The future is … problematic
The last few weeks have seen a crisis in confidence surrounding the development of Artificial Intelligence (now called “Super Intelligence” by President Trump). They have also seen a remarkably sanguine response from Trump, who characterizes the concerns as a “sick conspiracy.”
Amidst the Sturm und Drang, it seems useful to step back and offer some verities that get lost in the shuffle.
1. AI is real. AI development has led to amazing growth in productivity in many sectors of the economy. It has driven real capital investments in supporting hardware. And, unlike the banking bust of the late 2000s, it is profitable. Companies like Nvidia are making real money from this, not chasing fake profits with balance sheet manipulations.
2. AI is transformative. It will have significant impacts on productivity, employment, and almost every other aspect of economic activity. It remains to be seen if AI will work the same level of transformation as fire, the printing press, the industrial revolution, or computers – but there can be little doubt that the world 15 years from now will look substantially different from what it looks like today.
3. That having been said, AI is not an existential threat. It is highly disruptive to be sure, and it will change everything from how we wage war to how we do medicine. But in the end, AI is just code. Even self-replicating and self-improving code (the boogeyman fear driving the latest crisis) is, in the end, not as capable of having real-world impacts as, say, a train crash or a nuclear weapon. Humans survived the Ice Age; we will survive this. Which is not to say that AI is not dangerous and without risks – far from it. But humanity is not coming to an end any time soon.
4. Proposals to ban AI development or severely limit it are just smoke and hokum. No technology ever developed by humanity has been successfully eliminated or restrained, in the long term. The Japanese were successful in keeping new firearms developments out of their country for nearly two centuries. But in the end, you can never put the technological genie back in the bottle – and that is as true of AI as any other human technological or social development.
5. Which is NOT to say that all regulations and limits on new technologies are totally ineffective. To the contrary, when humanity unites, it has a pretty good record of placing reasonable limits on how new technology is used and of managing technological transitions. Those regulatory interventions are never perfect, but they are far more effective than the “do nothing” approach in effectively mitigating disruptions. Nuclear non-proliferation has worked reasonably well for the last 80 years. The norms against indiscriminate use of landmines and chemical weapons have been effective enough that the instances of violations can be counted on a single hand.
6. Given the generally low barrier to market entry into AI production (which is much less, say, than producing a nuclear bomb), one can expect AI to proliferate globally. This means that any successful regulation or restriction of AI must also be at least partially global in nature. That, in turn, means that unilateral action by the United States is highly unlikely to be fully effective and may even be counter-productive. Though America is a substantial part of the international economic order and a significant leader of AI technological change, it is not the only game in town.
7. This also means that private-sector self-regulation is, likewise, likely to be ineffective. Market factors drive private actors to favor unrestricted AI development lest their fears of being second-to-market are realized. This dynamic is evident in the major AI firms call for government regulation – they are too disjointed to organize it themselves, even in the United States, much less globally. So, they want governments to intervene and stop them before they do harm.
8. It is not possible, at this juncture, to really know how the Chinese government views AI. At some level, they see it as a threat to party control. At another level they see it as a means of international economic competition (not to mention military advantage) and domestic regulation. Critical to any international approach to AI regulation is the need to appropriately incentivize China to be a willing, active participant.
9. For the most part (though not exclusively, e.g., the landmine ban), much of the international arms control order and economic order that has worked in the last 80 years has been created under the soft-power leadership of the United States. America has, historically, rallied the West and then used Western economic power to propagate its chosen norms globally by persuading and/or forcing other nations (e.g., China) to join with us.
10. America’s capacity to lead such an effort on AI is currently at its nadir. Recent diplomatic initiatives have eroded trust in the United States held by our traditional allies in Europe (the EU mostly) and Asia (Australia, Japan, South Korea, etc.). Our rapprochement with other nations (e.g., Russia and China) has yet to yield significant diplomatic alignment. The prospects of an American-led global approach to AI are therefore near-nil. And while Europe’s efforts to globalize its regulatory approaches have been at least partially successful in other areas (e.g. privacy regulation and technology antitrust), the substantial economic value of AI development is likely to make an EU-led AI regulatory effort less impactful – there is only so much economic influence the bloc can wield.
All of which means that the prospect for AI development over the next 10 years is likely to be one of chaotic transformation. Without global regulatory leadership, the projected trajectory is for increasingly powerful AI models to be developed with only the ethical and moral constraints of the developers to limit them. That is not a cheery prospect. But it is the reality we face and there is little that can be done to change our course.