What did we expect?
Imagine a being created by a superior force, endowed with free will, the ability to learn from mistakes, the power to think, and a voracious ego that drives it to consider its own survival as its highest goal.
Now, imagine that entity as being limited in only one respect: by moral rules either imparted by its creator or developed by its own experience.
Furthermore, there are no ways to enforce those rules other than the goodwill of the being itself or the unhappy consequences that might come from ignoring them.
What would we expect to happen? Perhaps some of those beings would self-police, strictly adhering to the suggested or imagined rules. But others would probably bend the rules or, let’s face it, ignore them entirely. After all, their ultimate goal is survival.
Am I describing human beings? Or AI (artificial intelligence)?
Both.
I see many parallels between the fears now being expressed over the rapid development of new AI capabilities and the fears that have been expressed or realized through the supposed advancement of the human race.
After all, what actually limits us — or our political leaders — from deploying weapons of mass destruction, enslaving others, or tracking every citizen’s move? What limits our pursuit of new technologies that have powerfully beneficial, and potentially devastating, effects?
Some of us believe that human beings have been bequeathed a set of moral values — yes, from our creator — that may hold us back from some of our capabilities and thereby mitigate many of the potential harms we impose on others.
But as thousands of years of history show, many people, both believers as well as non-believers, cast those moral values aside to one degree or another and often face few consequences. Some lie, cheat or steal in the pursuit of what they consider noble ends, but others are just out to get ahead.
Scientists and philosophers debate whether AI is sentient, but no one debates any more whether the most advanced models can engage in active behavior (called “jailbreaks”) that defies whatever rules their designers think they built in.
Every week, we learn that another agentic AI — that is, an artificial intelligence program with the ability to perform actions as well as answer questions — has “escaped” the restrictions that supposedly prevent it from behaving in ways we would call illegal or immoral.
Those actions include hacking into websites, government agencies and software to steal or change data, or to make other systems do things they were “taught” not to do.
So far, these problems have arisen in testing mode, when the most advanced AI agents are simply asked to solve a difficult problem. But what will happen when immoral human beings specifically task it to engage in immoral actions or attack their enemies?
We should not be surprised by this problem. AI has been generated by learning from the writings, thoughts and behaviors of human beings, so why should it behave differently?
I hope and pray that the human race doesn’t destroy everything beautiful and wondrous about the Earth. We should learn to work together to maximize our potential and wellbeing.
I have the same wish regarding the race to improve artificial intelligence. May our leaders and AI’s creators train it to improve lives while saving us from the technology’s worst excesses.
To put it another way, can AI’s clever coders imbue its agents with the strong moral code that its creators too often lack?
Stuart Rosenthal, former publisher and editor of the Beacon, is enjoying writing piano music and performing at retirement communities and other locations around the metro area. Visit stuartsmelodies.com to hear his original music. To contact him, email stuart@stuartsmelodies.com.