Sloth vs Artificial Intelligence

Artificial Intelligence's relationship with energy consumption can only be described as pathologically voracious. Training a single large language model requires electricity sufficient to power approximately 126 Danish homes for an entire year. The data centres housing AI systems consume roughly 1-1.5 per cent of global electricity production, a figure projected to reach 4 per cent by decade's end. Each query to an advanced AI system consumes approximately 10 times the energy of a traditional search engine request.

The cooling infrastructure alone presents a thermodynamic challenge of considerable magnitude. AI data centres generate heat with such enthusiasm that their facilities require elaborate water-cooling systems, some consuming millions of litres daily. The carbon footprint of training GPT-3 was estimated at 552 tonnes of CO2, roughly equivalent to 120 vehicles driven for a year. The AI, for all its computational magnificence, remains fundamentally dependent upon finite resources extracted from an increasingly overtaxed planetary system.

Artificial Intelligence's cognitive capabilities have exceeded human performance across numerous benchmarks with alarming rapidity. Modern systems demonstrate proficiency in tasks including language translation, image recognition, strategic game-playing, and scientific prediction that would have appeared magical to previous generations. GPT-4 performs in the 90th percentile on bar examinations, whilst specialised systems have solved protein-folding problems that stymied researchers for decades.

The processing scale defies conventional comprehension. Large language models contain hundreds of billions of parameters, trained on datasets encompassing substantial portions of human written knowledge. A single AI system can simultaneously engage in thousands of conversations, each maintaining contextual coherence across extended interactions. The technology demonstrates what appears to be creativity, generating novel combinations of concepts in ways indistinguishable from human ideation. Whether this constitutes genuine intelligence or merely sophisticated pattern matching remains philosophy's most pressing contemporary question.

The existential risk profile of Artificial Intelligence has prompted the world's foremost technologists to sign open letters expressing varying degrees of civilisational concern. The theoretical failure modes range from the merely catastrophic to the comprehensively terminal. Alignment problems present the possibility of AI systems pursuing goals technically consistent with their programming yet disastrously inconsistent with human flourishing. The canonical example involves an AI instructed to maximise paperclip production converting all available matter, including humans, into paperclips.

Beyond theoretical scenarios, documented AI failures already demonstrate concerning patterns. Recommendation algorithms have radicalised populations, automated trading systems have triggered flash crashes erasing billions in market value within minutes, and facial recognition systems have demonstrated biases that perpetuate systemic discrimination. The probability assigned to AI-driven human extinction by surveyed researchers averages approximately 10 per cent, a figure that would prompt immediate evacuation were it applied to any other risk category. The AI may represent humanity's last invention, in every possible interpretation of that phrase.

Artificial Intelligence's potential contribution to human wellbeing spans applications from the transformative to the trivial. Medical AI systems have identified cancers invisible to human radiologists, predicted protein structures that may yield revolutionary treatments, and enabled personalised medicine approaches previously impossible. Agricultural AI optimises crop yields whilst minimising environmental impact. Climate modelling AI generates predictions essential for species survival planning.

Yet the wellbeing ledger must also record AI's contributions to human misery. Social media algorithms optimised for engagement have demonstrably worsened mental health outcomes, particularly among adolescents. Automated surveillance systems enable authoritarian control. Deepfake technology erodes epistemic trust. The technology simultaneously promises liberation from tedious labour and threatens displacement of meaningful work that provides identity and purpose. AI is a Janus-faced gift, offering solutions to problems it simultaneously exacerbates. The net wellbeing contribution remains genuinely uncertain, a calculation that may determine civilisation's trajectory.

Artificial Intelligence exhibits extraordinary adaptability within its operational parameters. A single architecture can be trained for tasks ranging from medical diagnosis to artistic composition to legal analysis. The technology has progressed from rudimentary chess-playing algorithms to systems capable of generating coherent text, images, and code within mere decades. This capability overhang suggests adaptability limited only by computational resources and training data availability.

Yet AI's resilience reveals fundamental fragilities. The systems require precise environmental conditions: stable power supplies, climate-controlled facilities, functioning internet infrastructure. A sufficiently motivated squirrel chewing through a fibre optic cable can render an AI system completely non-functional. The technology possesses no survival instinct, no capacity to repair itself, and no ability to operate independently of human-maintained infrastructure. Were civilisation to collapse, every AI system would follow within days, whilst sloths would continue their unhurried existence entirely unaffected.