AI Life Cycle Core Principles

This table contains a list of AI core principles, which I will refer to as the “Core Principles”. Some of these were compiled using publications from UNESCO, IEEE ISO, NIST and FTC. These AI Core Principles can be applied throughout the AI lifecycle. This table can be used as a guideline by many stakeholders to help them develop common law (courts, regulations, drafting contracts and audit, design, etc.

The Core Principles can be divided into three groups: Operational (Green), Valuable (Orange), or Both (Blue). Operational refers to the Core Principle that is most closely related to the operation characteristics of the AI application. It is also related to the promotion of values/norms. The Core Principle is a combination of operational and value elements.

This table will be updated periodically.

Core principle	What it is and how to promote
1	Accessibility	User-friendly interface and experience (UI/UX), that is affordable; aids in understanding the algorithm.
2	Accountability	Examines output (decision-making and prediction); identifies gaps in predicted and actual outcomes.
3	Accuracy	Uses reliable data (authentic, nonrepudiated and protected against unauthorized modification or destruction); the dataset is derived using reasonable selection criteria to minimize harm; output measurement possible.
4	Bias	Protects against disparate impacts, discrimination against protected groups, and unjust outcomes; more generally, protects from inaccurate results; a subset ethics.
5	Big data	High-quality data used; compatible with decreasing dependence upon labeled architectures; maintains context relevance; promotes data accessibility.
6	Consent	It aligns with the consent of the end user to the application’s goals.
7	Collaboration	Facilitates global development; facilites information sharing between internal and externe, which translates into transparency.
8	Efficiency	Assists in a cost-effective training and time ratio; makes the best decisions regarding resource and objective utilization.
9	Enabling	Conforms to government-sponsored controlled environments for testing and scaling AI.
10	Equity	Protects against the widening of gender and protected class gap; maps to bias
11	Ethics	It encompasses a wide range of values that aim at eliminating or reducing risk to human life, privacy, and property, and to increase and maintain public trust.
12	Explainability (XAI).	Facilitates understanding of operations and outcomes; increases accountability; improves transparency.
13	Fairness	Favors treatment that is based on similar characteristics, policies and procedures. Reduces unintended disparate treatments. Uses anonymized or pseudonymized information.
14	Fidelity	Applicant’s performance can be measured relative to its code or across the deployment population. This supports measuring ongoing compliance with Core Principles.
15	Fundamental rights	Open data access is compliant. This contrasts with closed models (proprietary), which can restrict access. Maps to accessibility.
16	Governance	Developped in an environment that adheres to documented policies and procedures that ensure consistent data quality.
17	Human-centered	Compatible with law, privacy and democratic values; provides safeguards to ensure fair and just societies; protects from enhancing and perpetuating social inequality, promotes equality and social justice; aligns with best practices for user interface and experience (UI/UX); compatible with human-collaborative compatibility; compatible with experiential AI; maps to human-like dexterity in robotic applications.
18	All Inclusive	Contribution to society in large numbers; does not exclude some parts of society; is consistent with the Core Principles of ethics.
19	Interpretability	Interchangeable with explanation; maps to trust.
20	Metrics	Capable of measuring compliance with and effectiveness of the Core Principles; encourages standardization.
21	Permit	Permits are required for application development and access by end users.
22	Predictable	Maintains compatibility with the Core Principle throughout its entire lifecycle.
23	Privacy	Data protection maintained; compatible with Fair Information Principles and data minimization principles. Compatibility with and maintains anonymized, pseudonymized and encrypted data; resistance to re-identification
24	R&D	Promoting research and putting humans at the heart of AI development (also known as human-inthe-loop, see Experiential AI).
25	Reliability	Design, development and deployment are guided by best practices. They also promote Core Principles.
26	Resilience	Recoverable in the event of failure
27	Robust	Operates efficiently with minimal downtime; resists adversarial attacks; maintains operational integrity through its lifecycle; can identify and handle input/output reliability; resists unintended behavior by the end user; exhibits high level of problem flexibility; autonomous behavior maintains line-of-sight with human developers and end users; supports information sharing best practices; employs sophisticated learning techniques to minimize bias.
28	Safety	Avoid unintended behaviour; follow permit-related policies, procedures.
29	Security	Resistant against adversarial attacks, inference attacks; compatible information sharing best practices.
30	Sustainable	Promoting long-term growth; compatible with information sharing best practices.
31	Track record	Application is the result of a developer who has been known to design AI compatible with Core Principles.
32	Transparency	Encourages transparency, discovery, accessibility and non-discriminatory output; facilitates end-user understanding; encourages consensus; enables audit; compatible avec experiential AI.
33	Trustworthy	This is a catchall for accuracy, explainability and interpretability, privacy as well as reliability, robustness, safety and security (resilience), bias, and other related terms.
34	Truth	It does not lead to unfair or deceptive results.
35	Wherewithal	Developer is financially sound, can operate with resilience, and has implemented policies and procedures that fully support Core Principle compliant AI.
36	Workforce Compatibility	Take into account issues related to worker displacement; encourages effective worker interaction and training with AI.