More artificial than intelligent

Even small, incremental changes, says Z, result in massive increases in complexity:

The company that started out as two people, but grew to one hundred people, is at least a thousand times more complex than when it started. Obviously, the small town that grew into a city seems infinitely more complex than when it started. Even your social circle can suddenly feel wildly complex if your circle of friends expands to include people outside your initial peer group. Complexity grows at a rate faster than the growth of the organization. That’s an iron rule of life.

The people working in artificial intelligence are running into this same problem. Replicating even the most mundane human task requires millions of lines of complex code. What we take for granted as humans is actually quite complex. For the same reason no one person can understand the complexity of their small town, the creators of AI cannot understand the complexity of their creations. Algorithms to handle one small task get unwieldy in a hurry once they are interfaced with other algorithms to handle other small tasks.

This is why the robot future is a lot further away than the futurists want to believe. The cost of labor to automate a warehouse is a grain of sand on the beach, compared to the cost and complexity of automating a highway. Just as important, the cost of maintaining it is orders of magnitude higher. As every business owner knows, just because something can be automated, the cost of doing so often outweighs the savings. Put another way, just because something can be done does not mean there is a reason to do it.

And even if there is a reason to do it, hidden costs inevitably emerge. It is an article of faith in the fast-food industry that an automated order-taker is ultimately less costly than giving some youngster $15 an hour.

Until the customer says “Hold the mayo.” Okay, you can do that. But suppose he wants an extra pickle too? As the number of choices goes up, the cost of handling them also goes up — only faster.


  1. Francis W. Porretto »

    6 September 2018 · 12:33 pm

    Which is why instead of human-scale AI, many developers are aiming at what were once called “expert systems:” systems designed to do one thing properly and nothing else at all.

    The order taker robot at the Golden Arches can cope with a great many options if it controls the means and therefore the range of input. Therefore, the customer will not say hold the pickle, extra mayo; he’ll tap options widgets on a touch-sensitive screen. It’s the same sort of logic that made the GUI a better and more reliable interface than the command line: the user’s degree of freedom is delimited by the program, rather than the program having to cope with whatever input the user, bless his spiny little heart, might care to type in.

    Complexity in an automated system is often merely a matter of not having canalized the user suitably.

  2. McGehee »

    6 September 2018 · 2:48 pm

    Considering how order-takers at McDonald’s use those registers for inputting orders, I sincerely doubt their kiosks are any more complicated except for the customer.

    And those of us who’ve been using the self-checkout at the supermarket have a leg up.

RSS feed for comments on this post