Unnoticed by many, artificial intelligence (AI) has long been intervening in our everyday lives – be it as a translation system, search engine, personal language assistant or robots programmed for specific activities. “The fact that AI will be an enormously important criterion for success in the future for science, society and especially the economy, is now sufficiently accepted,” says Kate Saslow, project manager for AI and foreign policy at the German Stiftung Neue Verantwortung (New Responsibility Foundation). This is largely to do with the fact that the potential applications of intelligent computer programmes are extremely versatile and, as a cross-cutting technology, connect the diverse areas.
What has become clear is that AI is likely here to stay, and will only develop further in the future. Instead of being stuck in the still debated question of “whether we need or want AI at all” – and thus missing the chance to have a say in developments – we should be discussing how and where we can use AI to serve the general good of humanity.
What is Artificial Intelligence?
Voice or face recognition, search engines, self-driving cars, computer games, social bots (chatbots that mimic human communication) or robots – all of these areas heavily involve AI.
But what exactly is artificial intelligence? So far, there is no single universally valid definition of AI. If we look for the lowest common denominator, artificial intelligences can be described as computer programmes that can learn and improve themselves. They are intelligent to the extent that human cognitive decision-making structures are replicated by programming – this includes visual perception, speech recognition and generation, reasoning, decision-making and action, as well as the ability to adapt to changing environments.
Simple, “non-intelligent” algorithms (for example, navigation devices) are programmed to answer clear questions: this is the problem and this is how it is solved (e.g. “find the shortest way from A to B”). The very heterogeneous AI applications, on the other hand, are not based on classical deterministic programming, but on statistical data analysis. In a learning process – also called machine learning – the artificial system learns from examples based on training data, which it can generalise independently after the learning phase. An AI-supported translation system, for example, is fed hundreds of thousands of data in the training phase (for example inputting a sentence in German, and outputting a sentence in English) and does not learn the examples by heart, but patterns and regularities (in this case, which word is translated into which word, which sentence into which sentence), which can then also be applied to new, previously unseen data.
Intelligent Computer Programmes – a Rapid Development
The European market for artificial intelligence will grow from around three billion euros in 2019 to ten billion euros in 2022. This corresponds to an average annual growth of 38 percent, as the European Information Technology Observatory (EITO) states in its study “AI in Europe – Ready for Take-off”.
The decisive factor for the rapid developments in the field of AI applications in recent years is the fast-growing computer power – because it is only this that makes it possible to process the large amounts of data of the rapidly advancing digitalisation. An end to these developments is not in sight. Computer scientists Dario Amodei and Danny Hernandez, who both work for OpenAI, a non-profit research company founded by, among others, the AI-critic Elon Musk, conclude in their study that an increase by a factor of 300,000 took place in the period from 2012 to 2017. This means that the new AI systems are currently, on average, doubling their speed every three and a half months and are thus able to process ever larger amounts of data.
Artificial Intelligence and Sustainability – a Good Team?
AI is being treated as a key technology in many social – and especially economic – areas. Is it perhaps also a key to transforming our society towards ecological and social sustainability? In view of the urgent need for action to avert or at least contain climate collapse, a wide variety of projects, start-ups and research projects have set out to find new solutions with the help of intelligent computer programmes.
In earth observation, for example, machine learning methods – such as pattern recognition and the linking of large amounts of data – together with data collected by satellites can help to make precise statements about climatic changes on our planet.
- The German Research Center for Artificial Intelligence (DFKI), for example, wants to combine machine learning algorithms with high-resolution satellite imagery to quickly and accurately access ground changes around the world. Hopefully allowing action, while something can still be done.
- Launched in 2017, the Sentinel-5P satellite is a collaboration between three European nations under the EU’s Copernicus Earth-Monitoring programme. Using a specialised instrument, Sentinel-5P can highlight a wide range of pollutants, such as ozone, methane, carbon monoxide and sulphur dioxide, with unprecedented resolution and accuracy. The information will then be decoded with the help of AI to track down polluters entering our atmosphere.
- Munich-based start-up Hawa Dawa is using sensors and AI to tackle the issue of air pollution in our cities: Hawa Dawa has developed a mobile measuring device that is networked via ‘Internet of Things’ technology with the Hawa Dawa software. This therefore allows it to compare weather and traffic data in real time. The software, which is equipped with artificial intelligence, constantly learns from the data compilation and can thus create forecasts as well as simulate data for areas where no sensors are installed. This creates an area-wide air quality map in real time.
In the energy sector, AI can help to regulate the growing complexity of the decentralisation of the energy system that goes hand in hand with the switch to renewable energy sources, as well as to use infrastructures more efficiently and increase the flexibility of the energy system by intelligently networking energy systems (smart grids), intelligent building control and energy data management.
- Green traffic lights not only save time, but also fuel. The savings potential, especially for larger vehicles such as trucks or buses, are enormous. The Hamburg Port Authority (HPA) is currently testing the “Green4Transport” project in the Port of Hamburg. The aim is to optimise traffic flow by networking vehicles with each other and with traffic lights.
- Together with Oxford University, the Environmental Defense Fund Europe and the WWF, the UK National Grid has developed a so-called “Carbon Intensity Forecast“, a kind of “weather forecast” for clean electricity. The software can indicate the proportion of renewable and non-renewable energy in the UK’s electricity, which then results in a forecast of CO2 emissions. Consumers receive the predictions via an app, allowing them to shift their energy consumption to periods of low carbon intensity.
And AI applications could also contribute to making the transport system more environmentally friendly, as well as assist in waste and recycling management.
- Food waste is a huge problem, including in the restaurant industry. The problem is compounded by the fact information on how much is thrown away and when is rarely recorded. The company Winnow Vision, founded in London in 2013, wants to help solve this issue using artificial intelligence. Winnow has developed a smart system that recognises and records the discarded food photographically. With this information, the system produces regular reports calculating the volume, value and environmental impact of the waste. With this knowledge, restaurateurs can make conscious decisions to reduce food waste in the kitchens.
For more examples of how AI is helping to protect wildlife or preserve our forests, click here: Sustainability and AI.
Sustainability AI in Policy and Research
Even though AI applications are expanding in all sectors, the use of the technology in the sustainability context is still restrained. This is evident from a look at the number of start-ups as well as research projects and funding.
As part of a study by the German Federal Environment Agency, the Crunchbase database was searched for start-ups that use AI technologies in the context of sustainability. The search revealed a total of 155 companies. Compared to the total of almost 12,000 startups that indicate references to artificial intelligence in the database, startups with a sustainability reference are therefore very much a niche.
How Trustworthy is AI?
In addition to a very technologically driven discourse on the feasibility and possibilities of AI, there is currently also an ethical debate on the topic.
Central questions are:
Autonomy: How autonomously does an AI act? The concept of autonomy of AI systems calls into question the self-determination of humans. This is currently being discussed above all in the application case of autonomous vehicles.
Trustworthiness and Transparency: How does an AI reach a decision? There is a certain lack of transparency in the principles of machine learning algorithms themselves, since the use of statistical correlations makes the algorithms more difficult to understand. For example, it is not always possible to deduce from an algorithm why a certain input value produces a certain output value. The answer: “Because the algorithm learned it that way from the sample data” is unsatisfactory for a deeper understanding. This can also become a problem from an environmental perspective; namely, when decisions made by AI systems cause environmental damage.
In order to prevent AI systems from reproducing the discriminatory patterns that exist in society and from making unsustainable decisions, it is necessary to ensure the greatest possible transparency of the algorithms on the one hand and to ensure that the training data for algorithms is designed in a more inclusive way on the other.
AI as a Personal Shopper
Most start-ups, financial service providers and established e-commerce companies have long since integrated AI into their mobile applications, shopping systems or the data processing of their online stores. The intelligent algorithms classify and predict, suggest products to customers and make purchase recommendations.
What is worrying is that the dominance of the large IT and internet corporations (Google, Microsoft, Apple, Facebook, IBM, Amazon, etc.) is continuing in the AI sector – as in all other areas of the digital economy. These globally active providers of digital services now have a near-monopoly position and have huge pools of data at their disposal. From a sustainability perspective, the main problem is that the digital corporations are using AI applications to further personalise their services and create increasingly accurate forecasts about the purchasing interests of their customers, thus ultimately further increasing consumption. The German Federal Environment Agency study, “Consumption 4.0” states: “The lowering of technical barriers, the integration of purchase recommendations and evaluations into consumers’ everyday lives via social networks and personalised marketing can encourage consumers to make more frequent purchases, limited only by their disposable income.”
Thus, the current priority use of AI applications that increase consumption ensures higher resource use and rising CO2 emissions.
The Energy Hunger of Machine Learning
In addition to a discourse on the risks of AI that is heavily influenced by ethical concerns, the question of the life cycle assessment of AI systems is also relevant from a sustainability perspective. While some see enormous opportunities in the fact that we can tackle environmental challenges with AI, others fear it will merely create new problems – with the further increase in our already enormous demand for electricity and massive increases in consumption.
The concerns are not unfounded: AI applications such as deep learning, simulations and forecasts require more and more computing power and thus ensure that the energy demand of data centers increases. And this is already very high with the rapidly advancing digitalisation: according to the Berlin Borderstep Institut, the energy demand of all data centers in Germany amounted to around 13 billion kilowatt hours in 2017 alone – and the trend is rising. As researchers at the University of Massachusetts have determined, training a single AI application for speech recognition can generate five times as much CO2 as a car emits over its entire lifetime.
At the same time, it must be taken into account that, for example, training such a network generates services for a large number of users, which can lead to emission reductions in the long term. So it is not very easy to make a blanket statement about the life cycle assessment of AI applications in general and must be considered on a case-by-case basis.
Future AI applications could also have lower energy consumption, as there are still refinements that can be made, especially in the area of machine learning: “That is why many research groups are working on making deep learning less energy-hungry. Different hardware, new models and learning methods – all this is being investigated,” says Prof. Dr Kristian Kersting, Head of the Machine Learning Department at the Darmstadt Technical University.
Sustainable AI in Policy and Research
A closer look reveals that AI for sustainable development has so far received little attention and hardly any funding in both political strategies and research.
- A first step: The German AI Strategy
A step in the right direction is the German AI strategy, which was published in December 2018 and in which sustainability references are at least made.
It outlines, albeit on a rather abstract level, the potential of AI for achieving the national, as well as the international sustainability goals. In particular, it is emphasised that AI applications can improve the “understanding of complex systems [such as] nature”. In this context, the authors point out that new methods must be developed to evaluate the data generated by satellites and to combine it with data from other sources (e.g. geoinformation, social media and citizen science data).
The proposals of the German AI Strategy on the potential of AI for the environment, resources and climate are somewhat more concrete. The goals of AI applications for socio-ecological change are mentioned: Measures should be strengthened, evaluation criteria for the environmental impact of AI should be developed and an environmental data cloud should be established. This is to be achieved by means of the 50 lighthouse applications funded by the Federal Ministry for the Environment’s (BMU) funding programme “AI Lighthouses for the Environment, Climate, Nature and Resources”
- The “AI Lighthouses for the Environment, Climate, Nature and Resources” Funding Programme
As part of the funding programme, the BMU is looking for projects from business, science and civil society that use artificial intelligence to tackle ecological challenges and that are exemplary for environmental, climate and nature-friendly digitalisation. In 2019, a funding budget of 27 million EUR was made available, which was increased to 45 million EUR in 2020
- Little Happening Elsewhere in Europe
At the European level, there are very few direct references between AI and sustainability in the “Communication Artificial Intelligence for Europe” plan to support AI, which was presented in April 2018. Even though AI is at least mentioned as a possible solution to combat climate change and, in the EU’s view, can also contribute to achieving the SDGs, there is no development of this concept.
- Research still very early
For AI in the environmental field, there are no studies to date, either at European or international level, that allow for a comprehensive interpretation of research activities in the context of AI and the environment/sustainability.
However, there are a handful of short studies that explore the research field and raise questions rather than provide concrete answers or assessments of individual applications, including the WGBU report: “Our Common Digital Future” and the study “Artificial Intelligence in the Environmental Domain. Application examples and future perspectives in terms of sustainability” by the Federal Environment Agency (UBA).
- Further Funding Programmes
The search for specific funding programmes also yields few hits. Besides the BMU’s AI lighthouses, only two other funding programmes of AI with sustainability relevance can be found – at the large technology companies Google and Microsoft.
Microsoft launched the “AI for Earth” funding programme in summer 2017. According to its own information, 435 projects in 71 countries have been funded so far. The majority of the projects funded are concerned with monitoring, modelling and management in various areas (e.g. forestry, agriculture, fisheries) as well as the observation of ecosystems and biodiversity.
Google launched the “AI for Social Good” programme in October 2018. In addition to its own (partner) projects, the programme also supports non-profit organisations, researchers and social enterprises that use AI to address social, humanitarian and environmental problems as part of the Google AI Impact Challenge.
Conclusion: It All Depends on the Framework Conditions
With this overview, we have set out to find an answer to the question of how far AI applications can contribute to environmental and climate protection. Two things should have become clear. As initial projects show, AI applications certainly have the potential to support sustainable development by contributing to a more efficient use of resources, providing us with accurate information about the state of our planet or making forecasts easier. At the same time, however, it is also quite clear that the fact that AI leads to more sustainability is not a foregone conclusion in view of ethical concerns, the ecological balance sheet due to electricity consumption and the superiority of globally active IT companies, also in the AI sector, which ensure that intelligent computer programmes continue to boost global consumption in particular.
It is therefore necessary to conduct a comprehensive debate on this issue right now, in which the opportunities and risks of this technological development are weighed up not only for people and social interaction, but also for the environment and climate. While ethical-social aspects are already in the foreground in the general discussion about AI, ecological-sustainable effects and technology consequences are still hardly addressed.
In order for AI applications to really contribute to environmental and climate protection, suitable framework conditions must be created. This includes, for example, guidelines that keep energy consumption within limits and a political framework that guarantees data protection and transparency. If, for example, AI makes suggestions for the use of different transport routes or issues warnings for air pollution, it should be comprehensible according to which criteria the recommendations are made. In addition, the social question should always be asked for which concrete applications a comprehensive linking of data, for example on resource flows, consumption styles, energy consumption or financial transactions, is desirable at all and whose goals this serves. This also includes ensuring that data and AI systems are not controlled by just a few global players.
In order to drive innovative, sustainable AI developments, it is important on the one hand to promote research and young start-ups more strongly and on the other hand to create new interfaces between research communities. Currently, research communities that deal with AI and those that deal with climate and environmental problems work together far too little.
To the question of whether AI can save the world, the answer likely remains, “No”. AI technology alone will not save the world – as, incidentally, no other technology will. But artificial intelligence can make an important contribution – if we use it sensibly. We need to create the framework conditions for this.
This article was originally published in December 2019. It has since been updated on 04/20/2022 by Marharyta Biriukova.