest time period is the processing of the intent (to put a hard-to-hit ball through the strike zone) into a large, coordinated set of muscle movements, mediated by neurons. William Calvin teaches us about the surprisingly high accuracy needed to deliver a high speed ball to a small distant target - his example is hitting a rabbit at 14 meters distance, which requires open loop timing accuracy of 100 microseconds, formed by an intent occuring long before the fingers open to release the ball. A strike zone is bigger than a rabbit, but a baseball moves much faster than a stone aimed at a rabbit. Since nerves are jittery, with perhaps 10 milliseconds of timing noise, a LOT of signal averaging and a LOT of neurons are needed to fire all those muscle twitches at exactly the right times.
3 hours
The length of a major league baseball game. A pitcher is not an automaton; she is both a strategist and a tactician, as is everyone on the field. She has studied the opposing team on video, in her imagination, and in prior encounters, hours, days, and years before a particular game. But every game is different, and while she is evolving her strategy, minute by minute, to thwart the opposing batter and team, the batter is evolving his strategy to thwart her and her team. In a three hour game, there is plenty of time to learn and adapt.
20 years
The pitcher did not just walk in off the street; she has spent most of her life obsessed with baseball, watching it, playing it, practicing it. She probably was playing catch with her parents as soon as she could stand up. She learned more in little league, high school, and college. And she spent years on the team getting better and better, tutored by coaches and other players. She has probably aimed a baseball about ten million times, every time laying down vast numbers of neural circuits devoted to every kind of pitch she might ever expect to perform. Beyond the mechanics of striking out a batter, she must also be ready to throw the ball to a base(person) to strike out a player attempting to steal a base (did I get that right? What about tackles, putts, and free throws? 
). She must have a vast quantity of rehearsed moves in her brain, which takes decades to put there. Fortunately, brains are capacious, and they also share neurons amongst many related kinetic programs. The axons and dendrites and synapses in her brain are an elegant baseball decision map, a physical structure far too complex to describe with any programming language extant. Most importantly, this is a vast parallel machine, implementing millions of parallel computations at once, and accurately selecting the one computation she will implement with her muscles. Oh, and she must also cross streets, sign baseballs, pay taxes, and find a mate to produce more baseball players with, assuming she cares about the future of the game.
2 million years
She is on the mound because all of her ancestors found good mates, and produced children who found better mates. Homo erectus was the first animal capable of throwing accurately. Larger brains produced more signal averaging accuracy and thus wider range and more successful hunting. A very strong driver to evolve the twenty watt power hogs we perch above our necks. Calvin's throwing story is plausible - only humans can throw with such accuracy - but the paleo diet was mostly plant based, few hunts resulted in surplus calories, on average the energy cost of the brain was wasted on gathering. But dexterous picking of small fruits also required fast hand-eye coordination, and survival depends on gathering more faster than the competition. In times of drought and famine, catching one animal might prevent starvation. In any case, her ancestor's ability to throw required a large brain, and the large brain was fed by the throwing.
190 million years
Mammalian learning is plastic, and dependent on specific neural learning processes not found in any other animal family. Birds come closest functionally, but use different synaptic mechanisms. Scientists are still arguing about the mechanisms that produce long term potentiation in humans (I favor Gary Lynch'es kinase-modified dendritic spine synapse reshaping model), but this is not simple "selection from a pallete of evolved behaviors" like Kandel's Aplysia sea slugs. But then, Eric Kandel has a Nobel Prize and I don't, so don't take my word for it. In any case, mammalian learning started with the olfactory bulb and the sense of smell, building odor maps of the world that helped our mouse-like ancestors navigate between food and nest in the warm tunnels underneath the frozen ground caused by the Meteor Winter following the Chixhulub impactor. Without situational learning, dependent only on wired instincts, our ancestors would not have survived in order to evolve into human baseball players.
600 million years
It is difficult to throw anything underwater - animals had to move from water to land, and learn how to survive there. That learning was genetic; our ancestors were fixed-program machines. Sorta like IBM's Watson, which can "learn" to make clever Jeopardy answers from the facts and algorithms it was fed beforehand, but will need a new set of facts and algorithms to play Wheel of Fortune. Humans are the computer's evolutionary environment, Successful software gets more tweaking (by humans), and failures are fossilized on backup tapes.
2.1 billion years
Multicellular life - the first team players, and the first time life forms learned strategy, developed tiny brains and reactive behavior. We also learned about sex, and formed primitive wired responses like "girls can't throw". Which actually was true for billions of years, until the evolution of girls.
4 billion years
The emergence of life from the collaboration of autocatalytic molecules. These almost-creatures did not aim, but the ones pointed in the wrong genetic direction lost to the ones pointed in the right direction by sheer dumb luck. 4 billion years later, we still rely too much on dumb luck, like a ball thrown from the outfield that misses the infielder it is thrown to, and goes to another infielder who, after an extra second of game evolution, is who the ball actually needs to go to. The sports reporters will praise the insightful move, and the lucky and canny outfielder will keep her mouth shut and accept the salary-boosting acclaim. Nature rewards the combination of fortune and prepared mind more than she rewards one or the other in isolation.
Back to A.I.
Nature is prolifigate, with atoms, energy, and situations. Nature uses massive parallelism followed by selection to discover novel solutions to problems - our brains recapitulate this, as do the immune systems that keep us alive (and coincidentally, develop from the same stem cells as neurons). So do our societies and families and organizations, all aimed at survival and supremacy. This vast process, layered in information and space and time, seems simple and natural because it is essential for our existence, and we presume (without evidence) that the whole multilayered structure can be projected onto machines with a finite amount of effort. The more ambitious A.I. advocates presume that machines can duplicate these processes independently, and more successfully without the tether to natural processes. While such A.I. exceptionalism cannot be disproved, there is more evidence for flying saucers.
A more plausible scenario is that multilevel physics, chemistry, biology, and humanity will enhance and be enhanced by machines, forging new connections between isolated past levels and to the future levels that we (from insects to Intel) collaboratively evolve. The future will have less "them" and more "us", boundaries dissolving between the atoms and bits, protein and transistor, microscopic and galactic, femtoseconds and gigayears. Whatever is valuable and shared will thrive; connectivity is far more productive than contention. The future will have a place for autistic dweeb A.I. advocates, but the universe will not evolve into a single gigantic autistic dweeb A.I. Optimally coupled parallelism will always outperform single cores, as it always has. Slow and vast always outperforms quick and tiny; otherwise, a single gigawatt microbe would rule the planet - until it vaporized femtoseconds later.
Intelligence isn't speed - it is the efficent solution of important problems by the reliable elimination of inconsequential detail. Humans and machines together can collaboratively combine history-laden domain competencies to define importance, reliability, and consequence, then acting as a team, process a whole universe of detail. We will do best if neither arm, machine or human, is tied behind our backs.