The stratospheric growth of the AI economy seems unsustainable and headed for a significant crash. No commenter on this topic brings more receipts than Ed Zitron, whose June 30 article, “The AI Industry is Losing,” continues his years-long tirade. Reasonable minds may disagree as to the scale and imminence of a collapse, but to me, some retraction seems inevitable. When that happens, many companies in tech and beyond will suffer, as will the retirement and savings accounts of everyday Americans, and for that matter individuals and businesses around the world as the shockwave hits.
I want to say up front that if one were to list the 50 most important consequences of this, effects on consciousness study would not make the cut. Nevertheless, I’m writing about consciousness, and it remains an interesting and I believe important long-term question to study, so let’s ask: what does the coming economic retraction mean for the study of consciousness? I think it brings us simultaneously a step back, a step sideways, and a step forward. (Don’t try to visualize. It’s just metaphor.)
First, the step back: money is going to be more scarce, which will have consequences for any research into consciousness that requires AI compute, including my own. Not only will there be less money to spend, but the resources needed to do the work will become expensive at the same time, and everything will become more difficult. It’s funny, in a way, because money is an artificial construct, a societal agreement on a shared token that facilitates the exchange of goods and services and supports long term investments in the means of production. (Caveat: while I may pantomime the language a bit, I am not remotely a scholar of capital or labor.) Nevertheless, looking at the big picture, there are times when money is cheap and times when money is expensive, and we’re in sort of a “medium” place now (after an extended “cheap” phase), and likely headed to an “expensive” one. And that’ll cost us. (Pun intended.)
Practically speaking, there will almost certainly be less interest in, and less money going towards, the use of resources to study questions without near-term practical relevance. And while I continue to believe, as I have written before, that consciousness and safety are convergent matters, I won’t claim urgency in the study.
Next, the step sideways: there will still (always) be universities, and the liberal arts, and the non-GPU-powered dimensions of inquiry into these questions. We can look at what has happened, and what we’ve learned, and what people think about it, with non-digital tools and methods of study (and even non-AI digital ones, actually). Consciousness conversations still present a huge gap between those who believe that artificial consciousness in current systems is impossible, and those who believe the incredible growth we’re on will deliver something that we universally accept as conscious any day. Cutting off that growth curve may let us have a more grounded discussion about what kinds of things would allow us collectively to determine consciousness. (And, uh, I have some thoughts on this, if anyone is wondering.)
In my opinion, we could benefit from a few decades of liberal arts-centric inquiry into the techno-political-socioeconomic changes of recent years to arrive at a greater shared understanding and collective wisdom around them and their intersections with humanity. Maybe pausing development and economic growth for a while will be good in the long run. (Shout out to the University of Puget Sound here in Tacoma for its AI & Human Values Initiative.)
Last, the step forward: a crunch will give us some space from LLM-specific investments in particular, and perhaps a little more intellectual appetite for creative new directions. This is a bit of a thin silver lining, I know. But Gary Marcus, among others, has been professionally perturbed for years at the disproportionate level of investment in a single, specific direction for AI technology, and that money-driven bias may evaporate quickly. To the extent that LLMs have significant roadblocks in climbing higher up the curved plane of consciousness—something I argue in Chapter 5 of my book—moving to new algorithms and approaches might help significantly.
For my own research directions, I think this could create more interest in concepts like abstraction and resilience. Abstraction offers the potential for more efficient data and memory handling; and resilience lets us get more out of the systems we have, and assume less disposability. Focusing on a longer-term horizon in every sense offers some sensibility and some reason, and little about the current environment feels particularly sensible or rational.