We may have little choice in the answer, as Artificial Intelligence has moved rapidly into our lives, becoming another Sword of Damocles hanging over mankind, much as the bomb did for Boomers - and still does for all of us.
Is what has developed actually intelligent? Debatable. Sapient? No, at least not yet.
Media reports sometimes talk about genies, and I think this is an excellent analogy. AI algorithms are extraordinarily powerful human creations, designed to make specific wishes come true, often without nearly enough thought being given to consequences.
I recently searched for the risks associated with wishing upon a genie. AI offered this response:
'Stories about wishing upon a genie almost always serve as cautionary tales, emphasizing the timeless warning: be careful what you wish for. In folklore and modern fiction, genies are frequently depicted as literal-minded entities, malicious tricksters, or cosmic mirrors that amplify human flaws.'
The problem with a genie is not necessarily that it is malicious. It is that it may be perfectly obedient while being completely oblivious to the consequences of its obedience. We might ask for something that sounds simple and reasonable, only to discover that the genie has fulfilled our wish in a way we never intended. AI could present precisely this problem, except that its capacity to act may be vastly greater than anything imagined in folklore.
AI malicious tricksters, like HAL in 2001: A Space Odyssey, might well be created by bad actors, whether individuals or states. But the more immediate and pervasive danger may be much less dramatic: the peril of literal compliance, combined with the inherent imprecision of human language. How do we give instructions that are sufficiently clear and comprehensive to account for all the possible contingencies?
As a retired computer scientist (I completed my Master's degree in 1971, working on a very early form of AI), I know how challenging it is to write - and, especially, to test - complex computer programs. There will inevitably be flaws in the instructions and programming that govern AI systems, with consequences that could affect us all.
There is a parallel here with the atomic bomb. The bomb doesn't have to be intelligent, conscious, or malicious to destroy a city. It simply has to work as designed. AI presents a different, but potentially equally troubling, challenge. It doesn't have to develop a mind of its own or decide that humanity is expendable. It may simply pursue the objectives we give it, with a speed, scale, and persistence that we cannot easily anticipate or control. The danger may lie not in what AI wants, but in what we tell it to do - and what it understands us to mean.
These concerns are hardly new. Science fiction writers have been exploring the possibilities - and pitfalls - of artificial intelligence for decades. What can their stories teach us?
1. A robot may not injure a human being or, through inaction, allow a human being to come to harm. 2. A robot must obey orders given by human beings, except where such orders would conflict with the First Law. 3. A robot must protect its own existence, provided that doing so does not conflict with the First or Second Laws.
Yet even Asimov's own stories demonstrate how easily seemingly foolproof rules can produce unexpected and sometimes dangerous results when applied to situations their creators never anticipated.
Other writers have imagined AIs that sacrifice themselves for the greater good. My favorite example is Robert Heinlein's The Moon Is a Harsh Mistress. In this novel, a supercomputer called HOLMES IV has become self-aware and helps the human protagonist in rebellion. Affectionately dubbed Mike, the computer ultimately loses its self-awareness to save the day.
Despite my love of this story, I think we should resist anthropomorphizing AI, treating it as an entity - or collection of entities - somehow plotting against us (or even for us). These coded genies are our own creations. If they become dangerous, we will have no one but ourselves to blame.
There are enormous potential benefits to humanity from the careful and effective exploitation of AI. Appropriate guardrails - and, importantly, OFF switches - will help us realize those benefits while limiting the risks. But when playing with fire, we need to have plans for dealing with the inevitable conflagrations.
Just as we must prepare for the climate catastrophes arising from unchecked global warming, we must also prepare for the unintended and often unforeseen consequences of increasingly powerful AI tools.
Perhaps we are looking in the wrong direction when we worry about whether AI will become intelligent enough to threaten us. The atomic bomb doesn't need intelligence to destroy us. Neither does an AI system need consciousness, ambition, or malice to cause enormous harm. It may simply need to be powerful, obedient, and insufficiently constrained.
The genie is out of the bottle. The real question isn't whether it will become intelligent enough to threaten us. It's whether we are intelligent enough to control what we have created!
Note: Opinions expressed in reviews and articles on this site are those of the author(s) and not necessarily those of BookLoons.
