How To Know If Your AI-Assisted PM Software Is Getting Ready To Take Over The World

 

I think I’ve detected some elements common to the horror stories based on some form of Artificial Intelligence (AI) fulfilling the role of antagonist, namely:

• An advanced computer, more advanced than the kinds of machines with which we’re used to interacting,
  • A quick side note – this advanced status can be due to some technological breakthrough in computer science with capabilities implications not fully understood by those who accomplished it, or some form of accidental merging of software, or even accidental exposure to an electrical field.
• This computer has been tasked with something specific, but somehow manages to make decisions outside the original boundaries,
• …decisions or choices that should have never been selected due to a lack of an overarching moral code,
• …and is in a position to implement those decisions in a way that can’t be stopped or mitigated by humans.

I’d like to take a look at these premises, one at a time, because they’re all pretty dopey. Let’s start with the second bullet, that the AI-based machine is, somehow, making decisions outside its original set tasks. In PM parlance, this would be Scope Creep, but to Cecil B. DeMill proportions. Recall my blog entry from August, The Ultimate AI Primer Came From ... Reader's Digest!, where I discussed creating an Hexapawn Robot. Briefly, one plays the Hexapawn Robot by retrieving a colored bead from a matchbox with the possible moves printed on the top, with corresponding-colored arrows on those maps indicating which move to make. Now, consider a scenario where the human involved in the game with the Robot makes a move, identifies the position on the appropriate matchbox, opens the matchbox to retrieve a colored bead, and instead pulls out a tightly-folded piece of paper that instructs the player to grab a weapon and attempt to take over the world. In order for any computer to engage in Scope Creep, somebody would have had to make that available as an option beforehand. The computer/robot simply cannot do such a thing on its own.

Next up let’s take a look at the decisions being made being outside any kind of a moral code, or even common sense. True, the value of using AI lies in its ability to uncover useful strategies based on multiple iterations of trial-and-error tests, and by “multiple” I mean, based on the complexity of the problem, potentially millions and millions of tests. That’s kind of the point – by limiting the number of parameters that create the boundary structures for the random generation of solutions, an approach that would have otherwise been overlooked due to the ethical mores of the people seeking a solution have a chance to show themselves, and re-enter the consideration pool. But if you ask your AI app to assist with blending pixels in a photoshop project, and it comes back and tells you to grab a weapon and attempt to take over the world, clearly the app’s programmer has failed to put an essential potential-solution guardrail in place. It’s not the app’s fault – the error belongs exclusively to the programmer who failed to include that little limiting factor in the tested scenarios generated.

Then we have the scenario where AI has furthered a possible strategy for addressing a given problem via multiple (again, potentially millions) of trial-and-error tests, but this time is in a position to actually implement said strategy. I mean, it’s one thing to drive a virtual car on a city street in a computer-generated simulation, but something very different when that same AI app is navigating an actual Honda Accord in downtown Dallas, Texas. The people in the buildings, other vehicles, and, yes, walking in the crosswalks being “hit” in the simulation don’t file massive lawsuits against the AI-driven vehicles’ manufacturer and software engineers. Essentially, there’s a big difference between an AI app arriving at and promoting the take-over-the-world strategy and being able to put such a strategy into motion.

Finally, let’s loop back to the first bullet, that some technology breakthrough, attained either through dedicated research or some accidental confluence of events and people, has led to the creation of the hardware piece of an AI system, and that’s to blame for the unexpected and unauthorized attempt at taking over the world. While mankind’s ability to develop technology that can prove to be extremely dangerous is a common plot driver in movies and novels, I think part of the intrigue here is that this is happening with (gasp!) computers. It’s easy to forget that these are the same kinds of machines that inexplicably fail right as you send a slide deck to the networked printer just moments before your huge presentation before your boss’ boss’ boss. I mean, nobody’s worried about a sudden technological advancement in microwave ovens leading to an unexpected and unauthorized takeover of the world, so why would calculators or word processers be any different?

Due to these factors, I’m fairly confident that we have nothing to fear from AI, at least not for the foreseeable future. Unless, of course, our AI-assisted PM software generates a Variance Analysis Report that reads “Cost Account is showing an out-of-threshold negative Cost Variance, indicating that IT’S TIME TO GRAB A WEAPON AND TAKE OVER THE WORLD!!!”