We are hearing a great deal about "artificial intelligence" these days. It's an attractive topic for speculation because it can mean a number of different things, and there really is no consensus on how to define "intelligence" in the first place. Some very simple machines can exhibit behavior that mimics intelligence.
Very few people would suppose that the centrifugal governor on a steam engine -- which dates from the end of the 18th century -- actually is thinking, although an unsophisticated spectator watching it control the speed of the engine might regard its purposeful action as the result of thought. A few other people, by contrast, consider that human thought is essentially mechanical, and that steadily improving technology therefore will permit the development of machines that can do it, too. The rest of us are willing to wait and see.
It seems to us that vending machines usually have been ahead of the curve in terms of behavior that mimics intelligence. The most sophisticated mechanical coin discriminators could identify, separate and validate several (usually three) coin denominations, recognize each, add up their values and enable a vend. They did this by applying Newtonian physics, sometimes supplemented by a magnet or two. We don't recall anyone ever asking, "How does it know to accept this dime while rejecting that penny?"
Another perennial ability of vending machines that has grown more sophisticated over time is customer communication. The little red "sold out" tab that popped up near the actuating knob on a closed-front machine was an early example. And we recall being impressed, half a century ago, by a postmix cold drink machine under a staircase at a New York City subway station (this was before the Transit Authority banned vending from its platforms). The vender was designed with three small horizontal windows stacked above the coin entry. These consisted of strips carrying transparent lettering on an opaque ground, mounted in front of small filament lamps and behind a piece of frosted glass. The uppermost window was illuminated from behind while the machine was not in use; it invited passersby to "insert coins here." Those who accepted that invitation and initiated a vend saw that message wink out and the middle window light up to explain that the machine was "preparing your drink," which told them not to reach into the cup well just yet. When the vend cycle was complete and the cup was full, that message blinked off and the bottom window lit up to say "thank you." This all was accomplished without the use of a microprocessor; our guess is that three switches did the job, but their sequential action might look a lot like "intelligence" to a naive observer.
The advent of solid-state electronics and its application to vending gave rise to "self-diagnostic" circuits to aid in troubleshooting, and ultimately to work with a telemetry system informing the operator of a machine fault, perhaps before it's noticed by anyone in the location. And, as we grapple with computers in our daily activities, we believe the time is ripe for the next step: the self-repairing machine, or one that is able to restore minimum functionality while it awaits the arrival of a technician.
Computer users are familiar with error messages, more often than not reporting that an "exception" has occurred at some address, and so the machine is going to shut down -- which it proceeds to do. This has not changed much from the days of MS-DOS, regardless of all the claims for transformative disruptive innovation. We are not particularly impressed by the ability of "agents" like Apple's Siri or Microsoft's Cortana to respond to a spoken inquiry, perform an Internet search that we might have performed for ourselves, and report the results, e.g. nearby pizzerias.
What would impress us would be an "agent" in touch with a very robust monitor program and able to appear when necessary, to report that an error had occurred -- and the agent had taken care of the problem. That would be much closer to our idea of "artificial intelligence" than a database search routine which, discovering that you have purchased a toaster, sets in motion a series of processes that show you lots of ads for toasters.
Vending machines actually do better than most devices in this regard. Today's positive vend assurance circuitry has eliminated nearly all the suspense of making a purchase from a glassfront machine by monitoring the vend right through the arrival of the product in the delivery bin and, if this does not happen, initiating procedures to dislodge the refractory item or, if those fail, to encourage the customer to make another selection.
Those who compare today's machines with the equipment available even two decades ago may not see how the machines themselves can be much improved. The hope now seems to be for increasing customer engagement, and some of the ideas we've seen for this look promising. We do believe, however, that no worthwhile progress will be made with "artificial intelligence" if it's not designed and implemented by people employing a sufficient quantity of the genuine article.