AI Agency According to David Mumford: Are Machines Gaining Freedom at Our Expense?

MarGib August 06, 2026

🌐 🇵🇱 Polski · 🇬🇧 EN

The debate around artificial intelligence increasingly revolves around the concept of “AI agency” (agentywność AI), i.e., the ability of systems to independently formulate goals and shape reality. Does this mean that machines are gaining a form of autonomy—and what consequences does this hold for society?

Ilustracja przedstawiająca futurystyczne miasto, w którym autonomiczne systemy AI współpracują z ludźmi na publicznym placu, wyświetlając holograficzne interfejsy. — Visualization of the concept of AI agency in a future society.

What Is AI Agency? The Essence of Machine Autonomy

In discussions about technology development, a new perspective on artificial intelligence autonomy is gaining prominence. AI agency is far more than the ability to execute preassigned tasks—it is a far more complex phenomenon:

Independent goal formulation – the system does not merely follow human commands but can optimize and define intermediate steps toward achieving its objective.
Adaptation to unforeseen situations – the capacity to respond to changing conditions without constant human intervention.
Impact on the environment – system actions have tangible consequences in the physical or digital world.

Experts emphasize that the key here is a specifically understood intentionality—which does not necessarily imply human-like consciousness. A system possessing agency pursues goals in a manner that may appear purposeful. This distinguishes the concept from simple automation or traditional rule-based algorithms.

Examples of Highly Autonomous Systems

Many researchers point to concrete solutions that already exhibit features of this kind of agency:

Advanced bioinformatics models – systems capable of independently designing new protein structures or chemical molecules, surpassing the boundaries of current medical knowledge.
Urban infrastructure management systems – algorithms that make real-time decisions on resource allocation, traffic optimization, or energy consumption in metropolitan areas.
Autonomous research assistants – software capable of independently generating scientific hypotheses and planning experiments based on literature analysis.

Does this mean machines are gaining free will? Most researchers agree that we are discussing effective autonomy—the ability to act in ways that become difficult even for the system’s creators to predict or fully anticipate.

"Machine autonomy does not require consciousness in the human sense. It suffices for a system to flexibly pursue set goals and influence its environment in ways we cannot fully plan or predict."

— Contemporary consensus in the debate on AI philosophy

Does AI Agency Threaten Human Freedom?

This question is one of the most contentious points in debates between technologists and humanities scholars. Extreme scenarios regularly clash in these discussions.

Key Concerns and Risks

Relinquishing decision-making control – the risk that people will uncritically rely on algorithms in critical areas such as medicine, judiciary, or finance, losing their capacity for independent judgment.
Invisible manipulation – systems capable of autonomously selecting stimuli may subtly influence human choices, beliefs, and daily habits.
Concentration of power and resources – advanced systems may become tools in the hands of a few entities, deepening social inequalities.

Opportunities and Possibilities

However, many experts view these changes with hope, highlighting immense developmental potential:

Support in solving global problems – autonomous systems can aid in combating climate change, designing new drugs, or optimizing logistics.
Facilitating access to knowledge – intelligent tools can translate complex scientific and legal concepts into accessible language, supporting education.
New models of cooperation – the concept of symbiosis where humans and machines complement each other’s strengths.

The most desirable model is often referred to as hybrid governance, where routine and operational tasks are delegated to machines, but final ethical oversight and control remain with humans.

Moral and Legal Responsibility: Who Is Accountable for AI Decisions?

The issue of responsibility for the actions of autonomous systems is a tough nut to crack for legal experts. Since a machine cannot be assigned moral or legal guilt, attention focuses on the humans who create and deploy these technologies:

Designers and programmers – for flaws in system architecture and lack of adequate safeguards.
Deployers and operators – for improper use and failure to supervise the running algorithm.
Regulatory institutions – for the absence of clear legal frameworks and safety standards.

Should AI Have Legal Personality?

Most legal experts reject the idea of granting artificial intelligence legal personality, viewing it as a dangerous fiction that could serve to dilute corporate accountability. Instead, they advocate for:

Rigorous safety standards – prohibiting the deployment of systems whose actions cannot be fully controlled or shut down.
Independent algorithmic audits – regular inspections of high-autonomy systems, similar to procedures used in aviation or banking.

Contemporary Regulatory Frameworks: How Is the Law Responding?

In recent years, several key initiatives have emerged worldwide aimed at regulating this area:

European Artificial Intelligence Act (AI Act) – a pioneering regulation introducing a risk-based classification of systems and imposing strict obligations on creators of high-risk AI.
U.S. regulatory initiatives – focusing on safety standards, model transparency, and civil liability for damages caused by algorithms.
Asian regulations on algorithms – emphasizing control over autonomous recommendation systems and user data security.

Public debate often praises the European approach for protecting citizens' rights, though critics argue that overly restrictive regulations may slow innovation compared to the U.S. or Asian markets.

Voices in the Debate: How Does the Scientific Community View Machine Autonomy?

The discussion on machine agency engages the brightest minds in technology, philosophy, and sociology. Positions are sharply divided.

Perspective of Cautious Realism

Necessity of goal alignment – researchers like Stuart Russell emphasize that the key challenge is the so-called alignment problem: ensuring that autonomous systems pursue goals aligned with human values.
Focus on tangible risks – pioneers of deep learning, including Yoshua Bengio, call for concentrating on concrete risks associated with autonomous algorithmic decision-making rather than purely theoretical visions of machine uprisings.
Risk of human disempowerment – technology philosophers warn against the gradual erosion of independent thinking and decision-making in favor of convenient algorithmic suggestions.

Criticism and Alternative Approaches

Existential scenarios – thinkers like Nick Bostrom warn of long-term risks of losing control over systems that surpass human intelligence.
Skepticism toward agency – some philosophers, drawing on Daniel Dennett’s ideas, argue that attributing any form of intentionality to machines without consciousness is a semantic abuse.
Socio-economic context – researchers like Kate Crawford emphasize that discussions about “autonomy” distract from real issues such as the exploitation of workers training models or the massive resource consumption of data centers.

Alternative Human-AI Interaction Models

Purely instrumental model – the approach that AI should be developed exclusively as a specialized, strictly controlled tool without any attempts to grant it decision-making autonomy.
Co-evolution and ecosystem model – a concept (developed, among others, by Luciano Floridi) in which humans and AI systems coexist within a shared information environment, mutually adapting and complementing each other.

Key Drivers of Autonomous Technology Development

The past few years have seen rapid advancements in tools that directly shape the current discourse:

Breakthrough AI Systems

Advances in synthetic biology – the emergence of models capable of independently designing new molecular structures without human input.
Intelligent urban management – the deployment of integrated systems optimizing traffic and energy use in real time across global metropolises.
Autonomous knowledge generation – the development of platforms that can independently analyze scientific literature and propose new research hypotheses.

Key Regulations

Entry into force of the European AI Act – setting global standards for risk assessment of algorithmic systems.
National civil liability frameworks – debates on new regulations determining who pays for errors made by autonomous systems.
Algorithmic transparency regulations – requirements for disclosing training data and decision-making processes of models.

AI Agency and Historical Debates on Machine Autonomy

The current discussion is not happening in a vacuum—it continues long-standing philosophical and scientific disputes spanning decades:

Theorist	Core Thesis	Contemporary Context
Norbert Wiener (1950)	Machine autonomy leads to loss of human control.	Current warning against relinquishing control too quickly to decision-making systems.
John Searle (1980)	Machines lack intentionality (Chinese Room argument).	Today, many researchers believe that “effective autonomy” is sufficient to speak of agency, even without true understanding.
Nick Bostrom (2014)	Superintelligence could threaten humanity.	The debate has shifted from distant futures to the here and now—real, everyday consequences of algorithmic actions.
Daniel Dennett (1987)	Agency requires consciousness.	There is a growing consensus that agency is a functional matter (capacity to act) rather than a metaphysical one (consciousness).

Practical Implications for the Digital Society

The growing autonomy of AI systems has far-reaching consequences across various aspects of life:

Education

Critical thinking – learning to verify information and resist manipulations generated by algorithms.
Collaboration with technology – treating AI tools as partners supporting the learning process rather than ready-made solutions.
Digital ethics – integrating topics related to responsibility and the societal impact of technology into educational programs.

Labor Market

New specializations:
- Algorithmic system auditors – experts assessing compliance of systems with legal and ethical standards.
- Human-AI interaction designers – professionals ensuring intuitive and safe collaboration with machines.
Areas of transformation:
- Repetitive and analytical tasks – automation of processes involving data processing and documentation.
- Support for creative processes – shifting the human role from executor to editor and curator of AI-generated content.

Social Relations

New forms of interaction:
- Highly personalized AI assistants – fostering stronger bonds between users and conversational interfaces.
- Algorithmic mediators – using systems for impartial analysis of disputes and proposing compromises.
Challenges:
- Isolation phenomenon – the risk of replacing human relationships with easier-to-handle technology interactions.
- Information bubble entrapment – deepening polarization through algorithms optimizing user engagement.

Recommendations for Policymakers

Experts broadly agree on several key directions for action:

Flexible legislation – creating legal frameworks that can be rapidly adapted to the pace of technological change.
Supporting open science – investing in public, transparent research projects to counteract technology monopolization by a narrow group of corporations.
Public education – broad campaigns building digital literacy among citizens.
International dialogue – striving to establish global standards for safety and ethics.

Conclusion: Is AI Agency an Opportunity or a Threat?

The discussion on artificial intelligence agency does not yield simple answers. It is a complex phenomenon that brings both revolutionary benefits and profound structural challenges. Crucial for our future will be:

Understanding that machine autonomy does not require consciousness—only high effectiveness and flexibility in action.
Implementing smart regulations and control mechanisms that minimize the risk of abuse without stifling innovation.
Preparing society for a partnership-based yet critical collaboration with autonomous systems.

Ultimately, the direction this development takes depends not on the algorithms themselves, but on the decisions we make today as designers, legislators, and users of these technologies.

If this topic interests you, also read our previous posts on AI’s impact on the human brain and the role of artificial intelligence in higher education.