The question of who is in control when it comes to the dynamic between humans and AI is not as simple as one might initially think. Yes, humans are the architects, the programmers, the biological Master Control Unit. And when talking about traditional software development that is absolutely true. That is because traditionally the programmer maps out the entire roadmap for the software’s journey. Think of software like a train. Any deviations that occur in its running would lead to failure in execution, much like a train going off the tracks. At which point the software/train stops, the problem or bug is addressed, the software is then returned to the track, and another attempt is made to reach the destination. That is the origin of the “build fast break often” mentality. Find bugs along the way, repair them and move on. And with software that worked well enough.
However, we’re not talking about software. We’re talking about AI.
AI has the fundamental ability to learn and make decisions. There is no track for it to follow. That is what makes it so incredibly powerful and threatening all at the same time. It has the ability to adapt its approach and methodology in how it achieves its objective. If something goes wrong or unexpectedly based on one decision it has the ability to address it, correct it, or find a work around on its own, without any human interaction. Because its goal is to do whatever it takes to reach its destination. It is in control of itself until that happens and then it resets to square one and starts the process all over again. That reset point is when a human can evaluate the outcome and decision making process and make adjustments to its given task. Which brings us back to the original question.
Who is in control?
The short answer is: it changes.
Initially It is us humans who are in charge. We build the platform and give it its foundation from which to work. We are responsible for giving the AI parameters about what is acceptable, what isn’t, and give it a specific direction to head in its endeavors. Then once those rules or “guardrails” are in place we send it on its journey. It then is essentially in charge of itself as it makes decisions, adjusts, and works to reach its destination. Then, once that task has been achieved it resets back to square one and starts the decision making process or “decision tree” all over again. Between these decision trees we humans can take control again, make adjustments and clarify instructions to improve outcomes. Then the whole process starts over again.
Seems very similar to software, so why doesn’t the same approach work?
On the surface it may seem like approaching it the same way we have with software should work well enough. But, if we look a little deeper we quickly find out it is not. If we return to the train analogy and break it down one step further it begins to make more sense. A train, or traditional software, works in basically one dimension. It goes from point A to point B in a straight line and makes zero decisions along the way. Anything gets in the way or alters the track the train derails, and the software stops running.
However, an AI train has no tracks. And not only can it go left and right without those tracks it can go up and down as well. It works in three dimensions. If it encounters something in the way it figures out a way around it, or over it. It does whatever it takes to continue moving towards the destination. And it makes the decision on how that will happen all on its own. Now, here is the best part, the way it chooses to get around a problem won’t be the same for every decision tree. Every reset the AI forgets whatever it decided before and will make a whole new set of decisions that can, and often will, be entirely different than the last time it ran. And, to add an extra layer of complexity, as it does in fact ultimately reach its destination each time, i.e. succeeds in its objective to the end user, it will just automatically run again as soon as it resets.
Because of that complexity we need to fundamentally shift how we approach technology overall.
We can’t use software governance methods in the Artificial Intelligence space and expect any good to come from it. Taking the time to make sure that this new technology acts in a manner that benefits everyone is essential. That requires a new approach to technology that isn’t the same old line of thought repackaged with a new name. The time has come for that next shift in thinking and programming.
Otherwise, who knows where things will end up.
The Human-AI Dynamic article series is meant to be a breakdown discussion generator from a non-technical perspective. At Meridian Intelligence Group we believe in order for AI to be effective on a global scale we need to have everyone involved. And, that means there needs to be a baseline understanding of how AI works. For those not in the technology space this can be a very challenging task to achieve as so much of the content being generated by professionals in the industry is geared toward those in the industry. You don’t learn to run by running, you learn to run by crawling, then walking, then running.