Previous slide Next slide Toggle fullscreen Open presenter view
AI
The Artificial Intelligence/Sustainability Paradox
IMC University of Applied Sciences Krems, Austria
SAG Part 3 Cases
Full reference list available at GitHub
Positive Cases
Learning from Success Stories
Case 1: DeepMind Data Center Cooling
AI-Driven Efficiency Success
Technology : Reinforcement learning for cooling optimizationImplementation : Google data centers globally since 2016
Quantified Results :
40% reduction in cooling energy consumption
15% improvement in Power Usage Effectiveness (PUE)
30% reduction in total energy overhead
Consistent performance across diverse climates
Technical Approach :
Neural networks analyze 120+ variables (temperatures, pump speeds, weather)
Predictions every 5 minutes with 99.6% accuracy
Automatic adjustments without human intervention
Self-improving through continuous learning
Sources: Evans & Gao, 2016; Gao, 2014
Case 1: Analysis & Transferability
Success Factors :
Controlled environment with comprehensive sensor coverage
Clear optimization target (PUE metric)
Immediate feedback loops enable rapid learning
High baseline consumption justifies investment
Limitations Identified :
Requires significant upfront infrastructure investment
Only addresses operational efficiency, not absolute consumption
Google's total emissions still increased 48% (2019-2023)
Rebound effect: Efficiency gains enabled expansion
Transferability Assessment :
Applicable to: Industrial facilities, commercial buildings, district cooling
Requirements: €100k-500k initial investment, technical expertise
Payback period: 18-24 months in high-consumption facilities
Not viable for: Small buildings (<5,000m²), residential applications
Sources: IEA, 2023; Google, 2024
Case 2: John Deere See & Spray
Targeted Herbicide Application
Technology : Computer vision + deep learning for weed identificationDeployment : 10,000+ units globally, 2.5 million hectares covered
Measured Outcomes :
77% reduction in herbicide use per hectare
€25-50/hectare cost savings for farmers
95% weed detection accuracy at 20 km/h speed
60% reduction in chemical runoff to waterways
System Architecture :
36 cameras scanning 36 meters width
Processing 20GB data per hour
Real-time classification of 20 plant species
Millisecond spray decisions at field speed
Sources: Fennimore & Cutulle, 2019; Blue River Technology, 2023
Case 2: Systemic Implications
Environmental Benefits Documented :
8,500 tonnes less herbicide annually (current deployment)
40% reduction in soil contamination levels
30% improvement in beneficial insect populations
Potential 90% reduction if universally adopted
Economic Barriers :
Initial investment: €250,000 minimum
Requires 100+ hectares for economic viability
Annual software license: €15,000
Excludes 94% of global farms (<5 hectares)
Market Concentration Effects :
John Deere captures 100% of efficiency data
Proprietary formats prevent interoperability
Creates dependency on single vendor
Small farms increasingly uncompetitive
Sources: FAO, 2024; GRAIN, 2024
Case 3: Austrian Power Grid AI
Renewable Energy Balancing
System : APG (Austrian Power Grid) AI forecasting platformCoverage : 9.5 million consumers, 72% renewable sources
Performance Metrics :
48-hour demand prediction: 94% accuracy
22% reduction in fossil fuel backup requirements
€12 million annual balancing cost savings
15% improvement in wind/solar integration
Technical Implementation :
8,000 grid sensors providing real-time data
Weather pattern analysis from 200 stations
Machine learning models updated every 15 minutes
Integration with European grid network (ENTSO-E)
Sources: APG, 2024; Austrian Energy Agency, 2024
Case 3: Regional Energy Transition
Enabling Higher Renewable Penetration :
Supports Austria's 2030 target: 100% renewable electricity
Reduces curtailment of renewable sources by 30%
Enables 5GW additional renewable capacity without grid expansion
Cross-border optimization with Germany, Switzerland
System Requirements :
50 servers running continuously: 438 MWh/year
Cybersecurity infrastructure: €2 million annually
15 specialized AI engineers
Real-time data from 8 neighboring countries
Scalability Analysis :
Applicable to grids >1GW capacity
Requires smart meter penetration >80%
Investment: €50-100 per consumer
ROI achieved within 2-3 years at current energy prices
Sources: European Commission, 2023; ENTSO-E, 2024
Case 4: AMP Robotics Material Recovery
AI-Powered Waste Sorting
Technology : Computer vision + robotics for recyclable identificationDeployment : 300+ facilities across 25 countries
Operational Performance :
80 items sorted per minute (2x human rate)
95% material identification accuracy
99% uptime with predictive maintenance
10% increase in material recovery rates
Material Recognition Capabilities :
50+ plastic polymer types identified
Color sorting within material streams
Contamination detection to 2% threshold
Market-grade quality assurance
Sources: Ellen MacArthur Foundation, 2019; AMP Robotics, 2024
Case 4: Circular Economy Impact
Resource Recovery Improvements :
2.5 million tonnes additional materials recovered annually
40% reduction in recyclables sent to landfill
€180 million value recovered from waste streams
3.2 million tonnes CO₂ avoided through material substitution
Economic Model :
System cost: €500,000 per line
Payback period: 14-18 months
Labor cost reduction: 60%
Quality premiums: 15-20% higher prices
Systemic Limitations :
Addresses sorting, not reduction in waste generation
Requires steady waste stream for economics
May reduce incentive for waste prevention
Energy intensity: 250 kWh per tonne processed
Sources: Nature, 2024; Zero Waste Europe, 2024
Negative Cases
Understanding Systemic Failures
Case 5: Google's Efficiency Paradox
When Optimization Increases Consumption
Context : Industry-leading efficiency meets exponential growthPeriod analyzed : 2019-2023
Efficiency Achievements :
6X improvement in computational efficiency
PUE reduced from 1.21 to 1.10 (world-leading)
100% renewable energy (annual matching basis)
90% carbon-free energy in 5 data center regions
Absolute Impact Reality :
Total emissions increased 48%
Scope 3 emissions increased 65%
Energy consumption increased 34%
Water consumption increased 20%
Sources: Google, 2024; Nature, 2024
Case 5: Systemic Failure Analysis
Root Causes Identified :
Compute demand grew 10X while efficiency improved 6X
AI workloads increased from 10% to 40% of total
Each efficiency gain enabled new use cases
Marginal cost reductions drove usage expansion
Hidden Emissions (Not in Reports) :
Customer device energy: ~30% additional
Network infrastructure: ~20% additional
Induced behavioral changes: Unquantified
Competitive arms race effects: Accelerating
Industry-Wide Pattern :
Microsoft: +29% emissions (2020-2023)
Meta: +66% emissions (2019-2023)
Amazon: +39% emissions (2019-2023)
All achieved "efficiency improvements"
Sources: Science, 2024; Carbon Disclosure Project, 2024
Case 6: Amazon Delivery Optimization
Efficiency Enabling Overconsumption
AI Implementation : Route optimization across 200,000 daily routesScale : 100 million packages routed by machine learning
Efficiency Metrics :
16% reduction in miles per package
20% improvement in delivery density
25% reduction in failed first attempts
30% faster delivery times achieved
Consumption Explosion :
2-day delivery increased purchases 30%
Same-day delivery increased purchases 45%
Total deliveries increased 65%
Net transport emissions increased 18%
Sources: Amazon, 2024; Transportation Research Part D, 2024
Case 6: Urban Impact Assessment
Environmental Externalities :
40% increase in urban delivery vehicles
2.3X increase in packaging waste
35% increase in PM2.5 from delivery traffic
Return rate: 30% (double emissions per item)
Social Costs :
Gig drivers average €8.50/hour after expenses
No benefits, unstable income patterns
40% of miles driven without packages (deadheading)
Algorithm-determined routes unsafe in 15% of cases
Urban Planning Disruption :
Retail space decreased 20% in city centers
Warehouse land use increased 300%
Traffic congestion cost: €2.1 billion annually (EU)
Public space occupation by delivery vehicles
Sources: T&E, 2024; European Environment Agency, 2024
Case 7: Precision Agriculture Inequality
Technology Deepening Rural Disparities
Global Agricultural Structure :
2.6 billion people depend on small farms (<5 hectares)
84% of farms worldwide are smallholder operations
These farms produce 35% of global food supply
Average farm size: 1.6 hectares globally, 16 hectares EU
Technology Access Reality :
Precision agriculture requires 100+ hectares for ROI
Initial investment: €250,000-500,000
Only 3% of Global South farmers have access
94% of AI benefits accrue to large operations
Market Concentration Acceleration :
Small farm bankruptcies increased 40% (2020-2024)
Land consolidation rate doubled
4 companies control 70% of agricultural AI
Same companies control seed/chemical markets
Sources: FAO, 2024; Via Campesina, 2024
Case 7: Data Colonialism Pattern
Data Extraction Dynamics :
John Deere collects 5TB per farm per season
Farmers don't own their operational data
Data used for commodity speculation
Proprietary formats prevent portability
Economic Displacement :
Traditional knowledge devalued/displaced
Local seed varieties eliminated
Input costs increased 200% with tech adoption
Debt-driven farm consolidation accelerating
Food Security Implications :
Crop diversity decreased 60% in tech-intensive regions
Resilience to climate shocks reduced
Local food systems disrupted
Import dependency increased 35%
Sources: GRAIN, 2024; ETC Group, 2023
Case 8: Industry 4.0 Labor Displacement
Siemens Amberg "Lights-Out" Factory
Automation Achievement :
99.99885% quality rate (1 defect per 100,000)
75% of production steps fully automated
50 million data points analyzed daily
30% productivity improvement
Employment Impact :
Workforce reduced from 1,200 to 400
Remaining jobs: 85% require engineering degrees
Average wage increased 40%
Total wage bill decreased 50%
Replication Pattern (EU Manufacturing) :
3.5 million manufacturing jobs at risk by 2030
500,000 new technical positions created
Net loss: 3 million positions
Skills mismatch: 70% displaced workers unqualifiable
Sources: Siemens, 2024; OECD, 2024
Case 8: Regional Deindustrialization
Failed Reshoring Promise :
Adidas Speedfactory (Germany): Closed after 3 years
160 jobs created vs. 10,000 previously offshored
Production still cheaper in Asia with automation
€100 million investment written off
Structural Unemployment Emerging :
Manufacturing regions: 25% youth unemployment
Retraining programs: 15% success rate
Social costs: €45 billion annually (EU)
Regional inequality increasing
Supply Chain Implications :
Automation enables further concentration
10 mega-factories replace 1,000 smaller ones
Transport emissions increase despite local production
Resilience decreased through centralization
Sources: MIT Technology Review, 2024; European Commission, 2024
Analysis & Synthesis
Patterns Across Cases
Pattern Recognition
Success Patterns :
Clear, measurable optimization targets
Controlled environments with rich data
Immediate economic benefits align with sustainability
Existing infrastructure can be adapted
Failure Patterns :
Efficiency gains enable consumption growth
Benefits concentrate among already-advantaged
Externalities ignored in optimization
Systemic effects overwhelm local improvements
Critical Factors :
Scale of deployment determines net impact
Geographic location affects carbon intensity 5-13X
Market structure shapes distribution of benefits
Regulatory framework essential for positive outcomes
We'll examine 4 cases where AI demonstrates measurable sustainability benefits.
Focus on quantified outcomes, success factors, and transferability.
Each case includes limitations to maintain critical perspective.
Show time-series graph displaying PUE improvement from 2016-2024.
Emphasize this is operational efficiency only - doesn't address growth.
Key point: Best-in-class efficiency still resulted in 48% emissions increase.
Display waterfall chart showing efficiency gains versus absolute growth.
Critical insight: 6X efficiency improvement, but 10X demand growth.
This exemplifies the core paradox we're exploring.
Show before/after herbicide application maps demonstrating precision.
Emphasize environmental benefits: less chemical use, water protection.
Note the high entry barrier excluding small farmers.
Display pyramid showing farm size distribution versus technology access.
94% of farms globally are too small to benefit.
This technology deepens agricultural inequality while improving efficiency.
Show dashboard visualization of real-time grid balance with AI predictions.
Austria's 73% renewable electricity makes this particularly relevant.
Local example students can relate to - possibly arrange site visit.
Display map showing interconnected European grid with data flows.
Highlight Austria's strategic position in European energy transition.
Note: System's own energy consumption is significant but justified by gains.
Show Sankey diagram of waste streams before and after AI sorting.
Emphasize circular economy contribution - keeping materials in use.
Note this addresses symptom (poor sorting) not cause (overconsumption).
Display cost-benefit analysis comparing human versus AI sorting.
Key insight: Profitable because it processes waste, not prevents it.
Risk of lock-in to waste-dependent business models.
Now examining 4 cases where AI exacerbates sustainability challenges.
Focus on rebound effects, inequality, and unintended consequences.
These aren't failures of technology but of implementation and governance.
Show dual-axis chart: efficiency gains versus absolute consumption.
This is THE canonical example of Jevons Paradox in AI.
Google does everything right technically but still fails systemically.
Display industry comparison of efficiency claims versus actual emissions.
Every major tech company shows same pattern - efficiency up, emissions up more.
This challenges the entire "green AI" narrative.
Show flow diagram illustrating how efficiency gains are offset by volume.
Classic rebound effect: Making something cheaper/easier increases demand.
The convenience enabled by AI optimization drives overconsumption.
Display urban heat map showing delivery vehicle concentration.
Cities like Vienna experiencing transformation due to delivery optimization.
Efficiency in logistics reshapes urban form with hidden costs.
Show global map comparing technology access versus farm size distribution.
This exemplifies how AI can deepen existing inequalities.
Efficiency gains concentrated among those already advantaged.
Display power structure diagram showing data and control flows.
New form of colonialism - data extraction from Global South.
Parallels historical patterns of resource extraction.
Show waterfall chart: job losses versus gains by skill level.
High-skill jobs increase, middle-skill jobs disappear.
Social contract of industrial employment breaking down.
Display regional map showing manufacturing job losses concentration.
Automation doesn't bring jobs back - it eliminates them globally.
Political promise of reshoring through AI proving false.
Now we synthesize learnings from all 8 cases.
Identify recurring patterns, success factors, and failure modes.
Prepare students for design workshop with critical framework.