The Strategic and Responsible Use of Artificial Intelligence in the Public Sector of Latin America and the Caribbean

Annex B. Examples of LAC AI instruments aligned with OECD AI values-based principles

Table A B.1.  Examples of LAC AI instruments aligned with OECD AI principle #1

	Reviewed Instrument	OECD Principle: Inclusive growth, sustainable development and well-being
Chile	AI Policy and Action Plan (2021)	AI for sustainable economic development: the incorporation of technology as an axis of the sustainable development of the country. (...) The policy and the actions derived from it must promote an AI that does not harm our environment and that, as far as possible, contributes to preserving and improving it. AI in society – all policies and actions related to AI must be approached in an interdisciplinary way, enhancing the contribution of the different areas of knowledge. [AI Policy] actions should integrate collective intelligence and feeling through open processes of deliberation. The actions derived from the Policy will seek to develop each region departing from its reality. Globalised AI – the Policy and the actions derived from it should consider how they are inserted in the international context and promote participation in bilateral and multilateral spaces of which our country is a part.
Brazil	Brazilian Artificial Intelligence Strategy (2021)	AI for sustainable economic development: Strategy includes significant discussion on the importance of the topic and lists some relevant initiatives. Key goal to “Promote sustained investments in research and development in AI”. Establish partnerships across sector to encourage training and to consider the new realities of the labour market. AI in society- Action to create awareness campaigns on the importance of preparing for the development and ethical use of AI. Social benefit – Commitments to “Share the benefits of AI development to the greatest extent possible and promote equal development opportunities for different regions and industries.” And to develop educational programmes at all levels. Globalised AI – Inclusion of a cross-cutting thematic axis on “international aspects”.
Colombia	AI Ethical Framework (2021)	Inclusion: “active participation of historically marginalized and diverse populations in the design, development, implementation, and evaluation of artificial intelligence systems used in Colombia” (p.37). This principle is also applied to data (representative data), algorithms (avoiding privileging specific groups), and practices. Prevalence of the rights of children and adolescents. Social benefit: “artificial intelligence systems implemented in Colombia must allow or be directly related to an activity that generates a clear and determinable social benefit” (p.40). Goals should be linked to recognised social ends, contextualised to Colombian main social challenges.
Mexico	AI Principles	Seek social welfare as the main goal of AI implementation, focusing efforts on generating a positive impact that contributes to solving public problems. Have an inclusive approach that encourages vulnerable groups to access the benefits associated with the use of these systems. Monitor and evaluate the impacts of AI implementation to ensure that it fulfils the purpose for which it was designed.
Uruguay	AI Strategy	Purpose: AI must enhance the capabilities of human beings, complementing them as much as possible, aiming to improve the quality of people ́s life, facilitating processes and adding value to human activity. General interest: AI-based solutions promoted by the State should tend to protect the general interest, guaranteeing inclusion and equity. For this, work must be carried out specifically to reduce the possibility of unwanted biases in the data and models used that may negatively impact people or favour discriminatory practices.

Source: OECD review of national AI principles and related instruments.

Table A B.2.  Examples of LAC AI instruments aligned with OECD AI principle #2

	Reviewed Instrument	OECD Principle: Human-centred values and fairness
Barbados	Data Protection Act (2019)	Fairness of Processing: personal data shall be processed lawfully, fairly and in a transparent manner in relation to the data subject. Right to prevent processing likely to cause damage or distress. Automated individual decision-making, including profiling: the data subject has the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or similarly significantly affects him.
Brazil	Brazilian Artificial Intelligence Strategy (2021)	Non-discrimination. Action to direct funds towards that aim to apply ethical solutions, mainly in the fields of equity/non-discrimination (fairness). Action in public procurement to include criteria related to the incorporation of ethical principles related to equity and non-discrimination. Diversity. Action to stimulate the diverse composition of AI development teams with regards to “gender, race, sexual orientation and other socio-cultural aspects”. Human-centred AI. Action to promote means to observe AI systems and how they promote human rights, democratic values and diversity. Action for human review and for human intervention in high-risk situations. In the public sector, establish mechanisms for prompt investigation of complains about violations of rights. Other strategic actions. Action to stimulate partnerships with corporations researching ethical AI technologies. Action to create open multi-sectoral spaces for the discussion and definition of ethics principles.
Brazil	General Law on the Protection of Personal Data (2018)	Non-discrimination: impossibility of processing for unlawful or abusive discriminatory purposes. The law includes two other related rights: Every natural person has ensured the ownership of his personal data and guaranteed the fundamental rights of freedom, intimacy and privacy, in accordance with this Law. The data subject has the right to request the review of decisions taken solely on the basis of automated processing of personal data affecting his interests, including decisions to define his personal, professional, consumer and credit profile or aspects of his personality.
Chile	AI Policy and Action Plan (2021)	People-centred AI: AI should contribute to people’s well-being and avoid direct or indirect harm. Inclusive AI: AI must not arbitrarily discriminate or be used to the detriment of any group, should incorporate a gender perspective, ad should consider children and teenagers.
Colombia	AI Ethical Framework (2021)	Privacy: “Artificial intelligence must be preceded by a respect for the privacy of individuals and their private sphere that prevents the use of information that they have not authorized and profiling” (p.31). Human control of decisions of an AI system (human-in-the-loop and human-over-the-loop): humans having control of data collection, decision making and systems that interact with citizens. The principle suggests a strict standard while AI technology matures in the country (p. 32). Non-discrimination: “Artificial intelligence systems cannot have outcomes or responses that under- mine the well-being of a specific group or limit the rights of historically marginalize populations” (p.36). This principle is applied to data, algorithms and practices.
Ecuador	Guide for the Processing of Personal Data in the Central Public Administration	Non-discrimination: personal data treatment cannot originate discrimination of any kind (see section 8).
Jamaica	Data Protection Act (2020)	Rights in relation to automated decision-making: “An individual is entitled, at any time (…) to require the data controller to ensure that no decision by which this section applies is based solely on the processing, by automatic means, of personal data in respect of the data subject for the purpose of evaluating matter relating to the data subject (e.g. the individual’s performance at work, creditworthiness, reliability or conduct).
Panama	Data Protection Law (2019)	“The owner of the data has the right not to be the subject of a decision based solely on the automated processing of their personal data, which produces negative legal effects or is detrimental to a right, whose purpose is to evaluate certain aspects of their personality, status health, job performance, credit, reliability, behaviour, characteristics or personality, among others.”
Peru	First draft provided for OECD review in May 2021.	Non-discrimination: Draft commits that public bodies will develop an impact study on achieving less bias in the use of algorithms for the classification of people in the private sector (e.g. AI systems that classify people for providing benefits, opportunities or sanctions). All AI use cases of this type would need to undergo a socioeconomic impact study to guarantee equity.
Mexico	AI Principles	Protect the will and freedoms of people in the implementation of AI, seeking respect for the integrity and the right to self-determination of individuals. Prioritise safety, integrity and human dignity in the design and application of AI tools, evaluating and monitoring the factors that may impact on it. To promote equality by reducing the possibility of discriminatory biases in the data and models used. Promote justice by offering human-operated mechanisms to review, appeal, and correct decisions or actions taken by AI-based systems. Ensure privacy by incorporating, from the design, consent mechanisms and personal control over the use of data.
Uruguay	AI Strategy	General interest: AI-based solutions promoted by the State should tend to protect the general interest, guaranteeing inclusion and equity. For this, work must be carried out specifically to reduce the possibility of unwanted biases in the data and models used that may negatively impact people or favour discriminatory practices. Respect for Human Rights: Any technological solution that uses AI must respect Human Rights, individual freedoms and diversity. Ethics: When the application and/or development of AI-based solutions present ethical dilemmas, they must be addressed and resolved by human beings. Privacy by design: AI solutions should contemplate people ́s privacy from their design stage. Personal Data Protection principles in force in Uruguay are considered components of this document.

Source: OECD review of national AI principles and related instruments.

Table A B.3.  Examples of LAC AI instruments alignedwith OECD AI principle #3

	Reviewed Instrument	OECD Principle: Transparency and explainability
Barbados	Data Protection Act (2019)	On Transparent information: “The data controller shall take appropriate measures to provide any information (...) relating to processing to the data subject in a concise, transparent, intelligible and easily accessible form, using clear and plain language,” (p. 37). Right of Access: where personal data of the data subject is being processed, [the data subject has the right] to request from, and to be given (...) a description of (...) the existence of automated decision-making, including profiling, referred to in section 18 and, at least in those cases, meaningful information about the logic involved, as well as the significance and the envisaged consequences of such processing for the data subject.
Brazil	Brazilian Artificial Intelligence Strategy (2021)	Transparency. Action to direct funds projects supporting transparency, and to stimulate transparency and responsible disclosure. Enable supervisory mechanisms for public scrutiny on AI activities. Action in public procurement to include criteria related to the incorporation of ethical principles related to transparency. Public scrutiny – Commitment to enable supervisory mechanisms for public scrutiny on AI activities Access. Commitment to stimulate disclosure of open source data and code.
Brazil	General Law on the Protection of Personal Data (2018)	Free access: guarantee, to the holders, of easy and free consultation on the form and duration of the processing, as well as on the completeness of their personal data; Transparency: guarantee, to holders, clear, accurate and easily accessible information on the performance of the treatment and its agents of treatment, in the light of trade and industrial secrets; The controller shall provide, whenever requested, clear and appropriate information regarding the criteria and procedures used for the automated decision, in compliance with trade and industrial secrets.
Chile	AI Policy and Action Plan (2021)	Safe AI: Due to the close dependence that exists between algorithm training and the use of data for it, transparency and explicability become relevant elements for the conception of a safe AI.
Colombia	AI Ethical Framework (2021)	Transparency: “openness to provide meaningful and understandable information about the design, operation, and impact of artificial intelligence systems” (p.29). Considers the sources of information, the model behind the algorithms, algorithm accuracy, and information about the design and development teams. Explanation: “Both for the developers and users of the system and for those individuals who may be affected by its decisions and results. The information should be easily accessible and understandable, in order to promote active citizen participation in the design, implementation and evaluation of AI systems” (p.30). Thus, it refers to accessible information over the importance of data for the system, how it is collected, the purposes for data collection, algorithms’ initial objectives and expected and obtained results, and roles of the people involved.
Jamaica	Data Protection Act (2020)	Rights of access to personal data: “an individual is entitled (…): -to be informed by the data controller, free of charge, whether personal data of which that individual is the personal data subject are being processed by or on behalf of that data controller; -if personal that are being processed as described in paragraph (a), to be given by the data controller, free of charge, a description of: personal data of which that individual is the data subject; the purposes for which personal data are being, or are to be, processed; and the recipients or classes of recipients to whom they are, or may be, disclosed; (...). -where the processing of personal data by automatic means of which that individual is the data subject: is for the purpose of evaluating matters related to that individual (such as, for example, individual’s performance at work, creditworthiness, reliability or conduct), and has constituted or is likely to constitute the sole basis for any decision significantly affecting the individual, to be informed by the data controlled, upon payment of the prescribed fee, of the logic involved in that decision-taking.”
Mexico	AI Principles	Explain to users who interact with AI-based services, the decision process that the system makes about them, the expected benefits and the potential risks associated with their use. Foster openness and trust by sharing as much information as possible to the public, preserving the required confidentiality, which allows understanding the training method and the decision-making model of the system. Publish the results of the system evaluations, trying to include versions aimed at different audiences.
Peru	First draft provided for OECD review in May 2021.	Transparency: Draft commits to implementing a platform that will serve as a registry of AI algorithms and associated data in the public sector.
Uruguay	AI Strategy	Transparency: AI solutions used in the public sphere must be transparent and comply with the regulations in force. This transparency must: -Make available the algorithms and data used for training the solution and its implementation, as well as the tests and validations performed. -Explicitly make visible, through active transparency mechanisms, all those processes that use AI, either in the generation of public services or in support of decision-making processes.

Source: OECD review of national AI principles and related instruments.

Table A B.4.  Examples of LAC AI instruments alignment with OECD AI principle #4

	Reviewed Instrument	OECD Principle: Robustness, security and safety
Brazil	Brazilian Artificial Intelligence Strategy (2021)	Security. Inclusion of a strategic axis on public security. Commitment to develop techniques to identify and mitigate algorithmic bias and ensure data quality for training of AI systems. Promotion of voluntary standards to manage AI risks. Action to conduct impact analyses on AI uses that directly affect the public. Commitment to implement a privacy and data protection sandbox for AI systems aimed at public security.
Brazil	General Law on the Protection of Personal Data (2018)	Prevention: adoption of measures to prevent the occurrence of damage due to the processing of personal data.
Chile	AI Policy and Action Plan (2021)	Safe AI: -Integrity, quality and security of the data. -Risk and vulnerability assessments. -Cybersecurity: confidentiality, integrity and protection of data, algorithms, processes and practices.
Colombia	AI Ethical Framework (2021)	Security: “Artificial intelligence systems must not affect the integrity and physical and mental health of the human beings with whom they interact” (p. 34). This principle includes measures such as immutability, confidentiality and integrity of base data, and establishment of codes of conduct and systems of risk to establish possible affectations.
Mexico	AI Principles	Mitigate situations of risk and uncertainty on an ongoing basis and generate response strategies to the effects derived from the use of AI. Prioritise safety, integrity and human dignity in the design and application of AI tools, evaluating and monitoring the factors that may impact on it. Establish solid mechanisms for the protection of user data, especially sensitive attributes, and notify about possible eventualities in their handling.
Uruguay	AI Strategy	Security: AI developments must comply, from their design, with the basic principles of information security. The guidelines and regulations related to cybersecurity in force in Uruguay that apply to the development of AI are considered components of this document.

Source: OECD review of national AI principles and related instruments.

Table A B.5.  Examples of LAC AI instruments alignment with OECD AI principle #5

	Reviewed Instrument	OECD Principle: Accountability
Brazil	Brazilian Artificial Intelligence Strategy (2021)	Accountability. Action to direct funds to projects supporting accountability. Commitment to encourage accountability practices for AI.
Brazil	General Law on the Protection of Personal Data (2018)	Responsibility and accountability: demonstration by the agent of the adoption of effective measures capable of proving observance of and compliance with the rules of protection of personal data and even the effectiveness of such measures.
Colombia	AI Ethical Framework (2021)	Responsibility: “There is a duty to respond for the results produced by an artificial intelligence system and the affectations it generates” (p.33). Responsibility from entities that collect and process data, and people who design an algorithm. It also suggests defining clear responsibilities on the chain of design, production and implementation.
Mexico	AI Principles	Determine clear responsibilities and obligations of different actors regarding the process of design, development, implementation and use of technology.
Peru	First draft provided for OECD review in May 2021.	Responsibility: Draft commits that the country will adopt ethical guidelines for sustainable, transparent and replicable use of AI with clear definitions of responsibilities and data protection.
Uruguay	AI Strategy	Responsibility: Technological solutions based on AI must have a clearly identifiable person responsible for the actions derived from the solution actions.

Source: OECD review of national AI principles and related instruments.