The human tendency to assign personality to objects or machines goes back millennia, with people anthropomorphizing their ships, swords, or cars. For centuries, people have been naming objects, talking to them, and attributing individual quirks to them. It’s poetic, but imaginary. Giving something a name and talking to it does not make it human.
Attributing personality to a device is different from a device actually having a personality, and this has become a problem in discussions around AI. Maybe it’s just the headline writers, but AI does not “escape” or “go rogue.” Instead, it’s an engineering problem.
Look at self-driving cars; they’re imperfect, but improving. They make mistakes when they stop suddenly, run over pedestrians, or are unable to select what set of instructions to execute in a power outage. But no one says a Tesla “went rogue” when its program fails to recognize a stop sign, because these terms imply intent.
Machines don’t have intent. They have instructions. When those instructions are poorly written or not fully thought through, they can cause errors. But these errors are not intent.
Intent would be a machine independently deciding to take an action to serve its own interests. This has not happened. Programs use probability to identify a course of action that best fulfills their programmed objectives. Sometimes it designs a course of action and then writes the code to implement it. But it is still ultimately carrying out human instructions.
Implying intention satisfies deep-seated human urges. First of all, it’s the kind of scary story that humans have told themselves for centuries about robots, golems, or haunted statues. Second, intentionally or not, it reassigns liability from coder to rogue AI. It’s like saying, “I didn’t misprogram the AI tool; it went rogue, so I’m not responsible.”
People love illusions and often willingly suspend disbelief. AI can give the illusion of personality and intent, which may be easier to understand than complex coding errors. The ability to assign weights in different AI models only increases the illusion of personality.
The problem is that this points policymakers in the wrong direction. It can increase fear and the sense that AI is uncontrollable. AI’s mistakes are either coding errors or the absence of rules to guide how code is written. Some people, like Rob Joyce, have suggested that we copy Asimov’s Three Laws of Robotics. Creating safety standards for programming AI tools would be helpful and would give people a way to measure error and assign liability.
Intentionality also has implications for cybersecurity. It’s not that rogue AI tools are out there plotting to hack your network; it’s either that someone has written imperfect code or intentionally aimed the tool at you. AI cyber-risks are the result of long-standing problems: credentialing and authentication, misconfiguration, segmentation of networks, and failing to patch flaws or fix coding errors. These issues create a poor cybersecurity environment. AI tools provide a better and faster way to take advantage of it. Rogue programs are not the problem.
Writing “company accidentally miscodes AI tool” is much less compelling than saying AI went rogue — but it has the virtue of being accurate. The next time you see a story or hear an interview where someone talks about AI as if it has intent, they are either pulling your leg or being intellectually lazy. If you do talk to your brick and it talks back, don’t ask AI what to do. AI is a tool designed and built by humans, still imperfect, and they are responsible for its actions.
James Lewis is a Distinguished Fellow on the Tech Policy Program at the Center for European Policy Analysis.
Bandwidth is CEPA’s online journal dedicated to advancing transatlantic cooperation on tech policy. All opinions expressed on Bandwidth are those of the author alone and may not represent those of the institutions they represent or the Center for European Policy Analysis. CEPA maintains a strict intellectual independence policy across all its projects and publications.
Tech 2030
A Roadmap for Europe-US Tech Cooperation
