AI, Ecological Civilization, and the Difference between Human and Algorithmic Intelligence
A few months ago I shared a post about the water and energy consumption needed to power generative AI tools such as LLMs such as ChatGPT. The post generated a mixed reaction, with many pushing back that the energy and resources required are more than worthwhile considering the productivity gains and the ways this can offset other sorts of environmental effects, our carbon footprint, etc.
I think this is almost certainly true for some intentional uses of generative AI by individuals and organizations. But these technologies are being deployed at staggering scales for all sorts of mindless purposes. Perhaps you, like me, have seen the proliferation of Facebook pages which are fully automated to post massive amounts of AI-generated images to farm engagement. And often when you look in the comments, you see a significant amount of Facebook profiles responding in very predictable, cookie-cutter ways — likely also fake, AI-powered accounts.
This, of course, connects to the “dead internet theory” that much of the content and conversations we encounter online are actually computer generated, with the subsequent engagement with this content also not coming from live human beings.
Obviously, there are economic structures in place which promote the growth of this sort of content-farming paradigm. This obviously isn’t sustainable long term, but there may be a good bit of longevity left here.
So, while I certainly am able to accept the resource cost of AI prompts in cases where the result of the engagement with the generative intelligence is personally meaningful, practically helpful, or productive of novel value, there’s no way I can view the broader proliferation and automation of these technologies in the context of global capitalism as not also constituting a new layer of our collective ecological crisis — and the same can certainly be said about cryptocurrencies.
I was recently watching a video by the popular STEM YouTuber Veritasium, outlining the centuries-long quest of mathematicians to find odd perfect numbers — or to prove definitively that all perfect numbers are in fact even.
It was a fascinating history of a quest for knowledge purely for knowledge’s sake — a beautiful testament to the ingenuity at the heart of the human spirit and the deep, collective desire for knowledge we have as a species.
However, there is a dark side to this aspect of our basic psychology as well. In contemporary times, this search for odd perfect numbers has been augmented by the use of computing technologies. As such, the quest to identify these numbers has now extended into the stratosphere of mathematics, computing numbers will be millions of digits. Of course, to do such calculations requires tremendous computing power, and thus, resource consumption.
At a later point in the Veritasium video Derek, the host of the channel, interviews mathematician Pace Nielson and asks if there is any practical utility to this search. Dr. Nielson answers emphatically in the negative — this is, at least for now, purely abstract curiosity on the part of mathematicians.
Of course, this does not preclude the possibility of some future practical use for knowledge generated in such pursuits, but at what point do we decide that, given our ecological embeddedness, the impacts of energy and resource consumption for such pet projects are simply unjustified or even immoral?
Considering this question alongside the concerns about the ecological impacts of generative AI, I am struck by the difference between human intelligence and the information resulting from engaging with LLM generative AIs or through running the kinds of massive algorithms that can search for specific kinds of numbers so staggeringly large that they are incomprehensible to our human minds.
Algorithmic computation requires energy inputs — independent of whether we are discussing the neural network of a large language model or complex algebraic calculations, these computer systems need an energy source to manipulate data in the ways their architecture dictates in order to produce a result.
The most unique and worthwhile aspects of human intelligence are fundamentally dissimilar to this. Of course, our brain as an organ has huge energy demands: the high-level predictive processing that takes place in concert with our sensory inputs, of course, has a certain metabolic demand. And there are, likewise, more algorithmic and calculative aspects of our own thinking and intelligence — if there were not, human beings would not have invented/discovered algebra, calculus, etc. and then subsequently developed technologies which aid in these investigations. However, this is only one small aspect of human intelligence and far from definitive or exemplary of our rationality and intellection as a whole.
Consider the example of sudden insights — those occasional moments where, at a sudden moment like a flash of lightning, something becomes clear to you. More often than not, these occur at moments of relaxation and ease — while in the shower, taking a casual walk, daydreaming by a window… These are expressly not instances where one’s body (and thus, the nervous system and brain) are under a lot of stress and consuming significant amounts of energy.
This type of insight cascade is fundamentally not algorithmic in nature, and thus comparing human intelligence with that of generative AI in this regard is inapt. This is one reason among many that we should be weary of outsourcing too much of our thinking to these technologies — one will always require resource investment and thus have ecological impacts, while the other is not tied to energy inputs in the same way. This is self-evident, as great luminaries like Einstein or Deleuze or Mullā Ṣadrā or Lao Tsu etc did not require radically more calories to live the sort of lives they did than the average person. To try to reproduce the kinds of contributions these great minds made using generative AI would require a massive energy investment, if it’s even possible in the first place with the limitations of the LLM architecture, which is merely predictively recombining materials from its massive database based on given inputs.
Of course, philosophers have been aware of this long before the advent of these AI technologies. I think here of Heidegger’s discussions of meditative thinking along with many pre-modern ideas of intellect, insight, and wisdom in religious traditions such as Christianity, Buddhism, Hinduism, and Islam. In all these cases, the rational faculty of the intellect is not a merely calculative function, but rather involves the whole person, connecting to our aesthetic and ethical senses to produce a surplus of coherence and meaning. The experiences of intellectual insights in these traditions were often so profound that connections to the sacred seemed self-evident.
In any case, it seems important to revive these broader conceptions of human intelligence, as they are both life-giving and humanizing while also being far more harmonious with our undeniable ecological embeddedness. To outsource too much of our thinking to generative AI, or to remain apathetic towards its environmental impacts simply because it has some obvious utilities, is likely to produce some devastating consequences.
One need not become a complete doomer or a reactionary Luddite in light of this situation, but it does require some real prudence and the exercise of wisdom. The good news, though, is this is something our ecologically emergent faculties of intelligence actually provide us with the possibility of! And while taking up this task is certainly difficult, your personal energy expenditure in doing so does not have any deleterious effects in the broader web of life — quite the opposite, in fact.
