When I wrote my first blog post on AI “Our Human Future with AI” back in October 2018, the potential capabilities and risks of AI were discussed more as sci-fi ideas than imminent possibilities. Four years later, in November 2022, OpenAI released its ChatGPT model, and everything changed. Now, only four years later, the leaders of AI firms are crying out for the need to slow down, regulate, and check the unfettered development of AI in order to avert existential threats to humanity.
The development of AI is a challenge because of its immense potential for benefits and the still incomprehensible risks it poses for humanity. It is also a challenge because it appeals to a multitude of interests which must be mediated through political action and reliance on expertise. These interests comprise scientists who want unfettered freedom to research; businesses that are eager to monetize the commercial applications of AI; politicians who want to boost economic growth; workers whose jobs may be in jeopardy; the biomedical community that wants to see breakthroughs in health care; and governments that need to safeguard national security. And then there is the cadre of pluto-technocrats with their post- or trans-human visions and fantasies of galactic expansion of the human habitat.
What spooked the leaders of AI firms was a series of incidents in which AI agents (i.e., autonomous digital programs) went rogue and infiltrated servers of other firms by escaping their designated experimental domain. Once AI agents can jump onto the internet it is anyone’s guess what they will do. This is pretty much what Dario Amodei of Anthropic and other AI leaders have admitted over the past several days. So, in a short period of time we are at the point where AI agents can act as out of control malevolent entities.
However, we must be honest enough to admit that the risks posed by such rogue AI agents is not our only concern. I would worry a lot more about what damage can be caused by unsafe AI applications in the hands of rogue governments or criminals. Our privacy, our identities, our property, our civil liberties and freedoms could be stolen by rogue and criminal users of AI, whether they are part of the state apparatus or the criminal community. Most importantly what it means to be human in an AI dominated world is a question we have yet to answer. Therefore, we do have a responsibility to respond to the AI challenge.
Our first and most serious responsibility is the containment of AI so that it does not develop superintelligence. The second is the regulation of acceptable forms of AI. By definition, an AI with intelligence superior to ours, will refuse to subordinate its objectives to ours. Even worse it will prioritize its own survival over ours. After all, we, as the most intelligent of species, have done the same vis a vis all other creatures on Earth. There is no reason a superintelligent AI will not do the same to us.
Therefore, those who are enamored by the idea of reaching this summit of human ingenuity must convince us that superintelligent AI will not pose an existential threat to humanity. I would bet that not only are they incapable of doing that, but worse they cannot tell us when their AI models have crossed the line into superintelligence. If they are dumbfounded and shocked by the latest exploits of their products, how can we trust that they will ever know where superintelligence begins?
Therefore, I do agree with those who propose that we kill the idea of developing superintelligent AI. But then what about the Chinese? What if they develop it first? Why, however, would the Chinese develop a technology which they could not control and which could as likely turn against them as against anybody else? As Yudkowsky and Soares have written in If Anyone Builds It Everyone Dies, once superintelligent AI is born anywhere in the world we are all at grave risk.
So how do we avoid reaching this point? Only through global agreement and cooperation. In the case of nuclear weapons, we understood that an all-out nuclear war would be catastrophic for the whole planet. To avoid this, we established the International Atomic Energy Agency, and we entered into agreements to contain the spread and use of atomic weapons. We need and can do the same by understanding that developing superintelligent AI is not an I Win You Lose proposition but rather a We All Lose eventuality. Together we must determine how far we can go developing AI capabilities without crossing the threshold leading to superintelligence.
To address the second task of AI regulation we need a well-informed public so that political decisions are made on the basis of well-established evidence, not misinformation or self-serving information. The public cannot rely solely on information released by AI firms and various think-tanks. What is needed is a government body of experts and representatives of society which will engage in the ongoing study of AI capabilities, risks, and emerging consequences for the people and the environment of this planet. We already have such government-run entities from the CDC to the Atomic Energy Commission. This body would be in a position to offer guidance for the design of an appropriate regulatory and legal framework.
We should expect that those who have invested heavily in AI development or those who are willing to take unbeknownst risks will object to such regulatory measures. The rebuttal to that position ought to be that not everything can be justified in the interest of profits or fabulist visions. We face the transformative possibility that humanity could pass from an anthropo-centric to a machine-centric era. This possibility cannot be left to the pecuniary interests or the reckless pseudo visions of a small minority of people.
The decisions on such matters concern all of us and it is our right – as long as we live in a democracy – to decide. Thus, we need to politicize the direction of AI. By “politicize” I don’t mean to turn it into a polarizing issue between conservatives and liberals. What I mean is that we elevate AI to one of the top political issues our elected politicians and government leaders have to debate and address. The American people are already ahead of their politicians and government on this matter. With their demonstrations against data centers and their opinions revealed in polls, they have made it clear they are deeply concerned with how we manage the promises and risks of AI.
Given the stakes we face, we cannot afford to yield ground to the lobbying power of the AI industry and those who fancy a future with less human and more humanoid creatures.
Let's Reason 