Control Engineering

Charlie Masi

How intelligent are computers, really?
May 14, 2007

As philosophical as this question sounds, it has real relevance for control systems, as we start asking them to run our infrastructure, control our processes, and even park our cars. It will become even more important in the near future as we attempt to have unmanned aerial vehicles (UAVs) share airspace with recreational aircraft, and automatic guided vehicles (AGVs) share our roadways.

What we’re talking about is not compute speed or memory capacity. Intelligence is more about the ability to find effective strategies to deal with novel situations.

The story goes that Sir Isaac Newton decided to enter a mathematics contest to solve what is called the brachistochrone problem. In a nutshell, the brachistochrone problem asks: “What shape does a track have to be so that a ball rolling down it from Point A to Point B under the influence of gravity will make the trip in the shortest amount of time?” It’s still used as an example in advanced physics classes. He won the contest by solving the problem in (if I remember correctly) 24 hours.

Sir Ike, being arguably the brightest of history’s brightest, didn’t attack the problem directly, as his competitors were doing. He realized that the available mathematics was inadequate to the task. Instead, he spent the first 20 hours inventing a new branch of mathematics called the Calculus of Variations, which is so powerful that one can solve the problem in about an hour using it. He then spent the last 3 hours checking his work and writing up a letter that said, basically, “I won!”

Today, physics students regularly spend a day or so learning the technique so that they, too, can solve the brachistochrone problem in about an hour.

While computers can solve the brachistochrone problem in about a millisecond, there ain’t a one of ‘em who can invent the calculus of variations.

So, the strategy I propose is to examine our question by comparing what computerized systems can deal with in the way of novel situations with what living organisms have to deal with.

If we start with your typical “supercomputer,” which regularly calculates the flow dynamics around supersonic aircraft, all it has to do is solve the Navier-Stokes equations over and over realfast! Its biggest brain-teaser really is planning how to break the novel geometric problem it has been given up among its array of processors. That’s about equivalent to a virus particle figuring out how to take over the metabolic machinery of a cell it’s invaded.

So, we have our first result:

Supercomputer = Virus (and a stupid one at that)

Things get more difficult for computerized systems that have to react to real-world situations in real time. A supervisory control and data acquisition (SCADA) system running a chemical processing plant, for example, has to react to upsets in perhaps a dozen or more measureables, such as ambient temperature, tank levels, and so forth. That’s about what a bit of phytoplankton floating around in the Sargasso Sea has to deal with. So, our second result is:

SCADA system = single-celled plant

Adding motion makes a big step up the evolutionary ladder. If we look for the computerized system that has to move through the most unpredictable environment with the least amount of outside help, I suggest a Mars Rover. Nobody really knew what those puppies would find when they got to the “office,” and they have to get themselves from Point A to Point B with the nearest human help many light minutes away.

That sounds like what a single-celled animal, such as a paramecium, deals with every day. So, our third result is:

Mars rover = paramecium

Slime mold cells, on the other hand, spend part of their lives cooperating together to form what’s called a fruiting body, which actually moves as a unit to find a less-crowded habitat to colonize. So, vehicles being designed to compete in the DARPA Urban Challenge, where they have to avoid bumping into each other as well as stationary obstacles, is probably the nearest computerized system. Note that no vehicle has yet successfully met the DARPA Urban Challenge, although the winner may have already been built. So, our fourth result is:

DARPA Urban Challenge vehicle = slime mold (on a good day)

According to this analysis, we’ve exhausted the abilities of the smartest computerized systems available, and haven’t even reached the level of multicellular organisms. You may disagree and, if you do, I’d be interested in hearing your thoughts in a comment to this posting.

Posted by Charlie Masi on May 14, 2007 | Comments (1)