Yesterday afternoon, the Foundation for Digital Sustainability – the first recognised research foundation in Italy dedicated to investigating issues of digital sustainability – announced the launch of the ‘Manifesto for the Digital Sustainability of Artificial Intelligence’, the Position Paper on Artificial Intelligence (AI) and Sustainability: the document is intended to be a guide to approaching AI in a conscious, responsible and sustainable way.

What happened to Tom Hanks – who found himself starring in a commercial that used his image created with artificial intelligence, without his knowledge – highlights a condition that could become increasingly common and that will not only affect VIPs, politicians and public figures but also, and increasingly, the general public.

The evolution of dedicated, simple and inexpensive Artificial Intelligence systems will drive the spread of deep faking techniques and thus the creation of ‘fake’ videos that cannot be distinguished from the original ones, where true and fake overlap. Three elements are needed to tackle it: regulation, and thanks to the AI act Artificial Intelligences must meet strict criteria; culture, through widespread and generalised cultural programmes that create aware and informed citizens; and technology. Just as AI can generate deep fakes, AI must be able to generate algorithms that unmask them and enable the user to understand the nature of the images that appear on the video.

The Manifesto serves precisely to analyse challenges and opportunities of AI in relation to society and, in the last part of the document, to identify, for each SDG (Sustainable Development Goal), those features of AI that contribute more than others to the realisation of the individual SDG goals. Starting from the assumption that AI has entered a phase of mass impact, involving not only users but also public and private institutions, the Position Paper aims to place AI in a digital sustainability context, providing a clear vision of the principles and characteristics that should guide the development and adoption of this technology. It is not about limiting AI, but about maximising its benefits while respecting the principles of sustainability.

“It is crucial for each of us to deepen our knowledge of these technologies in order to understand their potential,” said Marzio Bonelli, CIO of MM. “We would discover how current Artificial Intelligence, even generative intelligence, is not able to invent anything new but is a powerful articulator of existing knowledge capable of discovering weak relationships that escape our minds. We will discover how much algorithms can disrupt business models in the world of work, but also how much they can be conditioned, consciously or unconsciously, by those who generated them, and how important it is to recognise this risk in order to correctly interpret the results that AI proposes to us, approaching this technology in a constructive and conscious way, without prejudice.”

The Position Paper is developed in three main sections: the first contains a description of the AI domain and its definition, the second refers to the criteria and principles that need to be considered in order to evaluate and use it in a conscious way, and the third deals with the SDGs of Agenda 2030 as a key to defining the defining characteristics for the realisation of sustainable AI systems and solutions.

“The development of the manifesto saw the active engagement of a differentiated and multidisciplinary working group, made up of both professors from the universities in our network and experts from the companies that support the Foundation,” explained Stefano Epifani, President of the Foundation for Digital Sustainability. “In the reflection on the future of AI, multidisciplinarity is fundamental: it was exciting to see some of the brightest minds that are developing reflections in this field hybridising and combining legal vision (with the contribution of experts such as Giovanni Battista Gallus and Lara Lazzaroni), with technical-scientific expertise and a philosophical approach (with the contribution of scholars such as Tiziana Catarci and Fabio Ciracì). This is the modus operandi of the Foundation, which makes the interdisciplinary approach to problems related to digital sustainability one of its strengths.

Criteria for evaluating AI systems and solutions

The document presents the principles and characteristics needed to evaluate any product made from Artificial Intelligence.

It starts from the assumption that AI should not be regarded as a commodity that is accessed unconsciously, but rather as a technology that requires awareness, expertise and critical capacity. The goal is to use AI to build a more sustainable world by addressing the real problems that society faces.

To this end, the Foundation for Digital Sustainability has identified some criteria, listed in the Artificial Intelligence Manifesto for Digital Sustainability, that all organisations (public and private) should use when developing and adopting AI systems:

1. Respect for fundamental rights: AI systems must be designed from the outset to ensure respect for human dignity and fundamental rights, as defined by the Charter of Fundamental Rights of the European Union and the European Convention on Human Rights.
2. Privacy: AI must comply with data protection criteria and principles, giving users full control over their data.
3. Transparency: Artificial Intelligence must be based on transparency criteria both with regard to training data and reference sources, and with regard to the logics and algorithms adopted.
4. Non-discrimination: Artificial Intelligence must not create unequal treatment between subjects or groups of subjects, whether generated by algorithmic biases derived from training data or design choices.
5. Security: AI systems must guarantee the confidentiality, integrity and availability of information, take into account possible risks arising from interactions with people and other systems, and provide security mechanisms from the design stage to ensure that they are verifiably secure at every stage.
6. Interoperability: AI should not create new ‘walled gardens‘ but should be based on open standards and protocols that can optimally guarantee the exchange and reuse of information.
7. Portability: Users must be able to export their data (not only personal data) in a structured format and transfer them, even in an automated way, from one system to another.
8. Accessibility: Artificial Intelligence systems must guarantee, from the design phase and in subsequent phases, accessibility to people with disabilities, on an equal basis and ‘leaving no-one behind’.
9. Override: The possibility of effective human oversight, which can affect processes and actions governed or performed by AI, must be guaranteed.
10. Recognisability: Users must be enabled, in a simple and intuitive way, to know that they are interacting with an AI system. AI products must also be clearly identified, with standard and interoperable systems.
11. Proportionalityof risk: In the development of AI, there must be a proportionality relationship between implementation models, usage dynamics and regulatory processes in relation to the extent of risks that may be generated by its adoption.
12. Energy Efficiency: AI systems must be designed taking into consideration the environmental impact generated both in the training phases and during operation and use by users.

The criteria set out in the position paper serve to establish a scale for evaluating AI systems and solutions, and cannot disregard the scope of their use.

How AI contributes to achieving the Sustainability Goals (SDGs)

A crucial aspect of AI is its growing importance in the realisation of the sustainability goals set by the United Nations’ Agenda 2030. AI offers innovative solutions to address global challenges such as climate change, poverty, public health and gender equality.

By grouping the SDGs into 3 macro-categories, Well-being and Social Sustainability (SDGs 1, 2, 3, 4, 5, 8 and 10), Innovation and Environmental Sustainability (SDGs 6, 7, 9, 11, 12, 13, 15) and Peace, Justice and Sound Institutions (SDGs 16 and 17), it can be seen how AI, in its specific characteristics can impact the achievement of the goals of each SDG:

• Well-being and Social Sustainability Group: this group of SDGs appears to be the most sensitive to the intrinsic characteristics and purpose of AI systems and solutions. The non-discriminatory aspect of AI, such as attention to gender bias in the learning phase promotes inclusion and the promotion of equality(SDG5), while reducing inequalities(SDG10). Interoperability and portability are crucial for SDG3 as the possibility of real-time data exchange between multiple AI solutions. Their ease of adoption, on the other hand, can be instrumental in saving lives. The right to revoke action and human control over algorithms can avoid unwanted effects that, by relevance, always impact SDG3 the most. Accessibility is a feature that reduces inequalities and must be a benchmark for all digital technologies that must tend to favour inclusion. Finally, the environmental impact characteristics make special reference to SDG3 as a sustainable AI, operating in an optimised way, containing the consumption of computational resources has a positive impact on the environment and consequently on people’s health and well-being.
• GroupInnovation and Environmental Sustainability: this group of SDGs is the only one that is pervasively influential, and this confirms the leading role of AI in the development and innovation path. Indeed, AI can contribute to reducing the environmental impact of industrial activities, for instance, by optimising energy efficiency in industrial processes or by providing insights into water losses. Through monitoring and data analysis, AI-managed processes can promote a more sustainable use of resources(SDGs 6 and 7). Technical features are crucial for SDG9 (Innovation and Infrastructure) as these are implicit requirements. SDG11 (Sustainable Cities and Communities) can be achieved through the development of smart mobility, such as shared transport systems and traffic optimisation. Thus, contributing to reducing greenhouse gas emissions and improving air quality in urban areas. Data produced and managed by AI can be used to understand climate change processes and consequently develop new climate prediction models contributing to improving the resilience of communities and addressing environmental risks(SDG13).
• Peace, Justice and Sound Institutions group: this group of SDG targets is influenced by some of the characteristics of the IA, in particular the intrinsic and purposive ones. SDG16 aims to promote peaceful, just and inclusive societies. A non-discriminatory and ethically designed AI enables the implementation of systems and algorithms that avoid discrimination based on characteristics such as ethnicity, gender or religion. AI can help ensure fairness and justice in decision-making by promoting the impartial treatment of all people, regardless of their personal characteristics. The ‘inclusive’ aspect of IA fosters social cohesion and trust between different groups in society, helping to eliminate prejudice or stereotypes, reduce social tensions and promote cooperation and peaceful community building. As forSDG17 (Partnership for the overall goals), we have observed in recent years how AI is transforming several sectors and requires adequate literacy to be able to meet the challenges and seize the opportunities it presents. Literacy enables people to develop skills to use AI effectively and responsibly, promoting innovation and sustainability.