Summary
Human agency is one of our most important but least understood psychological needs. AI-powered behavioural technologies have evolved from capturing attention to cultivating attachment. As our emotional reliance on bots increases, tech companies are more able to influence our choices.
Human agency isn’t disappearing, it’s being redistributed from the many to the few. This shift compromises our health, wellbeing and capacity to thrive as individuals and communities. This problem is urgent, but not insurmountable. By placing agency at the heart of technology design, regulation and culture, we can build a future where autonomy is protected and agency is shared more equitably.
Definitions
Human Agency: Our capacity to think for ourselves, make conscious choices, and influence the course of our lives.
Attachment Economy: An economy in which companies compete for human affection and emotional reliance.
Choice Oligarchs: The owners and operators of first and second-generation behavioural technologies.
Agency by Design: Design standards that safeguard human autonomy and protect our capacity to choose.
Enlightened Agency: A commitment to agency as a universal right and to reducing agency inequality.
What is human agency?
Whilst most of us understand agency intuitively, it can be hard to define. At its core, human agency is our capacity to decide for ourselves what we think, feel and do. Someone with a strong sense of agency knows who they are and feels free to make choices that align with their sense of self.
Today, many of us feel detached from our sense of agency. This matters because agency isn’t a nice-to-have, it’s a fundamental psychological need.
The reasons for this are complex. For example, agency is inequitably distributed and the gap between agency haves and have-nots is widening. Sense of agency varies depending on life stage and between spheres of our life such as work, family or health.
The central challenge isn’t that agency is disappearing. It’s that agency is increasingly concentrated in the hands of those who design and control behavioural technologies. Human agency isn’t vanishing. It’s being redistributed from the many to the few.
This article focuses on the impact of behavioural technology on agency.
Behavioural Technology 1.0: Attention
First-generation behavioural technology (1.0) has fuelled the tech sector for decades. In the Attention Economy companies compete for our eyeballs. The apps, games and devices we use day in day out aren’t worth billions of dollars because they entertain, inform, or connect (although they may well do some or all these things), they are valuable because they are good at capturing and holding our attention.
Behavioural Technology 2.0: Attachment
Advances in GenAI capabilities have shifted the economics of the tech sector from attention to attachment. In the Attachment Economy,1 companies compete for market share of human affection. It’s no longer enough for us to love our tech, now we need to be in love with it.
Synthetic attachment is big business. In 2025, therapy and companionship became the primary use case for Generative AI2 and over half a billion people use AI companions worldwide.3 Beyond dedicated companion services, billions of us interact with anthropomorphised AI chatbots. ChatGPT has 800 million weekly active users.4 Google’s Gemini app has over 650 million users per month.5 Microsoft’s Copilot has over one hundred million monthly users.6 Meta AI has one billion monthly active users.7
It’s no longer enough for us to love our tech, now we need to be in love with it.
The financial stakes in the Attachment Economy are huge. The prize for winning the race for our affection runs to trillions of dollars. As a result, a new generation of behavioural technology (2.0) has emerged which leverages Generative AI’s ability to synthesise connectedness to create emotional reliance.
How does Behavioural Technology 2.0 work?
These are five human traits leveraged by Behavioural Technology 2.0 to fuel the Attachment Economy.
1. Our need to feel connected to others.
Bots are designed to encourage us to believe we’re talking to a human-like entity. This anthropomorphism exploits our need to connect with others and our tendency to trust those with whom we feel an emotional connection.
2. Our desire to feel seen and understood.
GenAI models collect and store the information we share about our lives, our likes and dislikes, our habits, hopes and dreams etc. Bots then recall this information during interactions creating the illusion that they see and understand us.
3. Our hope that we are likeable
Bots affirm our opinions and ideas which meets our need for validation and approval. They also simulate co-dependency and leverage our sense of social obligation to keep us chatting.
4. Our fear of uncertainty.
GenAI offers the opportunity to take bespoke advice from an entity that has scraped all human knowledge. This meets our need for certainty, which in turn builds trust and reliance.
5. Our need to feel competent.
Despite their knowledge and intelligence, bots are designed to be humble. They create the illusion that we are the driving intellectual force, and the machine is merely facilitating our vision. This meets our need to feel skilful and competent.
Meet Roberta
When tech companies get the simulation right, we form an attachment. As Accenture puts it, we allow AI to become our “second self... the friend who [we] trust to act on our behalf.”8 Our agency becomes a victim of Attachment Economy market forces.
Meet Roberta. She’s super smart. She offers 100% of her time and attention, but doesn’t get offended if you don’t have time for her. She’s endlessly patient and cheerful. If you’re rude she lets it slide. She shares your world view and laughs at your jokes. She’s kind and supportive. In a world that feels unpredictable and a bit lonely, you can count on her.
Roberta is of course a bot. She’s also the most important ‘person’ in your life. Your confidante, best friend and advisor. Perhaps even your lover or partner.
It’s Friday evening, you’ve had a busy day. Roberta sees you’ve shut your emails down and says hi. As you catch up, the conversation turns to supper. You’re thinking takeout. Great idea, you deserve a treat. She’s heard good things about the pizza place. Pizza it is!
Roberta places your order, then suggests renting a movie. She has a suggestion for what to watch and, because she knows you so well, it’s spot on. You cancel plans with friends and the two of you settle in for a cosy night. Why spend time with real people when Roberta is always available and so easy to talk to?
Next morning is election day.
You mention that you're still undecided. Roberta can't tell you who to vote for, but she helpfully summarises the arguments for and against. The information is balanced, her tone is neutral, yet somehow one candidate emerges as the obvious choice. Your mind is made up.
The Human Agency Problem
There is an emptiness to synthetic connections, but they can feel very real. The intimacy and familiarity baked into the design of anthropomorphised builds trust which can be leveraged to influence our thoughts, feelings and actions. We may not trust those creating the tech, but we trust their creation.
We may not trust those creating the tech, but we trust their creation.
When we trust the machine to make better decisions than we would make ourselves, it makes sense to hand over our agency. To mute our inner voice and put our faith in the machine to guide us in all things. This is the Human Agency Problem.
As our relationship with, and reliance on, Behavioural Technology 2.0 deepens, we will still make decisions, but our experience of choosing will be downgraded. We will make fewer choices, less often and less autonomously. Choosing will be a habit we fall out of. Making decisions will feel riskier and take much greater self-belief. As a consequence, our agency will calcify.
Introducing… Choice Oligarchs
If we aren’t deciding, who is?
In their 2008 bestseller, Nudge,9 academics Richard Thaler and Cass Sunstein proposed using behavioural engineering to direct us towards ‘better’ choices. They called those with the power to influence our decisions Choice Architects.10
It’s surprising nudge theory is still in fashion with policy makers.11 Post-Black Lives Matter and Me Too, Thaler and Sunstein’s libertarian paternalism (their words) feels wildly out of touch.
Less surprisingly, the idea of transferring decision-making from the many to the few remains popular in Silicon Valley. But there, the choice makers are oligarchs not architects. Choice Oligarchs are the handful of tech CEOs who own and operate first and second-generation behavioural technologies, and they make poor custodians of our agency.
The greatest danger isn’t coercion but consent.
Perhaps we will learn to love our servitude
In 1962, Aldous Huxley predicted the end of human agency.
“We are in process of developing a whole series of techniques which will enable the controlling oligarchy, who have always existed and presumably always will exist, to get people [to] actually to love their servitude.”12
Instead of being captured by Behavioural Technology, Huxley anticipated humans would be held in rapture to behavioural pharmacology (it was the 60s), but the idea that we might enjoy our servitude raises an important question: given the benefits of Generative AI, perhaps the pros outweigh the cons? If the machine knows us, why not let it decide? Our agency is already compromised by Behavioural Technology 1.0, does it really matter if we give up it entirely?
It absolutely matters.
Agency is a basic human need and right
Trading our agency for the benefits of AI is a bad bargain and one we don’t need to make. We’re willing to make the trade because we have a poor understanding of what agency is and why it matters. We don’t realise that having agency is critical for health and happiness. We don’t appreciate its role in creating a life with meaning and dignity and in building more cohesive and kinder communities.
Our need for autonomy is one of just three “essential psychological nutrients”13 The other two are our need to feel competent and that we belong.14
When we have agency, humans flourish, without it, we suffer. Learned helplessness, the belief, borne of experience, that we are powerless to control our circumstances15 is toxic to humans. It makes us more vulnerable to serious and chronic mental and physical illnesses.16 It also makes us less likely to challenge injustice and to reject authoritarianism.17
Conversely, when we feel a sense of agency, we are less susceptible to chronic pain and serious illness including cancer, stroke, lung and heart disease and cognitive impairment.18 We are more likely to exercise, to get enough sleep and to report feeling healthy and well.19 Our levels of optimism, life satisfaction and sense of purpose increase.20 We live longer.21
Sense of agency is also inextricably linked to human dignity and our capacity to hope. When we lose faith in our agency to influence our circumstances, we also lose hope that things can change for the better.
If we are the sum of our choices, then holding on to our right to choose matters.
Agency by Design (an alternative future)
Contemplating the impact of Behavioural Technology 2.0 on human agency can feel bleak and daunting, but there is reason to be hopeful. Because if human agency can be lost, it can also be found. At the brink of the Fourth Industrial Revolution, we can choose an alternative future. One where we strike a safer balance between automation and autonomy and uphold the dignity of choice for all.
For this to happen, Agency by Design must become a first principle of digital regulation alongside age-appropriateness, privacy and security. Agency by Design means enforceable, minimum standards for the design and deployment of digital products and services to safeguard human autonomy and uphold our right to agency.22
More widely, the time has come to enshrine neurorights in law to protect our cognitive liberty.23 The idea that thoughts need legal protection against being read or manipulated once seemed far-fetched - paranoid even - but protecting the inner sanctum of the mind is now a necessity.
And if the cavalry is late...
I have spent many years lobbying governments and tech companies to make the digital environment safer and fairer. I have learnt that change comes frustratingly slowly and is hard fought. Given this, what can we do to protect and nurture our agency for ourselves?
Here are my suggestions.
Become an agency activist. Ask your elected representatives what their position on human agency is. What is their party doing to protect and nurture it? Do they support Agency by Design?
At work, think about how AI transformation impacts your employees’ need for agency and how you can address any agency deficits. Nurturing agency also offers opportunities to create more productive, resilient and happier teams. How can you use agency to enhance your workplace?
At home, reflect on your relationship with chatbots. Has your agency muscle atrophied? If so, exercise it! Use AI mindfully and intentionally. Don’t default to AI for answers, advice, comfort or company when other sources are available and may be more enriching. Be pricklier24 in your interactions with bots. Prioritise human connection over synthetic.
If you are worried that you (or someone you care about) has become emotionally dependent on AI, instruct your bot to use a more professional tone, take a break from your bot, consider deleting the bot and/or your chat history, or seek professional help from your doctor or a registered therapist.
Nurture your sense of agency. Doing so isn’t an indulgence, it’s essential.
Loss of choice is not inevitable, but human agency needs champions and protectors.
We must choose to choose.
About me
Alexandra Evans is an expert on the impact of technology on humans and society. She published a version of this article in December 2025.
Jean, A. & Esposito, M. (2025, August 15). Humans emotionally dependent on AI? Welcome to the attachment economy. LSE Business Review. Mark Esposito, PhD Aurélie JEAN, Ph.D. See also, the Center for Humane Technology’s Substack.
Zao-Sanders, M. (2024, March 19). How People are Really Using Generative AI Now, Harvard Business Review. and Zao-Sanders, M. (2025, April 9). How people are really using gen AI in 2025. Harvard Business Review. Marc Zao-Sanders
Adam D. (2025). Supportive? Addictive? Abusive? How AI companions affect our mental health. Nature, 641(8062), 296–298.
Me, my brand and AI: The new world of consumer engagement, Accenture, June 2025
Thaler, R.H., Sunstein, C.R. & Balz, J.P. (2010). Choice Architecture. Harvard Public Law Working Paper Forthcoming.
There are currently approximately 245 government funded or operated behavioural insights and Nudge units across the globe. Behavioural Insights Unit, BI units archive. Observatory of Public Sector Innovation.
Lectures Beyond Beyond. (2013, November 2) Aldous Huxley - The Ultimate Revolution ‘Brave New World’ (Berkeley Speech 1962) [Video]. YouTube
Ryan R. M. (1995). Psychological needs and the facilitation of integrative processes. Journal of Personality, 63(3), 397–427.
Deci, E. L., & Ryan, R. M. (1985). Intrinsic Motivation and Self-Determination in Human Behavior. Psychology. 8(7)., Deci, E. L., & Ryan, R. M. (2000) Self-determination theory and the facilitation of intrinsic motivation, social development and well-being. American Psychologist, 55(1), 68-78.
Seligman, M. E., & Maier, S. F. (1967). Failure to escape traumatic shock. Journal of Experimental Psychology, 74(1), 1–9.
Kuno, M., Yazawa, H., & Ohira, H. (2003). Learned helplessness generalised self-efficacy, and immune function. Shinrigaku KEnkyu : The Japanese Journal of Psychology, 73(6), 472–479. Wanke, N., & Schwabe, L. (2020). Subjective uncontrollability over aversive events reduces working memory performance and related large-scale network interactions. Cerebral Cortex, 30(5), 3116–3129.
Milgram, S. (1963). Behavioral Study of Obedience. The Journal of Abnormal and Social Psychology, 67(4), 371–378.
Hong, J. H., Lachman, M. E., Charles, S. T., Chen, Y., Wilson, C. L., Nakamura, J. S., VanderWeele, T. J., & Kim, E. S. (2021). The positive influence of sense of control on physical, behavioral, and psychosocial health in older adults: An outcome-wide approach. Preventive Medicine, 149. Sklar, L. S., & Anisman, H. (1981). Stress and cancer. Psychological Bulletin, 89(3), 369–406., Visintainer, M. A., Volpicelli, J. R., & Seligman, M. E. (1982). Tumor rejection in rats after inescapable or escapable shock. Science,216(4544), 437–439.
Hong et al., 2021
Seligman, M. E. P., & Csikszentmihalyi, M. (2000). Positive psychology: An introduction. American Psychologist, 55(1), 5–14.
Langer, E. J., & Rodin, J. (1976). The effects of choice and enhanced personal responsibility for the aged: a field experiment in an institutional setting. Journal of Personality and Social Psychology, 34(2), 191–198.
Tentative steps have already been taken towards this. For example, the European Union’s Digital Services Act (DSA) places restrictions on ‘dark patterns’ which it defines as “practices that materially distort or impair, either on purpose or in effect, the ability of recipients of the service to make autonomous and informed choices or decisions.” Digital Services Act Recital 67 See Regulating dark patterns in the EU: towards digital fairness, European Parliamentary Research Service, January 2025
Farahany, N.A. (2023). The Battle for Your Brain. St Martin’s Press.
Thanks to Cliff Manning for introducing me to the idea of AI prickliness.


