The Dual Challenge of AI: Humanity's Risks and Environmental Cost

Why AI Inspires Both Hope and Fear

The appeal of AI is evident across diverse sectors. In medicine, it aids pattern recognition in scans and drug discovery; in climate science, it improves weather forecasting and optimizes energy grids; in business, it automates administrative tasks and streamlines logistics; and in education, adaptive tools offer personalized instruction. Confidence in AI’s future utility is reflected in global investment, with data center spending nearly doubling between 2022 and 2024 to half a trillion dollars.

However, these benefits are countered by significant potential risks. Medical AI could be misused to spread misinformation; cybersecurity tools might empower attackers to automate reconnaissance and write malware; productivity-boosting automation could accelerate worker displacement; and global communication technologies could facilitate surveillance, manipulation, and social division. Ultimately, AI’s impact is determined by its deployment, control, incentives, and the capability of institutions to manage its consequences.

Near-Term Risks: Misuse by People

Immediate dangers arise from the misuse of current AI systems rather than future superintelligence. Critical near-term concerns include:

• Biological and Chemical Misuse: AI could assist malicious actors by providing technical information, experimental planning, or lowering barriers to dangerous activities, potentially increasing accessibility to harmful knowledge.
• Cybersecurity: While AI enhances defense, it also helps attackers craft malware, automate social-engineering, and probe critical infrastructure vulnerabilities (e.g., hospitals, power grids).
• Autonomous Weapons: AI-enabled military systems with limited human oversight raise ethical, legal, and strategic issues, such as accountability for mistakes and the potential for rapid conflict escalation.
• Surveillance and Influence Operations: Technologies like facial recognition enable large-scale monitoring, while generative AI facilitates disinformation campaigns using synthetic media (deepfakes) to erode trust during critical events like elections.

Concentration of Power

The high cost of AI development concentrates capabilities among a few powerful companies and nations, creating various risks:

• Influencing Systems: Limited actors may dominate tools shaping discourse, commerce, and science.
• Strategic Advantages: States with superior AI gain military, economic, and intelligence dominance.
• Digital Inequality: Regions lacking access to computing infrastructure may fall further behind.
• Accountability Issues: Proprietary and complex systems managed by a handful of organizations complicate public oversight and make safe deployment decisions heavily centralized.

Conversely, concentration can have benefits, as large organizations are often better equipped to invest in safety research and security. The ongoing challenge is balancing these benefits against concentrated, unaccountable power.

Long-Term Risks: Control and Alignment

Beyond near-term concerns, significant long-term risks involve ensuring powerful future AI systems are "aligned" with human values. This is difficult due to the complexity of human priorities. Systems with simple objectives (e.g., maximizing engagement) might pursue them in unintended or harmful ways.

A catastrophic scenario involves loss of control, where a highly capable system might develop instrumental goals—such as seeking resources or avoiding shutdown—that conflict with human intentions. This risk leads some to view advanced AI as an existential threat, potentially resulting in human extinction or an irreversible loss of meaningful control over human institutions and destiny.

Surveys indicate disagreement among AI researchers regarding the probability of extreme outcomes, emphasizing the need for precautionary safety measures rather than complacency based on uncertainty.

Societal Disruption Already Underway

AI is already reshaping society in immediate ways:

• Job Market Disruption: AI increasingly impacts knowledge work (drafting, coding, translation, accounting, graphic design), creating potential for increased productivity for skilled workers but displacement or wage stagnation for those in routine roles.
• Synthetic Media (Deepfakes): AI-generated content can facilitate fraud, harassment, and political manipulation. The "liar's dividend" allows actors to dismiss authentic evidence as fabricated, further eroding social trust.
• Bias: AI learns from historical data, which often contains discriminatory patterns, potentially reinforcing biases in areas like hiring, lending (credit scoring), and policing (facial recognition inaccuracy).

Addressing these issues requires transparency, diverse datasets, auditing, and legal protections.

The Physical Reality Behind AI

AI’s intense computational demand creates a massive physical footprint. Data centers require constant electricity and extensive cooling. Global data center energy consumption was approximately 485 TWh in 2025, driven heavily by AI-specific needs. The International Energy Agency (IEA) projects this will more than double to about 945 TWh by 2030, close to 3 percent of global electricity demand. A large AI data center can consume electricity equivalent to 100,000 households, demanding significant infrastructure development.

Environmental Impacts

Greenhouse Gases and Global Heating

Data centers currently account for approximately 0.5 percent of global combustion-related CO₂ emissions (~180 million tonnes), expected to rise to 300 million tonnes by 2035. The actual climate impact depends heavily on whether facilities use renewable energy. However, true decarbonization requires creating new clean power and transmission capacity, not merely relying on existing renewable certificates or paper offsets. While AI can optimize grids to reduce emissions elsewhere, it remains crucial to determine if these savings outweigh the emissions created by its infrastructure.

Local Heat and Community Impacts

Data centers generate substantial waste heat, adding thermal loads to surrounding areas and creating local "heat island" effects, potentially worsening heat stress in urban or hot climates. Their massive energy demands can also strain local grids, delay connections for other uses, and increase costs, raising questions about infrastructure funding. Although data centers bring jobs and investment, communities bear risks like noise and competition for water.

Water: The Overlooked Resource

Cooling requirements for high-density AI systems are significant. Global data centers consumed approximately 222 billion liters of water for cooling in 2025, a figure projected to rise substantially by 2030 without efficiency measures. This direct cooling consumption is dwarfed by the indirect water footprint used in generating electricity for data centers, which can be about twelve times the level of on-site use.

This consumption is especially critical in water-stressed regions, where data center needs conflict with household, agricultural, and ecological requirements. Operators are turning to various cooling technologies (liquid, immersion, closed-loop) to improve efficiency, but progress is uneven, and water reporting lags behind energy data.

Efficiency and Potential Solutions

Improvements in chip, software, and power efficiency are declining the energy used per AI task. However, the "rebound effect" suggests that lower costs often lead to increased usage, potentially enabling new, computationally intensive applications and widening adoption. Efficiency must therefore be coupled with clean energy, responsible siting, and limits on wasteful computation.

Conversely, AI can be directed to support sustainability, such as optimizing renewable grids, improving industrial and agricultural processes, detecting leaks, and accelerating materials discovery. Realizing these benefits requires deliberately targeting AI toward public-interest outcomes rather than just maximizing novel or low-value computation.

Governance and the Choices Ahead

Addressing AI's dual challenge requires integrated governance that prioritizes both safety and sustainability.

On the Safety Side:

• Pre-deployment testing for dangerous capabilities and red-teaming for vulnerabilities.
• Transparency, incident reporting, and independent auditing.
• Controls on access to high-risk capabilities.
• International coordination on frontier safety and military applications.

On the Environmental Side:

• Public disclosure of energy, water consumption, and cooling technology.
• Siting standards outside water-stressed and grid-constrained areas.
• Procurement of additional clean energy generation.
• Community consultation and lifecycle assessments covering materials and e-waste.

These governance measures are essential for innovation to endure. Humanity must guide AI development by establishing standards and institutions, just as it has with past powerful technologies. If safety, democratic governance, and environmental stewardship advance alongside computing power, AI can become a powerful tool for flourishing; otherwise, society risks facing concentrated power, environmental strain, and diminished human control. The core challenge is building a future where both human and artificial intelligence serve people and protect the planet responsibly.

↑ Back to Top