Maine is the first state in the Union to employ rank-choice voting (RCV) for a Congressional office, in this case Maine’s Representative from the 2nd District. The end result was a close race in which the incumbent, Bruce Poliquin (R) had a lead, but not a definitive lead, and as the process of RCV unfolded, his lead disappeared and ultimately his opponent, Jared Golden (D) was declared the victor.
Poliquin has cried foul, and along with filing suit against the use of RCV (approved twice by the Maine electorate), he’s also asking for a recount, citing some interesting reasons, as reported by the Press-Herald:
“We have heard from countless Maine voters who were confused and even frightened their votes did not count due to computer-engineered rank voting,” said Brendan Conley, a spokesman for the Poliquin campaign.
Frankly, I don’t have a lot of sympathy for this group of voters, if they even exist. The process is simple to use, and I assume this change in voting has been well-advertised. The informed citizen should have shown up at the polling place with choices made and ready to fill in boxes.
“Furthermore, we have become aware that the computer software and ‘black-box’ voting system utilized by the secretary of state is secret. No one is able to review the software or computer algorithm used by a computer to determine elections,” Conley said. “This artificial intelligence is not transparent. Therefore, today, we are proceeding with a traditional ballot recount conducted by real people.”
This is where things get interesting, and a bit disingenuous. Why? Because the voting machines supplied for the traditional “1 person, 1 vote” style of elections come from private manufacturers, and, at least last I heard, the software was also a private, uninspectable affair. The resulting suspicions I discuss in the thread starting here.
Frankly, the good Representative may not want to stray into the swamp of voting machine politics, where revelations concerning the nature of voting machine software might lead to vast embarrassments for the GOP brand.
But there’s more to unwrap here.
This spokesman mentions ‘artificial intelligence’, and at this juncture I’d like to say that I’m becoming more and more convinced that this term should not be applied to any entity which lacks volition, or at least is intended to have volition, failed or not. That is, if your software entity is only intended to do is, say, detect whether or not someone has cancer by examining an X-Ray after having been trained on a collection of X-Rays, then I find it difficult to classify this as AI. Really, this is Machine Learning (ML), in which the entity has learned a set of rules and applies them.
Off of my rant stool and back to the story, the incumbent Representative and his spokesman have cleverly attempted to slip something by the reporters. When they mention that artificial intelligence [or ML] is not transparent, this is not a lie. Much of ML is opaque to everyone, from us common folks to the designers who designed the system and the programmers who wrote it. Let me defer the why of that for a moment, because it plays into my objection to their statement.
And that objection is that neither AI nor ML should be involved in this operation (and, if it is, someone needs a good smack upside the head)!
RCV is not a difficult problem to solve, at its core. The real problems are in security and transparency (see links above).
But let’s briefly discuss why I’m asserting this with such certainty, despite no real relevant experience in ML.
When a programmer is given a task to solve, typically the steps that we’re encoding for the computer to follow are either well-known at the time of the assignment, or they can be deduced through simple inspection, or they can be collected out in the real world. An example of the last choice comes from the world of medicine, where early attempts at creating a diagnosis AI began with collecting information from doctors on how to map symptomology to disease diagnosis.
These steps may be laborious or tricky to code, either due to their nature or the limitations of the computers they will be run on, but at their heart they’re well-known and describable.
My observations of ML, on the other hand, is that ML installations are coded in such a way as to not assume that the recipe is known. At its heart, ML must discover the recipe that leads to the solution through observation and feedback from an authority entity. To take this back to the deferment I requested a moment ago, the encoding of the discovered recipe is often opaque and difficult to understand, as the algorithms are often statistical in nature.
And this is not RCV at all. It has well-understood steps that lead to the final result. There’s no secret to it. In fact, the recount will be by humans, not by computer, so that proves the point.
So when the losing side complains about AI and it not being transparent, don’t be fooled. They may have legitimate worries about security and hacking by malicious entities, or even bugs (sigh), but the core algorithm should not be of an ML or AI nature. They may not realize it, but that’s how this really all plays out.
