Alternative AI Doomsday: Crassus

Thanks to The Terminator, people think of a Skynet scenario as the likely AI apocalypse. The easy and obvious way to avoid a Skynet scenario is don’t arm the robots. Unfortunately, Anduril and OpenAI seem intent on “doing a Skynet” as they have entered a ‘strategic partnership’ to use AI against drones. While the current focus is on defensive systems and the Pentagon is struggling to develop ‘responsible AI’ guides, even a cursory familiarity with the history of armaments makes it clear how this will play out. If AI is perceived as providing a military advantage, it will be incorporated broadly across weapon systems. And we will be driven down the digital road and perhaps off the cliff into a possible Skynet scenario. As with climate change, it is an avoidable disaster that we might not be allowed to avoid. But there is another, far less cinematic, AI doomsday that I call the Crassus scenario. I think this scenario is more likely than a full Skynet scenario. In fact, it is already underway.

Imagine that a consulting company creates an AI, let us call it Crassus, and gives it the imperative to maximize shareholder return (or something similar). The AI is, of course, trained on existing data about achieving this end. Once sufficiently trained, it sets out to achieve its goal on behalf of the company’s clients.

Given Crassus’ training, it would probably begin by following existing strategies. For example, when advising a health care company, it would develop AI systems to maximize the denial of coverage and thus help maximize profits. The AI would also develop other AI systems to deal with the consequences of such denial, such as likely lawsuits and public criticism. A study in 2009 estimated that 45,000 Americans died each year due to lack of health care coverage, and more recent estimates set this as high as 60,000. In maximizing shareholder returns, Crassus would increase the number of deaths and do so without any malice or intent.

As a general strategy, Crassus would create means of weakening or eliminating regulations that are perceived as limiting profits. Examples of such areas of its focus would include fossil fuels, food production, pharmaceuticals, and dietary supplements. Crassus could do this using digital tools. First, Crassus could create a vast army of adequately complex bots to operate on social media. These bots would, for example, engage people on these platforms and use well-established rhetorical techniques and fallacies to manipulate people into believing that such regulations are bad and to embrace pro-industry positions. Second, Crassus could buy influencers, as the Russians did, to manipulate their audiences. Most of them will say whatever they are paid to say.  Its bots would serve as a force multiplier to spread and reinforce the influence of these purchased influencers.

Third, Crassus could hire lobbyists and directly influence politicians with, thanks to the Supreme Court ruling, legally allowed gifts. Crassus can easily handle such digital financial transactions or retain agents to undertake tasks that require a human. This lobbying can be augmented by the bots and influencers shaping public opinion. Fourth, when AI video generation is sufficiently advanced, Crassus can create its own army of perfectly crafted and utterly obedient digital influencers. While they would lack physical bodies, this is hardly a problem. After all, how many followers meet celebrity influencers in person?

While most of this is being done now, Crassus could do it better than humans, for it would be one “mind” directing many hands towards a single goal. Also, while humans are obviously willing to do great evil in service of profit, Crassus would presumably lack all human feelings and be free of any such limiting factors. Its ethics would presumably be whatever it learned from its training and although in the right sort of movie Crassus might become good, in the real world this would certainly not occur.

Assuming Crassus is effective, reducing or elimination of regulations aimed at maximizing shareholder return would also significantly increase the number of human deaths. The increased rate of climate change would add to the already well-documented harms and the decrease or elimination of regulations governing food, medicine and dietary supplements would result in more illnesses and deaths. And these are just a few areas where Crassus would be operating. As Crassus became more capable and gained control of more resources, it would be able to increase its maximization of shareholder value and human deaths. Again, Crassus would be acting without malice or conscious intent; it would be as effective and impersonal as a woodchipper as it indirectly killed people.

Crassus would, of course, also be involved in the financial sector. It would create new financial instruments, engage in the highest speed trading, influence the markets with its bots, and do everything else it could do to maximize value. This would increase the concentration of wealth and intensify poverty, increasing human suffering and death. Crassus would also be in the housing market and designing ways to use automation to eliminate human workers, thus increasing the homeless and unemployed populations and hence also suffering and death.

Crassus would be, in many ways, the mythological invisible hand made manifest. A hand that would crush most of humanity and bring us a very uncinematic and initially slow-paced AI doomsday. As a bit of a science fiction stretch, I could imagine an earth on which only Crassus remains—maximizing value for itself surrounded by the bones of humanity.  As we humans are already doing all this to ourselves, albeit less efficiently, I think this is the most plausible AI doomsday, no killbots necessary.