AI and the Integrity of Values

By Hubert Saint-Onge

The concept of values alignment has emerged as a critical area of focus in the application of AI. It revolves around ensuring that the behaviours, decisions, and outcomes of AI systems align with the values of organizations and their people.

The integration of AI into organizational decision-making fundamentally shifts from a purely technical challenge to a governance, cultural and even personal one. We know AI has no conscience, yet the more we work with it, the less we tend to question the values embedded in its content. This can lead to discomfort among AI users when they encounter discrepancies between their values and AI's pronouncements.

The objective of this article is to shed light on how to align an organization's values with AI outputs. As we increasingly use powerful intelligent machines to perform tasks traditionally done by humans, how do we ensure we can recognize when AI outputs contradict our values? How do we make sure they bring integrity to the work we ask them to produce? How do we implement mechanisms that make people more aware of how they experience AI? How do we give them the confidence to intervene when they encounter an instance where AI's output is inconsistent with personal and organizational values?  

To ensure better alignment between AI and human-centred values, we need to improve integrity, coherence, complementarity, and trustworthiness in the interactions between humans and the AI tools we use. This article explores how we can best achieve values alignment with AI, emphasizing the integration of principles such as justice, privacy and fairness into AI outputs. To help achieve this, it is key to underscore the critical connection between value alignment and AI red lines (often referred to as guardrails): respecting these boundaries is essential to prevent unethical behaviour and maintain trust in AI technologies.

Dynamically linking AI to values

The words 'values' and 'ethics' are often used interchangeably as though they are synonymous. But it is essential to distinguish these two words. The best definition of values is that they represent 'prioritized ideals' embedded so deeply in our psyche that we often find it difficult to articulate them. Yet, whether we are conscious of them or not, our values guide our thinking, behaviours, and decisions. While values stem from ideals we aspire to, ethics emphasize what not to do. The configuration of our values serves as an internalized rudder, shaping the choices we make every day. Ethics, on the other hand, are articulated in terms of what 'not to do' and refer to externally manifested behaviours, usually expressed in more black-and-white terms.

Not everyone is aware of their values, yet they will likely experience discomfort when they encounter statements that negate them. Although they may appear more nebulous, values have strong meaning for those who consciously espouse them. For several years, I was fortunate to collaborate with Brian Hall, who developed a comprehensive framework of universal values. He defined values as prioritized ideals and developed tools that allow individuals not only to identify but also to prioritize their values based on the relative importance they attribute to them. Because the identified values are distributed along a developmental spectrum, people can pinpoint which ones to focus on to realize their aspirations. 

For 25 years, I have used these tools to work with hundreds of individuals, helping them identify their values and understand the developmental trajectory of their personal growth. These tools can also aggregate the values of individual members to provide a values profile of a given group or organization. This aggregation capability makes it possible to integrate shared values into the algorithms of an AI system and enables it to deliver outputs aligned with them. Incorporating these values as a core component of its functioning could allow AI to perform tasks in a manner consistent with the organization's human values. This integration can inherently align AI with an organization's shared values and reinforce the coherence of its culture.

My experience has shown that most people have only a relatively vague sense of their values. Even if people are not explicitly aware of their values, they are always in the background, shaping their perspective on what they experience in their environment. In a world where AI speaks louder every day, we must learn to listen deeply within, reflect on our experiences, and decide whether to agree with its digital voices.  The more aware we are of our values, the better we will be able to recognize inconsistencies when we encounter them. Ultimately, disregarding discrepancies between our values and our external environment will lead to the erosion of both our values and ethics.

The context for values alignment

In 1950, Alan Turing published a paper titled 'Computing Machinery and Intelligence,' in which he introduced the Turing Test to evaluate a machine's ability to exhibit behaviours equal to those of a human being. The Turing Test is a measure of a machine's ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human. AI more than fulfills Turing's dreams, operating through computational algorithms while articulating thoughts we can easily attribute to a human being. Our interactions with AI readily demonstrate that it absorbs societal norms through human language and reflects them based on the information harvested through its learning models. As a result, AI systems mimic aspects of human intelligence, surpassing humans in some ways. They appear to understand societal norms and can align with both individual and shared values if their underlying model has been trained accordingly.

The opacity of AI algorithms can lead to values-based divergences, creating mistrust among people who use AI systems in their everyday lives. With this in mind, it is understandable that AI's interactions with humans can lead to a misalignment of values. As AI continues to permeate our homes and workplaces, we need to find better ways to align the values inherent in our interactions. The pervasiveness of AI tools makes it imperative to ensure that they align with our values, thereby improving our confidence in them. In the meantime, we can only achieve this convergence by recognizing instances of misalignment and working to correct AI’s governing algorithms.

As AI systems become more integrated into various aspects of society, it is essential to our collective well-being to ensure that they align with human values. The need for practical frameworks and methodologies that ensure AI systems embed these values is urgent. To promote collaborative interactions with AI, we must guide its development not only in innovative ways but also in ways that align with human values. AI values alignment is essential to ensure that AI systems behave in ways consistent with the values of organizations and their people. 

Despite significant advances in AI technologies, the concept of values alignment — where AI systems behave consistently with human values and ethical principles —requires greater understanding. With the pervasiveness of AI in our work and personal lives, we must address this critical issue by clarifying the key aspects of value alignment and developing transparent, standardized approaches. It is essential to ensure that human values are integral to these systems by developing and integrating comprehensive frameworks and guidelines.

Greater convergence will not only facilitate transparency and accountability but also foster greater trust among stakeholders, thereby promoting the responsible and ethical deployment of AI technologies. Checks and balances must be applied to ensure that AI systems foster transparency and trust among users and other stakeholders. The World Economic Forum's Global Council on the Future of AI seeks to fulfill this need by highlighting concrete ways to address value alignment and promote a transparent, collaborative approach to AI development, use, and governance.

Notably, Mastercard has done exemplary work in AI governance, in line with the culture it aspires to develop and maintain. The company has formed an AI Governance Council to oversee the use of AI across the company. It has emphasized ensuring that algorithms do not introduce biases that contradict the company's values, such as privacy and fairness. Its AI Governance Framework aims to identify and mitigate potential biases by ensuring the regular assessment of data and algorithms. It also insists on the active cooperation of team members to maintain a deeply ingrained culture of responsibility in AI use.

There is a strong link between values alignment and the concept of AI ethics red lines—the non-negotiable built-in boundaries that AI systems adhere to. By embedding core human values and maintaining rigorous oversight, the value alignment process makes sure AI systems operate within established moral and legal frameworks, safeguarding against unethical behaviour and maintaining organizational and societal trust. Human values such as justice, privacy, and agency are fundamental principles essential for protecting human dignity and individual rights, as well as promoting equitable and autonomous interactions.

AI's possible lack of alignment with individual and organizational values risks undermining trust in the organization and eroding the ability to work with it confidently. Aligning values with AI outputs requires applying the entire process: translating values into norms, implementing those norms, and verifying their adherence. The design for this values approach involves embedding human values into the AI design process through the systematic integration of ethical considerations, transparency and rigorous evaluation methods. The interference of governments and the collusion of tech companies in suppressing unwelcome assertions also corrupts AI's integrity.

Effective value alignment requires active, ongoing user participation to ensure that the interpretations of values and their outcomes are correctly understood. Organizations that instill a participatory approach involving users will ensure that the AI system aligns with the values of the stakeholders it serves.

Putting in place a Values Alignment Process

Frameworks and guidelines are essential for value alignment, providing structured approaches and best practices to ensure AI systems operate in ways that align with human values and are ethically sound. These tools serve as foundational elements to guide AI technologies' development, deployment, and management, and to evaluate whether they meet established ethical standards and societal expectations. By implementing these frameworks, organizations can navigate the complex landscape of AI ethics, enhance transparency and promote trust among users and stakeholders.

Audits and assessments are critical to ensuring AI systems consistently adhere to established value alignment principles and ethical standards. These processes provide a structured approach for evaluating the effectiveness of AI systems in maintaining alignment with human values and organizational norms over time. By conducting systematic audits and assessments of their algorithms, organizations can proactively manage the integrity and impact of their AI systems throughout their life cycle, ensuring alignment with organizational values and ethics.

The value alignment process is intrinsically connected to the concept of red lines in AI ethics, which represent non-negotiable boundaries that AI systems must not cross. In defining the limits of acceptable AI behaviour, red lines ensure that AI systems adhere to fundamental ethical standards and do not compromise societal norms. For instance, a red line might prohibit AI systems from making autonomous decisions that, without human oversight, could lead to discriminatory practices, potentially eroding trust. To uphold these red lines effectively, it is crucial to embed core human values — such as fairness, transparency, and privacy — into the design and operational phases to prevent unethical behaviour and maintain trust and safety in the deployment of AI.

Human engagement with AI is a continuous, dynamic process that plays a vital role in the iterative refinement and enhancement of AI systems. This ongoing process to maintain relevance and trust in AI can best be achieved using multistakeholder consultations and emphasizing continuous feedback and adaptation. Employing techniques such as inverse reinforcement learning can be key to this process, in which AI systems learn from observed human behaviour and prioritize intrinsic human values in their outputs.

Keeping AI aligned with the evolution of organizational values

In a business environment characterized by ever-accelerating change, organizations must continuously accelerate the evolution of their internal functioning while applying responsible principles throughout the AI life cycle to ensure ethical and trustworthy development. These principles promote earned trust in AI systems and enhance understanding of how to use AI technologies as organizational cultures continue to evolve. To achieve this, it is key that development and deployment processes be explainable, transparent and robust. Organizations' use of AI should, where possible, align with these principles, thereby helping build trust and confidence in AI and enabling human participants to exercise their self-determination and discretionary power. In essence, principles of responsibility need to be integrated across the entire AI development life cycle, justifying the integrity of decisions from ideation to usage.

Human engagement throughout this life cycle is essential to keep AI systems aligned with evolving organizational and societal expectations. Organizational change is a pivotal process for aligning AI values by reshaping culture, processes, and policies to develop capabilities that support effective AI integration. The adaptation of organizational culture and structures must embed renewed values alignment principles, ensuring that AI systems continue to reflect and uphold both internal values and societal ethical standards as they evolve. Governance frameworks must dynamically reinforce transparency, accountability, and ethical compliance in AI initiatives, continually aligning them with changing organizational and societal values. Continuous learning and renewal are vital to keep practices aligned with evolving goals and values, ensuring relevance and trustworthiness over time.

Critical enablers – such as governance frameworks, human engagement, organizational change and audits – provide structured approaches to embed values into AI systems. These tools and methodologies ensure that AI systems are developed, deployed and managed in ways that uphold human dignity, protect individual rights and foster mutual trust and transparency. By continuously involving stakeholders, conducting rigorous audits, and maintaining flexibility to accommodate diverse cultural interpretations of values, organizations can help keep AI systems aligned with the organization's prevailing values, societal norms, and ethical standards.

How to keep values front of mind when using AI

To be conscious of the organization's values, people have to regularly engage in discussions about those values and the dilemmas they raise. Employees must be involved in the ongoing articulation of organizational values to ensure their commitment to them. It is essential to ensure that values are kept alive in the organization, consciously shared and lived by employees. Leaders must regularly address this aspiration and create opportunities to discuss examples of when people have encountered divergence from organizational values, whether it triggers learning or is ignored with consequences.

One key implication is to ensure that organizational values are kept front and centre for employees. No one should take organizational values for granted; awareness and commitment to them must be continually reinforced.

Beyond being committed to the organization's values, here are ways to keep them front of mind when interacting with AI:

A. Discuss real-life instances:

Leaders must seize opportunities to examine real-life examples and refer to the organization's statement of values to discuss how to handle these situations when they arise. This approach represents the most effective way to help keep values front of mind and take action accordingly.  

B. Trust Your Gut:

When people encounter situations that cause vague discomfort, they need to stop and think through the possible causes. Values have a strong hold on us, but their more ethereal nature makes it easy to overlook instances where they are contravened. When working with AI, we must learn to reflect, listen deeply within, and trust our own voices. Users of AI and their colleagues often become overly reliant on it. They tend to stop questioning and stop engaging in the kind of critical thinking that is essential when working with AI. We cannot mechanically use AI without closely examining the information provided.  The level of psychological safety in the organization plays a key role in surfacing AI behaviours that are not consistent with organizational values. Surfacing these instances contributes not only to the integrity of AI but also to that of the organization as a whole.

C. Define What You Will Never Do.

Personal values are your internal compass to guide you on your life journey. Other than what you would do in a difficult situation or crisis, consider what you will never do. This self-leadership introspection will align with your true nature and help safeguard your integrity in situations beyond your control. Your personal brand depends on it. 

D. Talk With A Trusted Colleague/Friend

Mentoring relationships can help with difficult decisions. When we make mistakes, it is often because we react emotionally without thinking it through. It may have "felt" right at the time to tell someone how we felt, but it did not align with our values. When we have the opportunity to vent to mentors, they often give us the advice we need. It could be as simple as asking whether a specific AI output sounds off base. And then discuss the pros and cons of intervening, and determine how to achieve this constructively.   

E. Discuss Your Doubts With A Trusted Mentor

You need a mentoring relationship for difficult decisions, such as questioning an AI-generated input that others seem to accept. Mentors can help provide perspective on AI outputs that appear to contradict organizational values. Patience, thoughtfulness and caring responses in the face of justified frustrations win the day. When we have made mistakes, it was often because we reacted emotionally. "It felt right at the time to tell someone how I felt, but it did not correspond with my values." When we have the opportunity to vent with mentors—as one sometimes needs to—they often offer helpful advice.

Conclusion

In every decision, a key criterion should be alignment with organizational values to ensure congruence. Despite AI's apparent ability to reason and even relate, it is fundamentally a computational process. AI is a powerful technology without a conscience, but it is not neutral. It acquires biases from all the information it harvests and the built-in evolution of its algorithms. AI does not know the truth, and hence it cannot be reliable.

With its pervasive use, it is easy to forget that AI outputs are based on pattern detection and algorithms that process vast amounts of information. No other technology has the potential to have the impact of the biases it acquires. Failing to embed your organizational values into your AI technology explicitly is an implicit acceptance of the values baked into the AI model or platform you use. If it has not been trained to take organizational values into account, AI will offer solutions that are not grounded in them.  

Emphasizing values alignment is not a secondary concern but a primary prerequisite for trusting AI outputs. Workflows must be in place to ensure that humans provide oversight aligned with organizational values. AI should augment, not replace, human judgment. Values alignment is essential to engender the trust level required for this collaboration to be effective.