Problems of Literalism in Machines

If people who have a strong tendency to take things literally are alarmingly easy to exploit in the current world, the same flaw is also affecting machines.

I wrote last month about the commercial exploitation of people who take things and especially words and assertions at face value. These are people who are slow to detect exaggerations or check factual assertions, and advertisers and others are responding to this tendency by supplying lies and misdirection on an unprecedented scale, especially online.

In some ways, the machines that run our online world are even more prone to this error of literalism. It’s a problem we have known about for along time. In 2002, a person who posted my music online could evade copyright filters by intentionally misspelling my name as “Rick Astr.” Every human would know who they meant, but the machines searching for unauthorized posts would not.

Machines have gotten better at detecting simple misspellings, but they still make hilarious errors context. I realized how big a flaw this could be when I wrote the parody song “Give the Violins Back to the Violinists.” I never released this song to the public, and only a couple of people ever saw it, but Google saw it, and for months afterward, I was seeing online ads for cellos and violin repair services. It would not occur to the Google targeting algorithm that a use of a word might be metaphor, rhetoric, or fiction.

The same flaws affect people who use a word in one of its less common meanings or contexts. A person who ordered Mars candy bars or planned a visit to Jupiter, Florida would be selected for space travel content. The person who bought a Yamaha harmonica might find herself seeing ads from dealers and competitors of Yamaha motorcycles.

The mistargeted ads and content are glaring and visible to all of us, but the problems of literalism in machines go much deeper. Other problems are less visible but more serious in their implications. The suppression of false information claiming that viruses do not exist also swept away important information about false results from tests for viruses, and the suppressed information covered both biological viruses and computer viruses, with machines not always able to separate one topic from another. The whole computer security topic of spoofing covers ways to defend against patterns that exploit machines’ tendency toward literalism. These exploits can lead to the loss of data and other more serious consequences.

Machines make decisions much faster than humans. One machine might make a million decisions in a second. This makes the errors of literalism by machines a problem on a different scale. In the end, though, it is still another version of the same problem that is found in the exploitation of people who are too ready take things at face value.

Fish Nation Information Station | Rick Aster’s World | Rick Aster