The Revival of a Forgotten Science: Cybernetics for Responsible AI

Introduction

The rapid advancement of AI systems with increasing autonomy, influence and agency has accelerated the spread of discussions around developing AI responsibly and ethically. As we create more powerful technologies, we need conceptual models to understand how AI can be directed and controlled for the benefit of humanity.

My theory is that the field of Cybernetics, focused on regulation and communication in complex systems, can provide valuable perspectives to guide the integration of AI into society, providing tools and models to find effective solutions to the AI Alignment problem.

I believe that it is time to revive Cybernetics as a global discipline along with its spirit of creative cross-collaboration for tackling the complex issues surrounding the development and implementation of ethical AI.

What is Cybernetics?

The field of Cybernetics emerged from a series of influential multidisciplinary conferences in the 1940s and 1950s attended by pioneers like Norbert Wiener, John von Neumann, Claude Shannon, and Warren McCulloch and other from fields including mathematics, biology, engineering and neuroscience, who came together to restore unity to science.

The definition of Cybernetics was officially coined in 1948 by Wiener, who created structure and coherence around shared conceptions emerging from the conferences, and defined it as the study of control and communication in complex systems, whether mechanical, biological, cognitive or social.

After much consideration, we have come to the conclusion that all the existing terminology has too heavy a bias to one side or another to serve the future development of the field as well as it should; and as happens so often to scientists, we have been forced to coin at least one artificial neo-Greek expression to fill the gap. We have decided to call the entire field of control and communication theory, whether in the machine or in the animal, by the name Cybernetics, which we form from the Greek κυβερνήτης or steersman. 

- Norbert Wiener

It examined how parts regulate other parts and the system as a whole through circular flows of information and feedback loops.

In other words, Cybernetics provided tools and insights into how systems, regardless of their ontological nature and structural definition, achieve goals, correct errors, and pursue stability.

Key concepts in Cybernetics include:

• Feedback loops: Circular flows of information where a system’s outputs are routed back as inputs, enabling self-regulation. Both positive and negative feedback play important roles.
• Circular causality: Mutually reinforcing causal relationships between elements, rather than linear causation. Leads to non-linearity.
• Self-organization: Spontaneous emergence of order from local interactions rather than external control.
• Homeostasis: How systems maintain equilibrium and stability through self-regulating processes.
• Autopoiesis: Theory of self-producing and self-maintaining systems bound by dynamical structures.
• Variety: The complexity or information-carrying capacity of different system states.
• Information: Quantifying information to understand system complexity, communication, and control.
• Control: Mechanisms systems use to achieve goals and maintain stability through feedback and circularity.
• Hierarchy vs Heterarchy: Contrasting structures of vertical control vs distributed control.
• Structure vs Organization: Distinguishing the architecture (static, temporary form) from functional organization (dynamic form) of a system.
• Observer’s Influence: Accounting for the subjectivity and influence of external agents in shaping systems they are observing.

Cybernetics provided a paradigm for studying regulation, control, autonomy and communication across natural, artificial and hybrid systems. Its interdisciplinary nature and emphasis on unifying principles differentiated it from more specialized fields, having had strong influence on early neural network research and machine learning and providing models for self-correcting computations.

Seminal texts like Wiener’s “Cybernetics” and von Neumann’s “Computer and the Brain” connected Cybernetics to computer science. Connections to information theory, systems theory, and operations research also contributed to the foundations of the field.

By the 1970s onwards, Cybernetics was increasingly side-lined as derivative fields like computer science, cognitive science and systems theory took more specialized paths forward. While the core cybernetic principles remained in use in a number of disciplines, cybernetic pioneers became less referenced and ultimately, the discipline itself progressively disappeared from scientific discourse.

Cybernetics for Responsible and Ethical AI

As AI systems grow increasingly powerful and autonomous, with rapid advances in foundation models like OpenAI GPT-4, Google PaLM, Anthropic Claude, Inflection Pi, open-source alternatives, and autonomous agents such as Auto-GPT and BabyAGI, cybernetic concepts can provide crucial perspectives for developing these technologies responsibly and ethically.

There are a few key reasons why reviving cybernetic perspectives could be valuable for developing effective AI safety paradigms and ethical systems:

• Cybernetics offers crucial insights into regulating complex, adaptive systems. AI safety fundamentally involves controlling behaviours in advanced, self-improving algorithms. Cybernetic theories and concepts can inform the mechanisms needed.
• Concepts like circular causality and ‘second-order Cybernetics’ highlight the role of the observer within systems. This reflexivity is important for ethical AI; recognising that our own biases shape how we program morality.
• It emphasises how parts relationally regulate the wider system. For AI, this underscores holistic, dynamic alignment approaches rather than just applying censorship to the generative capabilities or constraining specific behaviours.
• The cybernetic focus on flows of information, communication and control maps well to coordinating distributed AI systems through networks and protocol-based governance.
• Cybernetic models balance stability and adaptation. Ethical AI requires dynamic principles that can evolve responsibly as technology advances.
• The field draws from diverse disciplines. In the same way ethics involves integrating perspectives from law, philosophy, social science etc, so having interdisciplinary mindset helps.
• The early pioneers of Cybernetics were highly interdisciplinary, collaborative and creative:  qualities useful for tackling complex AI safety challenges.
• Cybernetics’ history as an unconventional field that challenged boundaries which mirrors the need for innovative, bold thinking to guide AI wisely beyond existing norms.

Cybernetics’ interdisciplinary nature, reflexive analysis, and focus on unifying principles provides an effective framework for studying and developing AI ethics. Integrating diverse ethical philosophies and social impacts requires a systems-level perspective.

The early pioneers of Cybernetics also displayed creativity, boldness, and a willingness to challenge conventions; traits useful for innovating solutions as AI enters uncharted territory beyond existing technical and ethical norms. Reviving this spirit could help break the status quo thinking around technology innovation dictated by Silicon Valley megacorps.

Rather than consolidated corporate control, cybernetic thinking opens up possibilities for more decentralised, participatory paradigms for shaping the evolution of AI technology responsibly. The field’s boundary-challenging ethos mirrors the need for fresh perspectives that question prevailing norms as we build beneficial AI.

I believe that integrating cybernetic thinking in research communities, both closed and open source, seems highly relevant to developing paradigm-shifting solutions for AI alignment and for AI Ethics implementation in general.

Closing Thoughts

As AI grows more capable and embedded in our lives, ensuring its safe and ethical development is an urgent priority. Concepts from the vital but overlooked field of Cybernetics offer timely insights into regulating complex adaptive technologies.

By reviving cybernetic thinking, we can build upon the pioneering work of thinkers like Wiener, McCulloch, Ashby and others who tackled foundational problems of control, stability, and communication in systems. Their interdisciplinary creativity is needed today.

For my own part, I’ve started developing the following roadmap to 1. stimulate Cybernetics revival and 2. steer its applications in support of AI Ethics implementation:

• Applying Cybernetics and socio-Cybernetics principle to solve real-world pressing problems.
• Promoting cybernetic perspectives, AI literacy and ethics education to encourage informed global discourse among diverse groups of people, and to aid critical thinking about emerging technologies.
• Fostering an interdisciplinary community of thinkers to catalyse technical breakthroughs in responsible and ethical AI development.
• Updating cybernetic theories with modern understandings from complex systems science, cognitive science, distributed computing and more.
• Developing frameworks for AI ethics implementation based on reinforcement learning, feedback control, and distributed consensus.
• Designing algorithmic moderation features and adversarial behaviour control processes to mitigate potential misuse and risks of AI systems.

If you are compelled by this vision, reach out to collaborate, share perspectives, or steer related projects.

The path forward requires an integrated effort from diverse voices and disciplines, including technology, ethics, governance, philosophy, society etc.

By resurrecting this forgotten discipline and updating its frameworks to meet today’s challenges, we can shape AI technology for the benefit of humanity.