1. [para. 1][para. 2][para. 3][para. 4][para. 5] The author asserts that AI will not become green on its own. [para. 2] The core argument is that developing AI is not an end in itself; the true goal is sustainable human development, meaning "AI for good" must first be "AI for green." [para. 1] A 2024 UN General Assembly resolution unanimously adopted a rare focus on safe, secure, and trustworthy AI for sustainable development, while the 2024 UNCTAD Digital Economy Report marked a sharp pivot, questioning the future trajectory of the digital economy. [para. 3][para. 4][para. 5]
2. [para. 6][para. 7] The environmental costs of AI are staggering. By 2026, global data centers are projected to consume 1,000 terawatt-hours of electricity, roughly double France's total 2022 consumption. [para. 6] Manufacturing a single laptop requires 800 kilograms of raw materials, and demand for critical minerals like lithium and cobalt is expected to surge fivefold by 2050. [para. 6] Furthermore, training a single large language model requires millions of liters of water. If left unchecked, AI will fuel a "black economy" characterized by resource depletion and runaway carbon emissions. [para. 7]
3. [para. 8][para. 9][para. 10][para. 11][para. 12][para. 13][para. 14][para. 15][para. 16] Developing green AI requires walking on two legs. [para. 8][para. 9] The first leg is the "greening of AI" itself through optimized chip architecture, algorithmic models, and training strategies. [para. 10] Encouraging breakthroughs include an MIT research team demonstrating autonomous driving with a brain-inspired network of merely tens of neurons [para. 11] and companies like DeepSeek boosting efficiency via innovative training strategies. [para. 12] Ultimately, AI must run on green, renewable energy. [para. 13] The second, more critical leg is trading AI's energy consumption for massive savings across other industries. [para. 14][para. 16] Examples include smart grids enabling millisecond-level fault diagnosis, precision agriculture minimizing ecological footprints, and smart manufacturing reducing energy consumption by 8% to 12% in energy-intensive industries. [para. 15]
4. [para. 17][para. 18][para. 19][para. 20][para. 21] A desperate need exists for a new "intelligent transformation services" industry to bridge the widening structural gap between frontier AI models and industrial application. [para. 17][para. 18][para. 20] This gap arises because AI providers lack understanding of specific sector demands, while traditional industries are unfamiliar with AI capabilities. [para. 19] These service providers must possess deep expertise in both AI and niche industry scenarios, such as understanding the distinct process characteristics and pain points of fine and bulk chemicals. [para. 20][para. 21]
5. [para. 22][para. 23][para. 24][para. 25][para. 26][para. 27][para. 28][para. 29] Two specific recommendations are offered to accelerate the shift. [para. 22] First, drive green AI through standardization, specifically a "green token" system. A token is the smallest semantic unit in AI and the pricing basis for services. Embedding green metrics into token metadata—such as the proportion of green electricity consumed, carbon emission intensity per unit of computing power, and hardware recycling rates—would create market mechanisms to drive the greening of AI from the bottom up. [para. 23][para. 24] Second, integrate green AI into international cooperation on AI capacity building, ensuring the global trajectory aligns with sustainable development goals, as highlighted by a China-spearheaded UN resolution. [para. 25] The author, Gong Ke, concludes that AI is at a critical crossroads; the choice between a greener future and resource degradation is ours, but the window of opportunity is limited. [para. 26][para. 27][para. 28][para. 29]
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新