Sunday, 31 May 2015

Freeing humans from redundancy

This has been, and should be, the ultimate goal of mechanizing and automating the world around us. The human mind is too precious of a resource to waste on any types of repeating and repeatable actions, and we've been working on automating the latter since the Industrial Revolution. With modern A.I., more of this is becoming possible.

Consider showing a robot once how to clean the window - defining the physical boundaries, and indicating preferences for the action sequence: you can perform this specification of actions once, the robot repeats this for a specified period of time (e.g. an hour), repeating this sequence at regular intervals (say, once a week). Consider, in such a way, seeding a large variety of actions - watering plants, making mashed potatoes for dinner, tuning the bike, drycleaning your suits, etc. I am not imagining a single robot with the A.I. to do all these tasks (not to mention knowing when to do them) - I am rather imagining an army of simple machines that can be individually programmed by their owners (programming not in the coding sense, but in the show-by-example sense).

You put on your VR (virtual reality) headset while sitting on your hotel bed in SF, you log into your Boston home, seeding a bunch of actions through the FPV cameras on your machines (machine 1 will water the plans on your balcony, machine 2 will set some bread baking, machine 3 will scan some documents for you after finding them in the relevant folders on your shelf). You do the seeding for your cottage country house on Long Island as well. In such a way, not moving off your bed, you have now prepared your Boston home for your arrival tomorrow, and have checked in on your cottage. Here's the critical point: none of these machines has had to be hard-wired for your house or for any of the actions you have assigned them - they simply have the capacity to learn an action and a schedule for it (which does not require any complex A.I. and is completely feasible already). It is up to you to make the difficult human decisions of setting the schedule - when and how much to water the bonsai and the petunias, how long to wait before the bread has just the right crust for your taste, which of your clothes need special attention during dry cleaning, etc. Then the machine executes a repeatable sequence. If a condition arises for which the machine requires a decision to be made, you are pinged. Next time this condition arises, the machine has a stored solution. With time, your army of machines has been customized to all your preferences, and you have been taken out of the loop for anything that does not require an expert opinion (yours) or an indication of specific preferences (yours as well). Your mind becomes freed from anything at all repeatable or redundant, with its capacities available for the decision making and heuristic decisions that are the hallmark of human intelligence. You spend your time delivering instructions and managing outputs with utmost efficiency.

I think we will much sooner see this type of future with simple customizable learning agents than one with the courteous all-in-one robotic butler you see in the movies. In fact, you can already remotely control your house heating via Nest and your music system via Sonos, as two examples, all from your Android devices (cell phone, watch). The next step is simply the augmentation of your control options (from clicks and menus) to actions with as many degrees of freedom as your arm gestures permit via a VR device. This puts a larger portion of your household devices and tasks at your virtual fingertips. The Internet of Things is here.