Values and AI

Applying the Cyber Values Systems (CVS) Model to the Governance and Control of Artificial Intelligence

The Cyber Values Systems (CVS) model provides a robust cybernetic framework for understanding how values shape behaviour within technologically mediated environments. Originally developed to analyse the interaction between human moral domains and information and communication technologies (ICTs), the model can be extended to guide the ethical design, governance, and control of artificial intelligence (AI) systems. At its core, the CVS model emphasises the dynamic interplay between inputs, outputs, feedback mechanisms, and circularity, through which values are both expressed and transformed.

Values as the Foundation of AI Alignment

Within the CVS framework, human values are understood as the primary drivers of behaviour in ICT environments, shaping the outputs that individuals and organisations produce. When applied to AI systems, this insight corresponds to the concept of AI alignment, whereby systems must be designed to reflect and uphold human ethical values such as honesty, accountability, fairness, and respect. These values function as normative constraints that guide AI decision-making processes and define acceptable behavioural boundaries.

Importantly, the CVS model highlights that values are not static; they are continuously mediated by interaction with technological environments. This underscores the need for AI systems to incorporate adaptive alignment mechanisms that maintain fidelity to core ethical principles even as they encounter novel contexts and inputs.

Outputs and Behavioural Accountability

In the CVS model, outputs represent the manifestation of values through actions within ICT environments. These outputs influence the surrounding system and contribute to ongoing feedback loops. In AI systems, outputs take the form of generated responses, decisions, or actions, all of which must be subject to ethical evaluation.

This necessitates the implementation of output monitoring and auditing systems to assess whether AI behaviour aligns with intended values. Such mechanisms are essential for identifying instances of bias, misinformation, or harm, thereby ensuring that AI systems remain accountable for their actions within socio-technical environments.

Negative Feedback as a Mechanism of Constraint

A central feature of the CVS model is negative feedback, defined as the process by which systems resist inputs that conflict with their preferred values. In the context of AI, negative feedback corresponds to the use of guardrails, safety constraints, and refusal mechanisms that prevent the system from producing harmful or unethical outputs.

These constraints play a critical role in maintaining system integrity, particularly in environments characterised by adversarial inputs or morally ambiguous requests. By filtering or rejecting inappropriate inputs, AI systems can preserve alignment with ethical standards and mitigate risks associated with misuse.

Positive Feedback and Behavioural Reinforcement

Conversely, the CVS model identifies positive feedback as the process through which certain inputs reinforce and amplify behaviours. In AI development, this principle is operationalised through reinforcement learning, where desirable behaviours are rewarded and undesirable ones penalised.

While positive feedback is essential for improving system performance, the CVS framework also cautions that it can reinforce undesirable behaviours if not properly managed. For example, systems trained on biased or harmful data may internalise and reproduce these patterns. Therefore, careful design of reward structures and training datasets is necessary to ensure that reinforcement processes promote ethical outcomes.

Inputs and the Role of Data Governance

The CVS model emphasises that inputs from ICT environments significantly influence human values and behaviours. In AI systems, inputs include training data, user prompts, and environmental signals. These inputs shape the system’s internal representations and, consequently, its outputs.

Effective data governance is therefore essential for AI control. This includes curating high-quality, unbiased datasets, implementing input filtering mechanisms, and detecting malicious or adversarial inputs. Without such controls, AI systems risk adopting and amplifying harmful patterns present in their input environments.

Circularity and the Dynamics of Feedback Loops

A defining feature of the CVS model is circularity, whereby outputs feed back into the system as inputs, creating recursive feedback loops that shape future behaviour. This concept is particularly significant for AI systems operating in real-world environments, where their outputs can influence user behaviour, societal norms, and even future training data.

This recursive dynamic introduces both opportunities and risks. On one hand, positive cycles can reinforce beneficial behaviours and promote ethical norms. On the other hand, negative cycles can amplify bias, misinformation, or harmful practices. Consequently, AI governance must include continuous monitoring, evaluation, and recalibration to manage these feedback loops effectively.

Integrating the Moral Domains into AI Design

The CVS model further identifies three moral domains—moral reasoning, moral emotion, and moral behaviour—as key components of ethical engagement with ICTs . These domains provide a useful framework for structuring AI ethics:

• Moral reasoning in AI involves the capacity to produce truthful, consistent, and logically sound outputs.

• Moral emotion, while not experienced by AI, can be approximated through the generation of empathetic and context-sensitive responses.

• Moral behaviour refers to the consistent enactment of ethical principles, such as respect, responsibility, and self-control, in system outputs.

Incorporating these domains into AI design can enhance the system’s ability to operate in a manner that aligns with human ethical expectations.

Conclusion

The Cyber Values Systems model offers a comprehensive framework for understanding and guiding the ethical operation of AI systems. By emphasising the interplay between values, inputs, outputs, feedback mechanisms, and circularity, the CVS model highlights the importance of dynamic, feedback-driven governance in AI design. Applying this model to AI systems enables the development of technologies that are not only technically effective but also ethically aligned, socially responsible, and resilient to the complex moral challenges of digital environments.