Artificial Intelligence in support and protection of the environment, according to research conducted by accounting and finance professionals

Artificial Intelligence in support and protection of the environment, according to research conducted by accounting and finance professionals

In an economic and social moment in which it is necessary to face the problem of sustainability and the safeguarding of values ​​such as the survival of our planet, the work carried out by accounting and finance professionals, members of the ACCA (Association of Chartered Certified Accountants ), a global professional body of 233,000 members based in 178 countries and regions, gathered to conduct research aimed at guiding the ethical and sustainable adoption of Artificial Intelligence, entitled ” ETHICS FOR SUSTAINABLE AI ADOPTION – CONNECTING AI AND ESG “.

The preface highlights the following: “ With this great processing and data power, great responsibilities come with it. The report highlights these responsibilities as they fall within the environmental, social and governance ESG ) spectrum . As for the environment, for example, ESG data is highly unstructured and suitable for AI analysis. Accounting and finance professionals should consider new Artificial Intelligence solutions as part of their toolkit for challenging ” greenwashing “, ie when organizations claim to operate sustainably without this being confirmed by the data . “

At the basis of the aforementioned research, 5,723 interviews were carried out, in order to detect, through the survey tool, a faithful picture of the current global situation. “ We need a human-centered approach that considers how people can collaborate with technology for the benefit of society . 

Artificial intelligence is often perceived in abstract terms, as “intelligence” evokes the idea of ​​the intangible. In reality, however, the AI ​​supply chain is very tangible, involving real materials and natural resources. At the center, there is the consumption of energy, which is expressed during the execution of the algorithms. The volume of data is huge and growing exponentially, not linearly. Furthermore, this data includes a significant component of unstructured data , according to some estimates up to 90%. The available computing power can therefore exceed the rates set by previous regulations, such as Moore’s law (according to which the computing power would double every 18-24 months).

Hence, looking through the supply chain, AI systems have an identifiable carbon footprint . And it is not trivial. A study showed that the carbon emissions required to form a natural language processing (NLP) model would be equivalent to 125 round-trip flights between New York and Beijing. And more complex algorithms like GPT-3, used for language and text analysis, would use even more.

Ahead of the UN Climate Summit COP26, there is a stated goal of significantly reducing UK emissions by 2035 (HM Government 2021) and reaching net zero globally by 2050 (National Grid Group 2021) . This has provided organizations with a focus, to demonstrate how they are delivering on these commitments.

Unfortunately, some will also try to misrepresent the true extent of their green credentials. This could be through ” picking and choosing ” ( selecting and deciding ) what to disclose or when to disclose it, or through high-profile public campaigns that implicitly or explicitly suggest certain actions, but in which subsequent fulfillment is absent, giving rise to the phenomenon. of ” greenwashing “.

Here the accounting world and other finance professionals play a fundamental role, as they should:

  • ascertain the effectiveness of the systems, processes and controls on the underlying ESG data included in the financial reporting;
  • validate ESG data and comprehensively assess the financial impacts of environmental, social and governance considerations ;
  • provide complete, accurate and valid financial reporting relating to ESG considerations;
  • adopt KPIs [key performance indicators], metrics, benchmarking, continuous monitoring and evaluation of relevant performance, aligned with recognized ESG reporting standards / frameworks;
  • ensure compliance with relevant regulatory reporting standards on ESG, sustainability and / or climate;
  • lead the transition to greener economies , sustainable environments and fair societies by contributing to the development and management of resilient and ethical organizations.

Machine learning works by training algorithms on datasets. Often the entities providing this data are not aware that they have done so. For example, when data is removed from a website (not illegal in some jurisdictions) and subsequently fed into an AI training engine, this practice, colloquially known as ” participation washing “, results in the participation of individuals in an enterprise. without their knowledge or consent and for which they receive neither recognition nor remuneration.

The next path of this study identifies the paradigms underlying any development of AI, which necessarily must pass through ethical values, governance (which contemplate a correct minimization of data and respect for confidentiality, with approaches of ” differential privacy ” focus on acquiring information from an aggregated top-down view of the data, without the need for specific details about each record (i.e. individual, in the dataset it creates), quality and correct review of the data used, and regulatory parameters that can protect suppliers and users in an appropriate and fair manner, strengthening the climate of trust in technology, an obligatory and fundamental step for the socio-economic growth of each country.

Complaints and compensation are channels, necessary to contest decisions and deal, for example, with dissatisfied customers. The role of whistleblowing mechanisms is also important in this context, especially given the potential for less familiarity with AI among those not on the front lines. More generally, all mechanisms for incident management, exception reporting, escalation and emergency planning apply.

Equally important will be the Protection of the Artificial Intelligence system : fraud and unethical behavior can take many forms, including data poisoning and model evasion , both of which work by corrupting the data used to train the model. of Artificial Intelligence. The operational focus on model protection is intensifying and will be considered for mass adoption.

Ethical issues can arise from a wide variety of sources, during initial setup and operational monitoring.

  • Documentation : One of the ethical challenges with an AI model is ensuring sufficient understanding of what it is doing. Documentation quality is critical: how complete it is, how regularly it is updated, and how understandable it is to new individuals when there are staff changes and handover. And in cross-functional teams, non-technical business users may need access to some documentation. This requires the introduction of strong discipline in maintaining and revising document versions . It may also be useful to explore automated documentation applications to establish an end-to-end path .
  • Access Controls : There should be clarity as to who has access to training data and who can make changes to it. More generally, this extends to close monitoring of privileges and access rights for all data and systems; and through human accounts and, if applicable, Bots.
  • Transparency : Ethical behavior requires that information relevant to the user be readily available in the public domain. Information about how people’s data is used shouldn’t be hidden behind long contractual jargon but should be explained in simple terms. Finding information about a consumer’s ability to opt out should be as simple as the user’s path to initially attract them.
  • Evaluations and Audits : Conduct periodic process and ethical evaluations through an independent internal function and, if possible, use an external expert agency for auditing. Algorithmic impact assessment frameworks can provide a structured way to assess the impact of AI systems.

Therefore, it is necessary to understand well how Artificial Intelligence works, so that decisions and events are not ” outsourced “, delegated to AI, making those who created or put them on the market unresponsible, but recognizing the primary role of supervision and human intervention.

All rights reserved

Avv. Raffaella Aghemo

0 Comments

Your email address will not be published.