Can AI Maturity Protect Human Agency?

Artificial intelligence is becoming increasingly capable of performing tasks that once required distinctively human skills. It can generate content, support decisions, interpret information, and interact with users in ways that resemble human communication.

But as these capabilities expand, another question deserves attention: are people becoming more capable through their interaction with AI, or increasingly dependent on systems they do not fully understand?

This question was one of the motivations behind our chapter, Neural Network Pressure: AI Maturity Versus Dehumanization, co-authored with Professor Liudmila A. Vasilenko and published by Springer Nature in 2026.

The chapter examines the social, psychological, and ethical implications of widespread AI adoption, with particular attention to human agency, information manipulation, deepfakes, and the concept of AI maturity.

Why We Looked Beyond Technological Performance

Discussions about artificial intelligence often begin with technological achievements: improved performance, expanding capabilities, and new opportunities for automation.

However, technological progress and human development are not necessarily the same process.

A system may become better at performing cognitive tasks without its users developing a corresponding understanding of how the system works, when its outputs are reliable, or what consequences may follow from accepting its recommendations.

This distinction becomes particularly important when AI moves beyond specialized professional settings and enters everyday communication, education, work, and personal decision-making.

Our interest was therefore not limited to what AI systems could accomplish.

We wanted to examine what increasingly capable AI systems might mean for the people and societies adopting them.

From Technological Promises to Social Consequences

In developing the chapter, we examined research and statistical reports on AI diffusion, technological development, and patterns of human–AI interaction.

The analysis drew on sources available in 2024–2025, including international reports on AI development, patent activity, and research into how people use conversational AI systems.

One important distinction emerged from the evidence on AI use: people do not always interact with AI in the same way.

Sometimes AI supports human work through collaboration, iteration, and learning. In other situations, users delegate tasks to automated systems with relatively little direct involvement.

These different modes of interaction raise questions about the changing distribution of cognitive activity between humans and machines.

The implications cannot be assessed through productivity measures alone. They also require attention to human understanding, independent judgment, and the ability to evaluate technological outputs.

A Small Exploratory Experiment

Alongside the analysis of published sources, we conducted an exploratory exercise in April 2025 using publicly accessible AI systems.

We asked questions involving values, cultural references, moral judgments, and the ways conversational systems represented humans and themselves.

The purpose was not to benchmark model performance or establish generalizable differences between AI systems.

Instead, the exercise helped us examine a less visible dimension of human–AI interaction: how seemingly helpful conversational responses may also reflect particular assumptions, cultural references, and normative perspectives.

One example concerned recommendations of cultural works illustrating justice and morality. The responses prompted questions about the diversity of cultural references offered by conversational AI and the extent to which users recognize that recommendations are shaped by the systems producing them.

The exercise illustrated a methodological challenge that remains relevant: evaluating AI interaction requires attention not only to the factual correctness of responses, but also to the values, assumptions, and interpretive frameworks embedded in them.

These observations were exploratory and should not be treated as representative evidence about AI systems as a whole.

Why Deepfakes Matter to This Discussion

Deepfakes represent another dimension of neural network pressure.

Generative technologies make it increasingly possible to create convincing synthetic images, video, and audio. These capabilities have legitimate creative and commercial applications, but they also introduce risks involving deception, identity misuse, reputational harm, and information manipulation.

The problem extends beyond the technical capacity to detect manipulated content.

It concerns the conditions under which people decide what to trust.

When synthetic content becomes easier to produce and circulate, individuals may face a greater burden of verification. Institutions, meanwhile, must establish procedures for authentication, accountability, and responding to harmful uses.

This raises a broader question: can technological safeguards alone maintain trust, or must they be accompanied by new forms of individual and institutional competence?

AI Maturity as an Alternative to Passive Adaptation

Our chapter proposes AI maturity as an important response to these challenges.

We approach AI maturity not simply as the ability to operate AI tools, but as readiness to engage with them responsibly, with attention to ethical principles, values, and the potential consequences of their use.

At the individual level, this implies developing the ability to critically evaluate outputs, recognize limitations, and preserve independent judgment.

At the organizational level, it involves appropriate governance, responsible data practices, interdisciplinary knowledge exchange, and the capacity to respond to emerging risks.

At the societal level, AI maturity also involves education, public awareness, and institutional arrangements capable of supporting responsible technological adoption.

These dimensions are connected.

Technical proficiency without critical understanding may encourage dependency. Regulation without adequate knowledge may struggle to address changing technological practices. And innovation without attention to human consequences may create new forms of vulnerability.

AI maturity therefore requires more than technological adoption.

It requires the development of capacities that allow people and institutions to make informed choices about how artificial intelligence is used.

What This Research Led Me to Ask Next

One of the most important questions arising from this work concerns the relationship between technological development and institutional adaptation.

Even when the risks associated with AI are recognized, the mechanisms needed to address them may develop at a different pace from the technologies themselves.

This observation has become particularly relevant to my subsequent doctoral research on the transformation of educational and labour institutions.

That research extends beyond the scope of our published chapter. It examines how digitally mediated practices reshape institutional roles, decision-making authority, accountability, and the possibilities available to individuals affected by technological systems.

The connection between these research directions lies in a shared concern: how can technological progress expand human capabilities without diminishing meaningful human agency?

Continuing the Conversation

The central question is not whether artificial intelligence is beneficial or harmful in itself.

Its consequences depend on how technologies are developed, deployed, understood, and governed.

As AI capabilities advance, we need to examine the corresponding development of human competence, organizational responsibility, and institutional safeguards.

I would be interested to hear how researchers in different fields approach the concept of AI maturity.

Should it be understood primarily as an individual competence, an organizational capability, or a property of the institutional environment in which AI operates?

Original research

Vasilenko, L. A., & Litash-Sorokina, E. A. (2026). Neural Network Pressure: AI Maturity Versus Dehumanization. In Digital Transformation: What Kind of Smart Life do we have today?, Vol. 1 (pp. 37–48). Springer, Cham.

https://doi.org/10.1007/978-3-032-20582-7_3

Author: Elena A. Litash-Sorokina
Doctoral Researcher, RANEPA
Founder, PST Lab
https://lita.sh