AI Fear Part 2: Is It AI or Humans We Should Fear?

I wrote the article AI Fear: Overhype or Valid Concern? just two days ago and I’m already compelled to write a follow-up article. AI fear is accelerating with more politicians jumping on the topic and another resignation, this time from a Google DeepMind employee, Bilal Chughtai, stating “I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome.” [1]. I believe the fear is heading toward hysteria, but are we aiming that fear at the right target? The danger may not be that AI suddenly develops malicious intent, gets catastrophically out of control in the pursuit of a directive or because it got off track. The more immediate danger may be what humans do with AI, what humans already do to each other, and what frightened humans may do in reaction to AI.

In the previous article I separated malicious intent from loss of control and mission drift; I won’t relitigate those mechanisms here.

In this article I want to address real current threats arising from the same human capacity for destruction, which people seem comfortable with likely because they are familiar with those threats and the AI threat is new and feels like prognostications are coming true (or the fear was seeded whenever and however we developed AI). Humans have shown a great capacity and capability for killing one another since the dawn of civilization, so much so that there will be very little need to reference most of what I say throughout this article.

Let’s jump to the dropping of the atomic bombs on Hiroshima and Nagasaki which demonstrated an unprecedented new capacity for destruction by humans. Did it cause us to destroy the technology? Not even close. We improved the bombs and made them much more capable of killing more people in a single act and then we made a lot of them so that collectively we could ensure the true destruction of humanity. However, we have exhibited restraint and have not intentionally used nuclear weapons against other humans since World War II.

However, the larger fear is not that large powerful nations will use a nuclear weapon but that a rogue nation state or terrorist group will get access to these weapons and use them to intentionally kill people (I won’t attempt to address the intent of terrorists). We just honored the 25th anniversary of the 9/11 attack on America last week which did not use nuclear weapons but involved the hijacking (not digital but physical) of airplanes used as weapons against the US. This was real destruction enacted by humans against humans and using our transportation infrastructure as the weapon. Does AI introduce a new threat in this realm? Possibly, if this same group of terrorists had access to AI they possibly would’ve disrupted more planes and/or the air traffic control systems. Terrorism is real and we’ve witnessed bombings of innocent civilians many times in the US and globally over the past half-century or more. This is not a potential but a real threat.

Looking at the disruption of our infrastructure, we have seen many reports of cyber crimes focusing on hijacking computer systems for ransom. This happens to businesses and hospitals. Yes, hospitals whose systems get taken over and frozen, denying access to charts, medications, surgeries and patient care until the ransom is paid. I have seen this happen in my own community and it is my understanding from someone with deep expertise that these cybercriminal groups are so consistent at this that they operate with a code of honor to restore service immediately after receiving the ransom so the next victim knows if they pay systems will be restored. Can AI make this happen faster and more routinely — likely yes, but the fact remains this is already happening and perpetrated by humans.

The stated concern is that recursively self-improving AI could escape meaningful human control. But even the phrase “human control” deserves scrutiny. Which humans? We already know there are people, organizations and governments willing to destroy lives and infrastructure. Human control is not automatically benevolent control.

This creates two different risks. The first is familiar: AI becomes a more capable weapon in human hands, making cybercrime, terrorism or warfare faster and more destructive. The second is genuinely new: an autonomous system improves its own capabilities, coordinates with other systems and gains access to consequential infrastructure beyond effective human oversight. That possibility deserves serious scrutiny. But each step is still a separate link in a causal chain, not a guaranteed outcome.

The AI fear is growing so rapidly and the wave of animosity over data centers so sweeping that I fear domestic terrorism on data centers and big tech/AI company facilities. This shouldn’t be a surprise either. We’ve recently seen corporate CEOs from financial firms attacked and murdered and we’ve witnessed terrorism against doctors’ offices and clinics for decades. Are we just numb to this kind of destruction?

We are imagining a future intelligence becoming dangerous while remaining strangely accustomed to our own capacity for violence. So why are we more frightened by currently hypothetical machine intent than by demonstrated human intent? Perhaps because human violence is familiar. AI violence is unknown. And the unknown is easier to mythologize. We don’t need AI to be the monster in the closet when we are the monster under the bed.

References