The PIO Model

How it works

The PIO Model is a new assessment tool on ethical and legal uses that is harmonised with legislative requirements and current ethical standards and recommendations. This assessment tool consists of a checklist on ethical uses structured around seven principles, and designed for the development, deployment and evaluation of AI data and systems.

The PIO Model starts with so-called zero questions or basic questions that reflect the essential ethical content that makes up the checklist of the model and, in this sense, allows a quick approximation of the fundamental ethical considerations to be taken into account (Phase 0).

After this initial positioning, the PIO Model allows for a pre-assessment according to the risk categories under the EU Artificial Intelligence (AI Act) regulations. In this sense, AI systems will be classified according to whether they are considered to be of unacceptable risk based on the AI Act norms. Unacceptable risk refers to a very limited set of uses of AI that are particularly harmful and contravene EU values, as they violate fundamental rights and are therefore prohibited (Phase 1).

Once this distinction on unacceptable risk or not has been made, the same pre-assessment will classify the IA systems according to the IA Act as follows (Phase 2):

1. High risk: this refers to a limited number of AI systems that may have an adverse impact on the security of people or fundamental rights under the EU Charter of Fundamental Rights and are therefore considered high risk.
2. Limited risk: this refers to the risks associated with the lack of transparency in the use of AI and, consequently, the need for specific transparency obligations to ensure that humans are informed where necessary, building trust.
3. Minimal risk: this refers to all other AI systems that can be developed and used in accordance with existing legislation without additional legal obligations. On a voluntary basis, providers of such systems may choose to apply ethical principles such as those of the PIO Model.

The outcome of this pre-assessment focusing on the IA Act risk assessment will allow users of the PIO Model to position themselves as a starting point on the assessment of some of the main legal requirements of the core EU IA regulation (Phase 3).

After the pre-assessment, the checklist with examples will start, which is organised through the seven principles of the PIO Model and presents each question accompanied by the legal provisions related to AI data and systems on which it is based (not only around the AI Act), as well as with different ethical standards and recommendations (Phase 4).

The collection of system assessment information is carried out according to labels related to risk categories (high, limited and minimum), which accompany the questions to facilitate navigation and response according to the data and systems being assessed. In this sense, next to each of the questions, the risk category to which it refers is indicated with the following labels:

If a question has been labelled with more than one colour, this implies that the issue or question falls into more than one risk category and therefore also implies that there is a mandatory requirement according to the legal provisions.

The collection of information is done through binary answers (affirmative or negative) that do not apply according to the information that the users of the PIO Model have about the data and AI systems they are assessing in order to perform, a posteriori, a computation and photograph of the degree of compliance in terms of obligations and rights according to legal requirements and ethical standards (Phase 5).

With the information obtained through the pre-assessment and assessment, metrics (percentages) will be obtained on the degree of compliance with each of the 7 principles assessed and also according to the topics and risk categories of the AI data and systems assessed. These results are subject to the declaration of responsibility, whereby the organisation takes responsibility for the veracity of the answers provided, and will be observable from the profiles that can be created by each of the entities that carry out the assessment of their AI data and systems (Phase 6). 

 The last step, once the corresponding metrics have been obtained, will be the elaboration of a badge that will be the graphic image that represents the adequacy of the entity to the legal requirements and ethical standards according to the PIO Model through the seven principles of the model: (1) transparency and explainability, (2) justice and equity, (3) safety and non-maleficence, (4) responsibility and accountability, (5) privacy, (6) autonomy, and (7) sustainability. The same badge will be downloadable to reflect and evidence the adequacy and capacity building of an AI system in accordance with the PIO Model (Phase 7).

To whom it is addressed

The answer is simple: to all public or private organisations that have an interest in the design, development or deployment of AI technologies, and to all people who are using, buying or are recipients of these technologies and are asking themselves questions such as: What are the risks and benefits of these technologies? Who is affected by them and how? How can the well-being of people be improved with these technologies? We believe that these and other questions are sufficiently relevant to question all the organisations that form part of any innovative process: universities and research centres, public administration, business fabric and citizens.

Simplicity

The PIO Model is built with simplicity at its core, revolving around a fundamental and effective question that any public or private organisation involved in developing, managing or overseeing projects that incorporate AI data and systems can easily understand: have we done it?

Based on the premise that one of the most critical evaluations that public and private organisations can undertake concerns their use of AI data and systems, the PIO Model emphasises the importance of these evaluations. Regardless of whether organisational decisions are perceived (both internally and externally) as highly positive or negative, there is a growing need for these decisions to be aligned with ethical principles relevant to the context. Furthermore, it is imperative that this alignment is communicated transparently and implemented as accurately as possible, especially in light of new AI legislation such as the IA Act.

From a corporate social responsibility point of view, the PIO Model presents itself as a versatile tool, applicable to various stages of the AI data and systems lifecycle. This encompasses the design and modelling phase, which includes planning, data collection and model building; the development and validation phase, which includes training and model testing; and the development, monitoring and refinement phase, which involves troubleshooting and continuous improvement.