Principles

The UAE AI Ethics Principles and Guidelines give eight principles of AI ethics and guidelines for each principle (§ 1.1). The sections below show where the guidelines apply, the two types of organization, the decision tests, and the guidelines of each principle.

Before you start

The UAE AI Ethics Principles and Guidelines (UAE AI Office, DEC 2022) are "non-binding" (§ 1.1) and "non-mandatory" (§ 1.2). For more information, see Non-binding guidelines and Sources.

Most guidelines use "should", "should consider", or "could consider". Some guidelines use "must", but these guidelines are also non-binding.

Where the guidelines apply

The UAE AI Ethics Principles and Guidelines are for AI systems in "both the public and private sectors" (§ 1.2). In § 1.2, the guidelines are "applicable only to those AI systems which make or inform ‘significant decisions’". The guidelines also apply to critical decisions, which are "a subset of significant decisions" (§ 1.2). For the decision tests, see Significant decisions.

AI systems and AI subjects

Section 2, "Definitions", of the UAE AI Ethics Principles and Guidelines gives this information:

Name	Section 2
Artificial intelligence (AI), § 2.3	"The capability of a functional unit to perform functions that are generally associated with human intelligence such as reasoning, learning and self-improvement"
Functional unit, § 2.7	"Entity of hardware or software, or both, capable of accomplishing a specified purpose"
Artificially intelligent system (AI system), § 2.4	"A product, service, process or decision-making methodology whose operation or outcome is materially influenced by artificially intelligent functional units."
Subject of an artificially intelligent system (AI subject), § 2.12	"A natural person who is any of the following": "an end-user of an AI system", "directly affected by the operation of or outcomes of an AI system", or "a recipient of a service or recommendation provided by an AI system"

Section 2.4 gives this information about an AI system:

• In § 2.4, Note 1, it is "not necessary for a system’s outcome to be solely determined by artificially intelligent functional units".
• In Example 1 of § 2.4, AI gives a recommended decision, and "The judge determines the final outcome". But this decision method is an AI system.
• In Example 2 of § 2.4, "the chatbot is not an AI system" because it uses "pre-written responses" and "pre-programmed decision rules". But "If, however, the chatbot autonomously adjusted its treatment of customers based on the outcome of past cases, it would be an AI system."

The eight principles

Section 3 of the UAE AI Ethics Principles and Guidelines gives each principle and its guidelines:

Section	Name
§ 3.1	Fairness
§ 3.2	Accountable AI
§ 3.3	Transparent AI
§ 3.4	Explainable AI
§ 3.5	Robust, Safe and Secure AI
§ 3.6	Human Centered AI
§ 3.7	Sustainable and Environmentally Friendly AI
§ 3.8	Privacy Preserving AI

For the Modulos Requirement of each principle, see Requirements.

AI developer organizations and AI operator organizations

The UAE AI Ethics Principles and Guidelines give guidelines to AI developer organizations (§ 2.1) and AI operator organizations (§ 2.2). The guidelines also use "AI developers" and "AI operators" for these organizations. Some guidelines name AI developers, AI operators or organizations. Other guidelines are for stakeholders, persons who use AI systems, or other persons. Some name no organization.

In § 2.1, an AI developer organization is "An organization which does any of the following":

• "determines the purpose of an AI system"
• "designs an AI system"
• "builds an AI system"
• "performs technical maintenance or tuning on an AI system"

Section 2.1 also gives this information: "The definition applies regardless of whether the organization is the ultimate user of the system, or whether they sell it on or give it away." In the example of § 2.1, "A company develops an artificially intelligent facial recognition system and sells it to a country’s border force". The example gives this result: "The company is an AI developer organization and the border force is an AI operator organization."

In § 2.2, an AI operator organization is "An organization which does any of the following":

• "uses AI systems in operations, backroom processes or decision-making"
• "uses an AI system to provide a service to an AI subject"
• "is a business owner of an AI system"
• "procures and treats data for use in an AI system"
• "evaluates the use case for an AI system and decides whether to proceed"

Section 2.2 adds these two notes:

• "This definition applies regardless of whether the AI system was developed in-house or procured."
• "It is possible for organizations to be both an AI developer organization and an AI operator organization."

One organization can be an AI developer organization and an AI operator organization (§ 2.2, Note 2). Read the guidelines for each type. AI developer organization and AI operator organization are types of organization in the guidelines, and they are not the Organization level of Modulos. In Modulos, UAE AI Ethics is MFF-4 at the AI System level and OFF-4 at the Organization level (see Recommended AI System structure).

Significant decisions

The UAE AI Ethics Principles and Guidelines apply only to AI systems that "make or inform" significant decisions (§ 1.2). Section 2 gives these decision tests:

Name	Section 2	Notes and examples
Individually significant decision, § 2.8	"A decision which has the potential for significant impact on at least one individual’s circumstances, behavior or choices, or has legal or similarly significant effects on him or her."	Example: "A company decides to make an employee redundant."
Set of significant-at-scale decisions, § 2.10	"A set of decisions made by the same system or organization which, when taken in aggregate, have significant impact on society as a whole or groups within it."	"The decisions need not be individually significant in order to qualify" (§ 2.10). Example: "An AI system is used by a website to determine which content to show users." If "the website is popular", these decisions can be a set of significant-at-scale decisions.
Significant decision, § 2.11	"A decision which is either individually significant or is part of a set of significant-at-scale decisions."	A significant decision is a decision of one of the two types above.
Critical decision, § 2.6	"An individually significant decision which is deemed to either have a very large impact on an individual or to have especially high stakes."	Critical decisions "can be especially sensitive, have the potential to cause high loss or damage, be societally significant, or set an important precedent". Example: a court decision with "a life sentence".

Section 2.10 also gives examples of areas with a "large impact on society". These areas include "the large-scale allocation of resources or opportunities amongst groups", "the structure of government", "the law, and its interpretation and enforcement", "conflict and war", and "international relations".

A critical decision is an individually significant decision with "a very large impact" or "especially high stakes". Thus, not all significant decisions are critical decisions.

Some guidelines apply only to some decisions or systems:

• Critical decisions — See Principle 2 for accountability, transparency, and external audit (§§ 3.2.2.7(a), 3.2.2.3(a)).
• Critical decisions that are "life and death decisions" — See Principle 2 for checks by a person (§ 3.2.2.3(c)).
• Decisions "related to lethal force, life and death" — See Principle 6 for human control (§ 3.6.2.3(a)).
• AI systems that "may directly impact people’s lives in a significant way" — See Principle 5 for control by persons (§ 3.5.2.4(d)–(e)).

Principle 1: Fairness

The UAE AI Ethics Principles and Guidelines give guidelines to AI developers, AI operators and organizations in § 3.1 (Fairness). The principle is: "We will make AI systems fair". The principle (§ 3.1.1) gives these guidelines:

• "Data ingested should, where possible, be accurate and representative of the affected population."
• "Algorithms should avoid non-operational bias." In § 2.9, non-operational bias is bias that is "not a design feature" or "not important in achieving the stated purpose of the system".
• "Steps should be taken to mitigate and disclose the biases inherent in datasets."
• "Significant decisions should be provably fair."
• "All personnel involved in the development, deployment and use of AI systems have a role and responsibility to operationalize AI fairness and should be educated accordingly."

Fairness of decisions

The UAE AI Ethics Principles and Guidelines start the fairness guidelines with § 3.1.2.1: "AI systems should not discriminate against people or groups in a way that could have an adverse impact." The same guideline continues: "Especially significant and critical decisions made or assisted by AI should be provably fair." Section 3.1.2.1 gives these guidelines:

• (a) "an organization should document its own definition of fairness for the context that the AI system is going to be implemented in". Section 3.1.2.1(a) continues: "Organizations should document what the implemented fairness objective stands for and why this choice was considered the most suitable for the given scenario."
• (b) "It is recommended to identify and document demographic groups that may be adversely impacted and mitigate the risk where possible."
• (c) "AI developers and AI operators could consider formal procedures such as Discrimination Impact Assessments as a means of ensuring fairness."

For a Discrimination Impact Assessment, § 3.1.2.1(c) also gives these guidelines:

• "This assessment should be documented as well as the mitigation measures that were implemented by the organization."
• "The impact assessment should be conducted pre-release and regularly after as an ongoing evaluation, results are advised to be documented."

Data and bias

The UAE AI Ethics Principles and Guidelines give this information in § 3.1.2.2(d): "Fairness has many different definitions in different cultures and for different contexts." The guideline continues: "Encouraging a diverse and inclusive AI ecosystem is thus all the more crucial to ensure that one definition of fairness does not contradict another, and that the process of defining fairness itself is fair, with under-represented groups represented in the discussion."

The UAE AI Ethics Principles and Guidelines give these data and bias guidelines to AI developers and AI operators (§§ 3.1.2.2 and 3.1.2.3):

• "AI developers and operators should evaluate all datasets to assess inclusiveness of identified demographic groups and collect data to close any gaps."
• "AI developers and AI operators should refrain from training AI systems on data that is not likely to be representative of the affected AI subjects, or is not likely to be accurate, whether that be due to age, omission, method of collection, or other factors."
• "AI developers and AI operators should undertake reasonable data exploration and/or testing to identify potentially prejudicial decision-making tendencies in AI systems arising from data inaccuracy and biases in the data."
• "When evaluating the fairness of an AI system, AI developers and AI operators should consider whether AI subjects in the same circumstances receive equal treatment."

The same guidelines give two guidelines to AI developers only:

• "When subjecting different groups to different decision-making processes, AI developers should consider whether this will lead to non-operational bias."
• "AI developers should consider whether their AI systems can be expected to perform well when exposed to previously unseen data, especially when evaluating people who are not well-represented in the training data."

Access and different groups

The UAE AI Ethics Principles and Guidelines give these guidelines about access and about groups of persons:

• "AI operators should consider whether their AI systems are accessible and usable in a fair manner across user groups" (§ 3.1.2.4, the guideline about access).
• "Organizations should seek to include people from diverse demographic backgrounds across the full lifecyle, to include design, development and deployment processes." The same guideline continues: "Organizations should seek to engage diverse internal and external groups also." (§ 3.1.2.5)
• "AI developers should consider whether the assumptions they make about AI subjects could be wrong or are likely to lead to non-operational bias" (§ 3.1.2.5). In that condition, "they should consider consulting the AI subjects in a representative manner during the design, development and deployment to confirm these assumptions".

Principle 2: Accountable AI

The UAE AI Ethics Principles and Guidelines give guidelines to AI developers and AI operators in § 3.2 (Accountable AI). The principle is: "We will make AI systems accountable". In § 3.2.1, "Accountability for the outcomes of an AI system lies not with the system itself but is apportioned between those who design, develop and deploy it". The principle also gives these guidelines:

• "Developers should make efforts to mitigate the risks inherent in the systems they design"
• "AI systems should have built-in appeals procedures whereby users can challenge significant decisions"
• "AI systems should be developed by diverse teams which include experts in the area in which the system will be deployed"
• "AI systems should be subject to external audit and decision quality assurance."

Damage and risks

The UAE AI Ethics Principles and Guidelines give these guidelines about damage (§ 3.2.2.1):

• (a) "Accountability for loss or damages resulting from the application of AI systems should not be attributed to the system itself."
• (b) "AI operators and AI developers should consider designating individuals to be responsible for investigating and rectifying the cause of loss or damage arising from the deployment of AI systems."

Section 3.2.2.2 of the guidelines gives these guidelines about risks:

• (a) "AI operators should only use AI systems that are backed by evidence-based academic or industrial research, and AI developers should base their development on such research."
• (b) "AI operators should identify the likely impact of incorrect automated decisions on AI subjects and, in the case where incorrect decisions are likely to cause significant cost or inconvenience, consider mitigating measures."
• (c) "AI operators should consider internal risk assessments or ethics frameworks as a means to facilitate the identification of risks and mitigating measures."
• (d) "In designing AI systems to inform significant decisions, AI developers should consider measures to maintain data accuracy over time". These measures include "the completeness of the data", the "timely update of the data", and "whether the context in which the data was collected affects its suitability for the intended use case".
• (e) "AI developers and AI operators should tune AI models periodically to cater for changes to data and/or models over time." The guidelines set no time interval.
• (f) "AI operators should consider whether AI systems trained in a static environment will display model instability when deployed in dynamic environments."
• (g) "AI developers should collaborate with AI Operators to train models using historical data from the Operator."
• (h) "AI Operators should consider working with their vendors (AI developers) to continually monitor performance."
• (i) "AI Operators should subject AI systems informing significant decisions to quality checks at least as stringent as those that would be required of a human being taking the same decision."

External audit for critical decisions

The UAE AI Ethics Principles and Guidelines start § 3.2.2.3 with this guideline: "AI systems informing critical decisions should be subject to appropriate external audit". Each guideline below has a different condition:

• "When AI systems are used for critical decisions, external auditing of the AI systems in question should be used as a means to ensure meaningful standards of transparency and accountability are upheld."
• "In the case that critical decisions are of civic interest, public release of the results of the audit should be considered as a means of ensuring public processes remain accountable to those affected by them."
• If critical decisions are "life and death decisions", "these decisions should be supported by further validation and verification via a human operator" (§ 3.2.2.3(c)).
• "Facilitate traceability and auditability of AI systems, particularly in critical contexts or situations."

Section 3.2.2.3(e) to (g) also gives these guidelines:

• "Adopt a Trustworthy AI assessment list when developing, deploying or using AI systems, and adapt it to the specific use case in which the system is being applied."
• "Keep in mind that such an assessment list will never be exhaustive." The same guideline also gives this information: "Ensuring Trustworthy AI is not about ticking boxes, but about continuously identifying and implementing requirements, evaluating solutions, ensuring improved outcomes throughout the AI system’s lifecycle, and involving stakeholders in this."
• "Be mindful that there might be fundamental tensions between different principles and requirements. Continuously identify, evaluate, document and communicate these tradeoffs and their solutions."

Challenge procedures and opt-out mechanism

The UAE AI Ethics Principles and Guidelines start § 3.2.2.4 with this guideline: "AI subjects should be able to challenge significant automated decisions concerning them and, where appropriate, be able to opt out of such decisions". The guideline gives these guidelines to AI operators:

• (a) "AI operators using AI systems to inform significant decisions should provide procedures by which affected AI subjects can challenge a specific decision concerning them."
• (b) "AI operators should consider such procedures even for non-significant decisions."
• (c) "AI operators should make affected AI subjects aware of these procedures and should design them to be convenient and user-friendly."
• (d) "AI operators should consider employing human case evaluators to review any such challenges and, when appropriate, overturn the challenged decision."
• (e) "AI operators should consider instituting an opt-out mechanism for significant automated decisions."
• (f) "AI operators could consider “crowd challenge” mechanisms whereby a critical number of complaints triggers an investigation into the fairness and/or accuracy of a decision-making process as a whole."

Section 3.2.2.4 recommends that AI subjects be able to opt out of significant automated decisions about them "where appropriate". The opt-out mechanism of § 3.2.2.4(e) is for "significant automated decisions", not only for critical decisions.

Other accountability guidelines

The UAE AI Ethics Principles and Guidelines also give these accountability guidelines:

• § 3.2.2.5(a) — "When informing an AI subject about significant choices they will make, AI systems should not unreasonably restrict the available options or otherwise attempt to influence their value judgements without the explicit consent of the AI subject in question."
• § 3.2.2.6(a) — "AI developers who develop AI systems which may be used to assist in making critical decisions should involve in the process experts with a background in social science, policy, or another subject which prepares them to evaluate the broader societal impact of their work."
• § 3.2.2.6(b) — "Development of AI systems informing significant decisions should include consultation with experts in the field in which the system will be deployed."
• § 3.2.2.7(a) — "In the case of critical decisions, AI operators should avoid using AI systems that cannot be subjected to meaningful standards of accountability and transparency."
• § 3.2.2.7(b) — "AI developers should consider notifying customers and AI operators of the use cases for which the system has been designed, and those for which it is not suitable."

Principle 3: Transparent AI

The UAE AI Ethics Principles and Guidelines give guidelines to AI developers, AI operators and organizations in § 3.3 (Transparent AI). The principle is: "We will make AI systems transparent". The principle (§ 3.3.1) gives these guidelines:

• "Developers should build systems whose failures can be traced and diagnosed."
• "People should be told when significant decisions about them are being made by AI."
• "Within the limits of privacy and the preservation of intellectual property, those who deploy AI systems should be transparent about the data and algorithms they use."
• "Responsible disclosures should be provided in a timely manner, and provide reasonable justifications for AI systems outcomes."

Traceability and documentation

The UAE AI Ethics Principles and Guidelines start § 3.3.2.1 with this guideline: "Traceability should be considered for significant decisions, especially those that have the potential to result in loss, harm or damage." Section 3.3.2.1 gives these guidelines:

• (a) "For AI systems which inform significant decisions, especially those with the potential to cause loss, harm or damage, AI developers should consider building in traceability". In this guideline, traceability is "the ability to trace the key factors leading to any specific decision".
• (b) "Organizations should ensure that harms caused through AI systems are investigated and redressed, by enacting strong enforcement mechanisms and remedial actions, to make certain that human rights and the rule of law are respected in the digital world and in the physical world."
• (c) "AI developers and AI operators should consider documenting the following information during the design, development and deployment phases, and retaining this documentation for a length of time appropriate to the decision type or industry".
• (d) "Where possible given the model design, AI developers should consider building in a means by which the “decision journey” of a specific outcome (i.e. the component decisions leading to it) can be logged."

The documentation of § 3.3.2.1(c) includes these items:

• "the provenance of the training data, the methods of collection and treatment, how the data was moved, and measures taken to maintain its accuracy over time"
• "the model design and algorithms employed"
• "changes to the codebase, and authorship of those changes"

Information for AI subjects

The UAE AI Ethics Principles and Guidelines start § 3.3.2.2 with this guideline: "People should be informed of the extent of their interaction with AI systems". Section 3.3.2.2 gives these guidelines:

• (a) "AI operators should inform AI subjects when a significant decision affecting them has been made by an AI system."
• (b) "If an AI system can convincingly impersonate a human being, it should do so only after notifying the AI subject that it is an AI system."

Principle 4: Explainable AI

The UAE AI Ethics Principles and Guidelines give guidelines to AI operators in § 3.4 (Explainable AI). The principle is: "We will make AI systems as explainable as technically possible". The principle (§ 3.4.1) gives these guidelines:

• "Decisions and methodologies of AI systems which have a significant effect on individuals should be explainable to them, to the extent permitted by available technology."
• "It should be possible to ascertain the key factors leading to any specific decision that could have a significant effect on an individual."
• "In the above situation we will provide channels through which people can request such explanations."

Explanations of how the AI system operates

The UAE AI Ethics Principles and Guidelines start § 3.4.2.1 with this guideline: "AI operators could consider providing affected AI subjects with a high-level explanation of how their AI system works". Section 3.4.2.1 gives these guidelines:

• (a) "AI operators could consider informing the affected AI subjects in understandable, nontechnical language of" these items: "the data that is ingested by the system", "the types of algorithms employed", "the categories into which people can be placed", and "the most important features driving the outcomes of decisions". The last item is "A comprehensive list of feature engineering and models that was considered during the model building phase."
• (b) "For non-sensitive public sector use cases designed for the common good, AI operators could consider making source code, together with an explanation of the workings of the AI system, available either publicly or upon request". The guideline adds this condition: "this should be done only if there is low risk of people ‘gaming the system’".
• (c) "The AI operator should maintain necessary documentation that provides elaboration and clarification on how the algorithm works, for example documentation of processes, decision making flow charts of the system, and how the appeal process is embedded."
• (d) "An individual should have the ability to contest and seek effective redress against decisions made by AI systems. These must be addressed by the group or team supporting these models."
• (e) "Develop appropriate impact indices for the evaluation of AI system technological interventions from multiple perspectives."

The "must" in § 3.4.2.1(d) is part of the non-binding guidelines. For the challenge procedures of AI operators, see Challenge procedures and opt-out mechanism.

Explanations of significant decisions

The UAE AI Ethics Principles and Guidelines start § 3.4.2.2 with this guideline: "AI operators should consider providing affected AI subjects with a means to request explanations for specific significant decisions, to the extent possible given the state of present research and the choice of model". Section 3.4.2.2 gives these guidelines:

• (a) "AI operators should consider providing a means by which people affected by a significant decision informed by AI can access the reasoning behind that decision."
• (b) "Where such explainability is not possible given available technology, AI operators should consider compromises such as counterfactual reasoning, or listing the most heavily weighted factors contributing to the decision."
• (c) "In the case that such explanations are available, they should be easily and quickly accessible, free of charge and user-friendly."

The guideline about "further validation and verification via a human operator" is different. It applies only to critical decisions that are "life and death decisions" (see External audit for critical decisions).

Principle 5: Robust, safe and secure AI

The UAE AI Ethics Principles and Guidelines give guidelines to AI developers and AI operators in § 3.5 (Robust, Safe and Secure AI). The principle is: "AI systems will be technically robust". The principle (§ 3.5.1) gives these guidelines:

• "AI systems should be technically robust with a preventative approach to risks which operates in a manner such that they reliably behave."
• "AI Developers should ensure AI systems will not cause any unintentional harm and adverse impacts."
• "AI systems should be resilient to attacks and security such as data poisoning and model leakage."
• "AI systems should have safeguards that enable a fallback plan in case of problems."
• "AI system results should be reproducible."

Vulnerability assessments and fallback plans

The UAE AI Ethics Principles and Guidelines give these guidelines to AI operators about vulnerability assessments and fallback plans:

• § 3.5.2.1 — "AI Operators should continue conducting vulnerability assessments, verification of AI system different behaviors in unexpected situations and for any dual-use case". These include "integrity and resilience" against "potential attacks", "Verifying how AI systems behave in unexpected situations and environments", and "Taking preventative measures against any possible system dual-case use".
• § 3.5.2.2 — "AI operators should define a suitable fall back plan and test it to maintain readiness in unexpected situations and in high safety risks level".

The fallback plan of § 3.5.2.2 includes these items:

• "Developing a plan to measure and assess potential safety risks to you or any third party from technology use accidental or malicious misuse."
• "Define thresholds for system acceptable results and governance procure to fall back on alternative defined and tested plans." For AI operators, this recommends "thresholds for system acceptable results" and a procedure to use "alternative defined and tested plans".
• "Considering an insurance policy to mitigate risks arising from potential damages."

Documentation and test results

The UAE AI Ethics Principles and Guidelines start § 3.5.2.3 with this guideline: "AI Operators should assure end users of the system’s reliability through documentation, and operationalizing processes for testing and verification of desired outcomes". Section 3.5.2.3 gives these guidelines:

• (a) "AI system should be designed with an approach to continue monitoring if the system meets goals, purposes and intended application."
• (b) "AI Operators should define an approach to ensure results are reproducible through clear process and documentation."
• (c) "AI Operators should publish documentation to assure system robustness to end users."

Safety, security, and control by persons

The UAE AI Ethics Principles and Guidelines start § 3.5.2.4 with this guideline: "AI systems will be safe, secure and controllable by humans". Section 3.5.2.4 gives these guidelines:

• (a) "Safety and security of people, be they operators, end-users or other parties, will be of paramount concern in the design of any AI system."
• (b) "AI systems should be verifiably secure and controllable throughout their operational lifetime, to the extent permitted by technology."
• (c) "The continued security and privacy of users should be considered when decommissioning AI systems."
• (d) "AI systems that may directly impact people’s lives in a significant way should receive commensurate care in their designs."
• (e) "Such systems should be able to be overridden or their decisions reversed by designated individuals."

For systems that "may directly impact people’s lives in a significant way", § 3.5.2.4(e) recommends that they can be "overridden or their decisions reversed by designated individuals". Section 3.5.2.5(a) gives this guideline: "AI systems should be built to serve and inform, and not to deceive and manipulate." Section 3.5.2.5(d) also gives this guideline: "Systems designed to inform significant decisions should do so impartially."

Principle 6: Human-centered AI

The UAE AI Ethics Principles and Guidelines give guidelines to AI developers, organizations and stakeholders in § 3.6 (Human Centered AI). The principle is: "We will give AI systems human values and make them beneficial to society". The principle (§ 3.6.1) also gives guidelines about stakeholders, norms, nudging, and decisions "related to lethal force, life and death". Section 3.6.2 gives more information about these items.

Stakeholders

The UAE AI Ethics Principles and Guidelines give these guidelines about stakeholders (§ 3.6.2.1):

• (a) "Society should be consulted in a representative fashion to inform the development of AI."
• (b) "Stakeholders throughout society should be involved in the development of AI and its governance."
• (d) "Organizations should prioritize having all their stakeholders learn about wellbeing metrics as a potential determinant of how they create, deploy, market and monitor their AI technologies."

Nudging and affective systems

The UAE AI Ethics Principles and Guidelines start § 3.6.2.2 with this guideline: "Systematic analyses that examine the ethics of designing affective systems to nudge human beings prior to deployment are needed." Section 3.6.2.2 gives these guidelines:

• (a) "Nudging in AI systems should have an opt-in system policy with explicit consent."
• (b) "We recommend that when appropriate, an affective system that nudges human beings should have the ability to accurately distinguish between users, including detecting characteristics such as whether the user is an adult or a child."
• (b), continued — "Additional protections must be put in place for vulnerable populations, such as children, when informed consent cannot be obtained, or when it may not be a sufficient safeguard."
• (c) "AI systems with nudging strategies must be carefully evaluated, monitored, and controlled."

The "must" in § 3.6.2.2(b) and (c) is part of the non-binding guidelines.

Human control of decisions

The UAE AI Ethics Principles and Guidelines start § 3.6.2.3 with this guideline: "Humanity should retain the power to govern itself and make the final decision, with AI in an assisting role." Section 3.6.2.3 gives these guidelines:

• (a) "Decisions related to lethal force, life and death should not be delegated to AI systems. Rules and standards should be adopted to ensure effective human control over those decisions."
• (b) "Responsible parties (e.g., parents, nurse practitioners, social workers, and governments) should be trained to detect the influence due to AI and ineffective mitigation techniques. In the most extreme cases it should always be possible to shut down harmful AI system."
• (c) "Those actions undertaken by an affective system that are most likely to generate an emotional response should be designed to be easily changed."
• (d) "Users should be able to make informed autonomous decisions regarding AI systems." The guideline continues: "They should be given appropriate knowledge and tools to comprehend and interact with AI systems to satisfactory degree and, where possible to reasonably self-assess or challenge the system."
• (f) "In situations where it is needed, human rights impact assessments and human rights due diligence, human determination codes of ethical conduct, or quality labels and certifications intended to promote human centered values and fairness should be considered."
• (g) "Mechanisms should be put into place to receive external feedback regarding AI systems that potentially infringe on fundamental rights."

Norms and values

The UAE AI Ethics Principles and Guidelines give these guidelines about norms and values (§ 3.6.2.4):

• (a) "AI Developers should design AI systems to adopt, learn and follow the norms and values of the community they serve."
• (b) "Organizations could identify the norms of the specific community in which AI systems are to be deployed in." The guideline continues: "These could be documented as well as how these norms are addressed by the AI system."

Principle 7: Sustainable and environmentally friendly AI

The UAE AI Ethics Principles and Guidelines give the Sustainable and Environmentally Friendly AI principle in § 3.7: "We will promote sustainable and environmentally friendly AI". The principle (§ 3.7.1) gives these guidelines:

• "AI systems should be used to benefit all human beings, including future generations."
• "An AI system’s development, deployment and use, should be assessed via critical examination of resource usage and energy consumption."
• "Mechanisms to measure environmental impact due to type of energy use and processing power provided by data centers should be established."

Energy consumption and carbon footprint

The UAE AI Ethics Principles and Guidelines start § 3.7.2.1 with this guideline: "Throughout the AI lifecycle, implementations should always be carried out after full understanding and acknowledgement of AI implications on sustainability and environment." Section 3.7.2.1 gives these guidelines:

• (b) "AI model development, and by extension commissioning, should therefore seek to balance technical performance with energy consumption and environmental impact."
• (c) "Efforts should be made to estimate and understand sustainability and environmental impact across the AI lifecycle". The examples are "the energy consumption costs associated with model training, and CO2 emissions and cloud compute costs associated with the deployment and running of the system".
• (d) "While there is no standardized means of reporting on the environmental impact of AI systems, where possible, model development and commissioning should make carbon impacts a core consideration alongside functional and business requirements."
• (e) "Smaller models should be considered – shrinking down the model size and using fewer compute cycles (to balance financial and performance costs with the end performance of the model)."
• (f) "Carbon awareness should be considered – adjusting the operational parameters of the AI system to dynamically select the best time and location for energy use from the grid can reduce its carbon footprint."

In § 3.7.2.1(c), the "energy consumption costs" and the "cloud compute costs" are examples of environmental impact.

Principle 8: Privacy-preserving AI

The UAE AI Ethics Principles and Guidelines give guidelines to AI developers and AI operators in § 3.8 (Privacy Preserving AI). The principle is: "We will respect people’s privacy". The principle (§ 3.8.1) gives these guidelines:

• "AI systems should respect privacy and use the minimum intrusion necessary."
• "AI systems should uphold high standards of data governance and security, protecting personal information."
• "Surveillance or other AI-driven technologies should not be deployed to the extent of violating internationally and/or UAE’s accepted standards of privacy and human dignity and people rights."
• "Privacy by design should be embedded in AI systems and where possible AI algorithm should have adequate privacy impact assessments."
• "Adequate data protection frameworks and governance mechanisms should be established in a multi-stakeholder approach and ensured throughout the life cycle of AI systems."
• "AI developers and operators should strive to strike the balance between privacy requirements, individual rights and innovation growth and society benefits."

The guidelines set no risk level for privacy impact assessments. For more information about privacy impact assessments, see § 3.8.2.2(f) in Data governance.

Privacy problems and consent

The UAE AI Ethics Principles and Guidelines start § 3.8.2.1 with this guideline: "Establish mechanism for users to flag issues related to privacy or data protection". Section 3.8.2.1 gives these guidelines:

• (a) "AI Operators should review the system for proper consent logging, ability of users to revoke permission whenever applicable."
• (b) "Consider training AI model without or with minimal user of potentially sensitive or personal data." Here, "minimal user" is minimum use of "potentially sensitive or personal data".
• (c) "Use measures to enhance privacy such as encryption, anonymization and aggregation."

Data governance

The UAE AI Ethics Principles and Guidelines start § 3.8.2.2 with this guideline: "Establish an oversight mechanism for data collection, storage and processing and use across your organization". Section 3.8.2.2 gives these guidelines:

• (a) "Assess who can access data and under which conditions."
• (b) "Assign specific responsibilities and role for Data Protection Officers."
• (c) "Prevention of harm to privacy necessitates adequate data governance that covers the quality and integrity of the data used, its relevance in light of the domain in which the AI systems will be deployed, its access protocols and the capability to process data in a manner that protects privacy."
• (d) "AI systems must guarantee privacy and data protection throughout a system’s entire lifecycle." This includes "the information initially provided by the user, as well as the information generated about the user over the course of their interaction with the system".
• (d), continued — "To allow individuals to trust the data gathering process, it must be ensured that data collected about them will not be used to unlawfully or unfairly discriminate against them."
• (e) "In any given organization that handles individuals’ data (whether someone is a user of the system or not), data protocols governing data access should be put in place in line with national privacy legislation (be these universal or sector-specific)." The guideline continues: "Only duly qualified personnel with the competence and need to access individual’s data should be allowed to do so."
• (f) "Algorithmic systems require adequate privacy impact assessments, which also include societal and ethical considerations of their use and an innovative use of the privacy by design approach."
• (g) "Establish data policies or equivalent frameworks, or reinforce existing ones, to ensure full security for personal data and sensitive data, which, if disclosed, may cause exceptional damage, injury or hardship to individuals."
• (h) "Promote mechanisms, such as open repositories for publicly funded or publicly held data, source code and data trusts, to support the safe, fair, legal and ethical sharing of data, among others."
• (i) "Promote and facilitate the use of quality and robust datasets for training, development and use of AI systems, and exercise vigilance in overseeing their collection and use."

The "must" in § 3.8.2.2(d) and the "require" in § 3.8.2.2(f) are part of the non-binding guidelines.

Sources

• UAE AI Ethics Principles and Guidelines — UAE AI Office, AI Ethics: Principles & Guidelines, DEC 2022, with "the UAE AI and Blockchain Council as its owner" (§ 1.5). Sections 1.1, 1.2, 2.1 to 2.12, and 3.1 to 3.8. PDF.

Related information

For the UAE AI Ethics Frameworks in Modulos, the Requirements, and the steps that Modulos recommends, see Use Modulos. For the FAQ, see UAE AI Ethics FAQ. For more information about the guidelines and their sources, see UAE AI Ethics.

Disclaimer

This page gives general information. It is not legal advice.