AI needs understanding

Human intelligence has proven a challenge to characterize, including to what degrees it is fixed or variable, inherited or nurturable, and unitary or multiple. In action, it might look like the deft distillation of information in context to arrive at efficient explanations and understandings, leading to reliable (dependable) predictions and effects.

In areas such as science and mathematics, theories or models with these traits are described as elegant, connoting seemingly effortless style and class in other contexts.

For most of us, artificial intelligence is a black hole containing mysterious processes while inhaling all available microchips, water and energy.

Despite the prowess and ingenuity behind AI, it is certainly not elegant from the perspective of energy. Perhaps the internet universe, like our own, leaves little choice if wanting to enter into it. Nevertheless, of the tremendous amounts of electricity AI consumes, most (about 90 per cent) is wasted in the form of heat loss. (See “Waste heat utilization of data centers based on heat pump technology from the perspectives of supply and demand: an overview,” at science.direct.com, which addresses how this heat might be used.)

For perspective, a gasoline-powered car converts in the order of a quarter (estimates vary) of the energy derived from the gasoline it burns into useful motion, the rest wasted in the forms of heat and wear and tear. Electric vehicles are substantially more efficient. (This does not account for lost energy associated with generating, transmission, transportation, etc., of the energy source.)

This brutish grind and energy intensity of AI data centres should, alone, give pause to their explosive development as well as to some applications of AI, many ranging from frivolous to pernicious. Their draw on resources and the direct or indirect additional atmospheric carbon load will also drive further growth in the imaginatively exculpatory carbon capture industry.

According to seemingly credible online videos, AI algorithms harness morsels of phrases, words, sounds, images and videos appearing on the Internet in the binary 0,1 language of computers.

Through exhaustive iterations and mathematical modelling, statistical probabilities are assigned to these morsels as next in a sequence of output designed to be satisfying to the user relative to the prompt. More apt monikers for it might be probabilistic pandering or predictive appeasement. These outputs may be nuggets of high value, or fool’s gold with unfortunate or tragic consequences when users mistake it for certainly credible and human(e)ly considered.

The relentless pursuit of AI capacity has wealthy players scrambling to mark out lucrative territory with some governments’ enthusiastic support.

Proponents and supportive governments promise, as usual, jobs and positive progress which, also as always, will not result in long-term benefits because the enterprise is premised on opportunism and exploitation, rather than on equity and sustainability.

Threats to cybersecurity, focus on the darker arts, harmful impacts on individuals and the growing dilution of instincts and criteria for credibility and reality are other reasons for greatly improving our understanding of AI and what it delivers, and for adopting a disposition of extreme caution over its expansion.

The AI train is gathering speed and weight. Existing control switches and signals will soon be inconsequential, and the tracks on which it rides will, in its current form, need ever more massive expenditures of energy and environmental capital along with compromises to accessible sustainable energy supplies.

There will be leapfrogging threats to cybersecurity on which we all depend, and to well-being in ways seen and yet to be seen, as described by Devi Narayan in “AI ownership is not security” (Free Press, July 11).

The Internet universe that we are creating is worth exploring, beyond surfing, for what can be uncovered to our benefit. It is crucial that our leaders and lawmakers deftly distill information in context to understand and then discern how AI is allowed to expand and evolve.

At the moment, it’s far from elegant.

Ken Clark, retired after career stops in industry, education and government, writes from Winnipeg.