Sustainability

Question 1

What is Sustainability?

Answer 1

Sustainability = “Meeting the needs of the present without compromising the ability of future generations to meet their own needs.”
(United Nations Brundtland Commission, 1987)

Question 2

What are the 3 Pillars of Sustainability?

Answer 2

1) Environmental Sustainability (Planet)
2) Social Sustainability (People)
3) Economic Sustainability (Profit)

Question 3

Explain this pillar of Sustainability: Environmental Sustainability

Answer 3

• Focuses on preserving ecosystems and natural resources for future generations.
• Involves reducing carbon emissions, minimizing waste, and protecting biodiversity.
• Example: Using renewable energy sources (solar, wind) instead of fossil fuels to combat climate change

Question 4

Explain this pillar of Sustainability: Social Sustainability

Answer 4

• Emphasizes the importance of equity, inclusion, and well-being for all individuals and communities.
• Aims to reduce inequality, provide access to essential services (healthcare, education), and protect human rights.
• Example: Creating fair labor conditions and opportunities for marginalized communities.

Question 5

Explain this pillar of Sustainability: Economic Sustainability

Answer 5

• Ensures that economic systems can support long-term growth without negatively impacting
  environmental and social factors.
• Encourages responsible production, consumption, and innovation that contribute to a resilient economy.
• Example: Promoting a circular economy where materials are reused and recycled instead of discarded.

Question 6

What is “The Triple Transition”?

Answer 6

Digital - Green - Social Transition:
On 5 July 2023, one of the first events organised under the Spanish Presidency of the Council of the European Union was titled “Triple Transition: shaping international cooperation to address digital, green and social transitions“
The triple transition refers to the simultaneous progress in digital, green (climate), and social transitions.
This concept has become crucial in the EU’s policy frameworks, aiming to create a sustainable and competitive economy.

Question 7

Explain the concept of the UN Sustainable Development Goals

Answer 7

The SDGs are 17 Goals that were adopted by all UN Member States in 2015. They are a blueprint for peace and prosperity for people and the planet by 2030 and require
collective action from governments, businesses, and civil society to achieve
sustainable development worldwide.
AI’s Role: AI can contribute e.g. to achieving SDGs by optimizing resource use, advancing healthcare, or supporting sustainable agriculture

Question 8

List the UN Sustainable Development Goals

Answer 8

The 17 Goals are:
no poverty; zero hunger; good health and well-being; quality education; gender equality; clean water and sanitation; affordable and clean energy; decent work and economic growth; industry, innovation and infrastructure; reduced inequalities; sustainable cities and communities; responsible consumption and production; climate action; life below water; life on land; peace, justice and strong institution; partnerships for the goals

Question 9

Why was the song “Hi, AI” created?

Answer 9

Song targets digital natives, who use AI tools on a daily basis, but don’t think about environmental implications (they are invisible to them) –> increase awareness.
Song is a medium where people listen to it repeatedly, content is internalised.
(in general AI literacy in Austria is not very high, only 11,3% say that they could explain what AI is)

Question 10

How much energy is consumed per year? (TWh)

Answer 10

De Vries (2023): “2027 (…) NVIDIA could be shipping 1.5 million of its AI server units (…) these machines could have a combined power demand of 9.75–15.3 GW. Annually, this quantity of servers could consume 85.4–134.0 TWh
of electricity.” (is an estimation, theoretical paper, newer estimations are even higher)
For comparison: Countries such as Sweden, Argentina or the UK have a comparably large annual total energy consumption

Question 11

How much water is consumed for a conversation?

Answer 11

Li et al. (2023):
“the enormous water footprint of AI models — many millions of liters of freshwater withdrawn or consumed for electricity generation and for cooling the servers — has largely remained under the radar.”
Depending on location and season, 500 ml of water are needed for running 20-50 inferences of GPT-3

Question 12

How is electricity for AI generated?

Answer 12

traditionally: coal, gas
green energy: wind, solar
“green” (is argued): nuclear power plants

Question 13

How to make AI more energy-efficient?

Answer 13

• Hardware efficiency improvements
• Innovations in model architectures and algorithms

Question 14

What is the “Rebound Effect”?

Answer 14

aka Jevons’ Paradox:
as technology becomes more efficient in using a resource, the overall consumption of that resource can actually increase rather than decrease (the improved efficiency often lowers costs or makes the technology more accessible, leading to higher demand and greater overall use).
example: more AI in cars make them more efficient –> more cars are used –> more energy consumed

Question 15

What is “Lifecycle Assessment of AI”?

Answer 15

Lifecycle Assessment (LCA) is a method used to evaluate the
environmental, economic, and social impacts of a product, process, or service throughout its entire lifecycle (raw material extraction - manufacturing - distribution - use - disposal/recycling)

Question 16

Why do we need “Lifecycle Assessments of AI”?

Answer 16

• understanding the environmental and social consequences of different choices
• making informed decisions to reduce the negative impacts of
  human activities

Question 17

Describe the project “Anatomy of an AI System”

Answer 17

explores the lifecycle of an Amazon Echo device (Alexa), map that can be zoomed in, breakdown of all lifecycle stages

Question 18

Which aspects have to be included to calculate the total environmental footprint for AI?

Answer 18

examples: mining raw material, commuting, data quality, efficiency of code, electronic waste, hardware, data storage, nuclear waste …
difficult: where to start and where to stop when defining metrics

Question 19

What is “green-washing”

Answer 19

a form of advertising or marketing spin that deceptively uses green PR and green marketing to persuade the public that an organization’s products, goals, or policies are environmentally friendly, when they are in fact not

Question 20

What is the organization “Climate Change AI (CCAI)”?

Answer 20

CCAI is an organization composed of volunteers from academia and industry who believe that tackling climate change requires concerted societal action, in which machine learning can play an impactful role.

Question 21

What is the “Impact Assessment Framework”?

Answer 21

Kaack et al. (2022):
- Framework for assessing the greenhouse gas (GHG) emissions impacts of machine learning (way of thinking, not a formula):
A) Compute-Related Impacts
B) Immediate Application Impacts
C) System-Level Impacts
colour code:
Green = effects to reduce emissions
magenta = increases.

Question 22

What are Compute-Related Impacts of the “Impact Assessment Framework”: A) Model development and deployment?

Answer 22

amount of energy consumption depends on algorithm, stage of ML model life cycle, size of models (number of parameters), number of FLOPs (floating point operations), …
example: different number of FLOPs has different error rates -> to consider if benefits outweigh the costs in terms of emissions

Question 23

What are Compute-Related Impacts of the “Impact Assessment Framework”: B) Computing infrastructure?

Answer 23

majority of ML-related workloads: cloud and hyperscale data centers (0.1-0.2% of global GHG emissions, 20% growth annually), smaller share occurring on distributed devices (e.g. PCs, in future also smartphones).

Question 24

Explain POSITIVE Immediate Application Impacts of the “Impact Assessment Framework”

Answer 24

ML models can:
* track GHG emissions (essential to monitor worldwide energy consumption)
* forecast renewable power production, crop yields, and
transportation demand
* improve operational
efficiency for industrial heating and cooling
systems

Question 25

Examples for POSITIVE Immediate Application Impacts of the “Impact Assessment Framework”: HVAC systems

Answer 25

Automated and predictive HVAC systems collect data from the humidity and temperature sensors -> data is analyzed in real time and combined with external information such as
weather forecast to predict the needed heating or cooling for the building -> turn down air conditioning before the room temperature is too low (money saved, comfort increased), better planning of maintenance downtimes

Question 26

Examples for POSITIVE Immediate Application Impacts of the “Impact Assessment Framework”: See & Spray – Blue River’s Technology

Answer 26

technology identifies a wide variety of harmful weeds, targets weeds in real-time, crops are not inhibited by weeds (space, sunlight, nutritients)

Question 27

Explain NEGATIVE Immediate Application Impacts of the “Impact Assessment Framework”

Answer 27

ML applied in ways that can make climate targets harder to achieve, examples:
* decrease cost of emissions-intensive activities (oil and gas extraction and exploration) -> larger consumption of fossil fuels, less reliance on renewable energy.
* manage livestock -> increase cattle-farming (already ca. 15% of total GHG emissions)

Question 28

Explain System-Level Impacts of the “Impact Assessment Framework”

Answer 28

changing the demand for certain goods and services, for example: advertisement recommender systems -> increase consumption of goods and services with high or low GHG emissions (personal vehicles vs. railway companies)
System level impacts may be hard to quantify, but have potential to outweigh immediate application impacts

Question 29

What does “134 TWh” refer to in the context of AI?

Answer 29

It refers to the annual energy consumption of AI systems, highlighting their environmental impact.

Question 30

Why is sustainability a key concern in AI development?

Answer 30

Because AI’s energy demands strain natural resources and contribute to ecological degradation.

Question 31

What are transformer models in AI?

Answer 31

Algorithms like GPT-3 that require extensive computational resources for training. This releases significant amounts of greenhouse gases.

Question 32

How does “Hugging Face” work to reduce AI’s carbon footprint?

Answer 32

By developing tools and practices to make model training more energy-efficient.

Question 33

What is an ecological footprint?

Answer 33

A measure of the demand placed on Earth’s biocapacity by human activities.

Question 34

What resources are heavily consumed by AI systems?

Answer 34

Energy, water, and materials like silicon and rare earth metals.

Question 35

What is “ghost work” in the context of AI?

Answer 35

The invisible labor of data workers behind AI systems, this includes for example data annotators and content moderators.

Question 36

What challenges do AI content moderators face?

Answer 36

Content moderators provide critical human input for ensuring that AI systems remain usable and safe. However, they have to face low pay, psychological stress, and lack of legal protections.

Question 37

How does outsourcing affect AI workers?

Answer 37

It leads to exploitative conditions (poor working conditions, low wages, lack of job stability), especially in the Global South.

Question 38

What is “lifecycle analysis” in AI?

Answer 38

An assessment of environmental and social impacts across the lifespan of an AI device.

Question 39

Why is resource extraction a concern for AI systems?

Answer 39

Because it relies on non-renewable materials and causes ecological damage.

Question 40

How does AI contribute to e-waste?

Answer 40

Through the disposal of outdated or non-functional devices containing hazardous materials.

Question 41

Why is annotation work critical for AI?

Answer 41

It provides the labeled data necessary for training AI systems.

Question 42

How does outsourcing influence the AI labor market?

Answer 42

It shifts work to regions with lower labor costs, often leading to exploitation.