Earthworming: Responsible AI

Ursula Franklin's 1989 Massey Lecture "The Real World of Technology" offers a sharp framework for thinking critically about "artificial intelligence" today. Delivered at the dawn of the internet and personal computing, the technocratic conditions she described have only intensified. Her analysis of how prescriptive technologies erode human agency and restructure social relations around compliance has anticipated algorithmic surveillance, predictive policing, and mechanized warfare. Today, with Big Tech's rapid deployment of LLMs and agentic AI, her thought proves only more prescient.

What would Franklin say about the largely unregulated roll-out of "intelligence tools" into our homes and workplaces? What might a more responsible approach look like?

Prescriptive Technologies

Franklin defines technology not as simply a collection of tools, but as a comprehensive social practice that dictates human relationships and structures power. She distinguishes between two fundamental categories: holistic and prescriptive. Holistic technologies are the tools of the craftsperson —the stonemason, the weaver, the independent developer who custom codes software— where a single maker oversees production from inception to completion, exercising judgment at every stage. Prescriptive technologies, on the other hand, fragment production into discrete, managed steps —the assembly line, the fulfillment centre— displacing autonomy into layers of management and monitoring. The maker no longer makes; they comply.

The fragmentation of work into prescribed steps institutionalizes isolation, dependency and standardization. While holistic practice thrives on communication, collaboration and shared judgment, prescriptive workflows reduce the relationships of a community of practice to mere process. Outputs converge toward what the system permits rather than what the makers envision. Variation is seen as waste, deviation as error, regardless of any unpredicted merit. As collaboration and judgment are excised from the workflow, skill and creativity atrophy. Automation bias takes hold, and the machine's answer becomes the default. Over time, prescriptive technologies do not merely change how work is done; they reshape social relations in their own image, organized around production, efficiency, and top-down management rather than the lived experience of the communities they serve.

Franklin's distinction serves as a diagnostic lens rather than a rigid taxonomy. Many technologies combine elements of both holism and prescription. To be sure, compliance, standardization, and efficiency are not inherently malign. Prescriptive tech underpins much of what keeps us safe, healthy, and productive. The problem is not prescription itself, but its dominance as the organizing logic of social life.

While we should not forget that these prescriptive technologies are often exceedingly effective and efficient, they come with an enormous social mortgage. The mortgage means that we live in a culture of compliance, that we are ever more conditioned to accept orthodoxy as normal, and to accept that there is only one way of doing 'it'. — Ursula Franklin ^[1]

Compliance Culture

Prescriptive technology does not merely reorganize work; it reorganizes people. Living and working within systems designed primarily for operations and not the communities within them becomes normalized. Compliance becomes the dominant culture.

Prescriptive technology operates on feedback, not reciprocity. A system operating on feedback registers a response and makes minor adjustments within its own predetermined logic. The design itself is never questioned. Reciprocity is communication that fundamentally alters the system, redistributing power among its participants. A platform that updates its terms of service with an "accept or leave" ultimatum is feedback. A platform restructured by genuine user governance is reciprocity.

Political decisions about access and power are increasingly buried in technical designs, deployed behind the scenes by private interests rather than debated publicly by the community or elected representatives. The adoption of technology is framed as inevitable "progress" and inherently "good", an ideology that renders the process unquestionable. While the system appears objective, its politics are —often by design— made invisible.

In cultures of compliance, "people are seen as sources of problems while technology is seen as the source of solutions" ^[1]. This "anti-people mindset" ^[1] favours a model of production over growth, where environments are engineered for technological optimization rather than the people within them. When the logic of production becomes totalizing, it flattens all other social relations —those based on compassion, obligation, ecological responsibility— until only the transactional remains. If the medium is the message, the message of prescriptive technology is compliance.

If technology is not merely material production but social infrastructure, then decisions about its design and deployment are too consequential to be left to markets alone. These decisions belong to public discourse, subject to the kind of debate that facilitates genuine communication and reciprocity rather than the mere appearance of consultation.

The answer is not more regulation for its own sake. Poorly designed policy can stifle the real gains technology offers just as effectively as unchecked markets erode them. What Franklin calls for is better governance, not a blanket brake on innovation. LLMs themselves may prove useful here: accelerating research and providing the impact modelling that better policy requires.

Despite short-term gains, ultimately compliance culture is a liability. Prescriptive systems trade resilience for efficiency, turning people into passive components, diminishing the judgment and creativity needed when things go wrong. Responsible production isn't just an ethical choice; it's a shift from unsustainable extraction to sustainable cooperation.

"As methods of materials production, prescriptive technologies have brought into the real world of technology a wealth of important products that have raised living standards and increased well-being. At the same time they have created a culture of compliance. The acculturation to compliance and conformity has, in turn, accelerated the use of prescriptive technologies in administration, government, and social services. The same development has diminished resistance to the programming of people." — Ursula Franklin ^[1]

Are LLMs Prescriptive Technology?

Franklin was not "anti-tech". She acknowledged the genuine gains of early computing, from word processing to instant communication. Yet she also observed how these tools systematically eroded essential human skills — spelling, arithmetic, handwriting.

Every new technology is introduced as liberating. But in the technology hype cycle, there is a steep drop from the peak of "youthful exuberance" into disillusionment. Social media is a cautionary example: marketed as connection and community, it has mostly delivered distraction, misinformation, algorithmic addiction, and surveillance. What begins as voluntary adoption becomes dependency on frictionless conveniences. The industry then rebrands intentionally designed "lock-in" as "platform stickiness". By the time a technology is normalized, opting out becomes too professionally and socially costly to be a real choice.

The fever-pitch adoption of LLMs has been no exception. Hawked as tools that extend rather than displace human authorship, they have arrived touting the language of holistic technology: "supercharge your productivity", "your AI assistant", "copilot". Used with intention, LLMs can be genuinely assistive. The researcher who retains authorship, the experienced programmer who uses code completion as a guide rather than blind instruction, the person with a disability navigating an inaccessible world not designed for them — these are instances where the technology extends human capacity rather than replacing it.

But as LLMs are aggressively deployed without meaningful reciprocity, their prescriptive design becomes impossible to ignore. Students are automating homework rather than cultivating the capacity for critical thought. In the workplace, craft is fragmented, not yet replacing workers wholesale, but systematically hollowing out the skills and relationships that make work meaningful and furthers innovation.

Perhaps most troubling: in a society increasingly isolated and dependent on prescriptive tech, many are turning to LLMs for companionship and emotional support, substituting the mimicry of communication for the real thing. AI platforms are inheriting the extractive business model of the social media giants, monetizing algorithmic engagement and manufactured connection. Isolation and dependency are not side effects of these systems; they are design objectives. It's not a bug, it's a feature. The fragmentation of social relations from the ubiquity of prescriptive technologies leads to isolation, isolation forms dependency, dependency demands more of the technology that produced it. Loneliness, it turns out, is enormously profitable.

"What does it say about our society, when human needs for fellowship and warmth are met by devices that provide illusions to the users and profits to the suppliers? The reason I find this particular application of technology so upsetting is that as a response to loneliness, it seems to me deceitful and fraudulent. There are no shortcuts to the investment of time and care in friendship and human bonding, and it is fraudulent to pretend otherwise. When human loneliness becomes a source of income for others through devices, we'd better stop and think a bit about the place of human needs in the real world of technology." — Ursula Franklin ^[1]

Despite being marketed as intelligent collaborators, LLMs are simply predictive models optimized for a predetermined output, not "thinking" machines exercising holistic judgment. They calculate the most probable response based on vectorized language probability and are incapable of genuinely questioning whether a goal is worth pursuing.

The prescriptive nature of LLMs is not a flaw to be corrected — it is what they are. But deployment is a choice. Current deployments prioritize operational control over user agency. Rules structuring their output remain opaque to the user. Outputs trend toward the predictable and familiar, suppressing divergent thinking, and agentic workflows reduce the human to the supervision and quality control of agents — or remove them from the workflow entirely.

While some platforms simulate responsiveness through interface-level feedback —asking clarifying questions or requesting permission before executing— this remains a closed loop. Even training on user feedback does not constitute reciprocity; the user remains a source of data rather than a partner in governance. This is refinement of the system's parameters, not any stake in its design. Interface-level responsiveness, even with a "human in the loop," does not substitute for the reciprocity that matters: the power to decide how these systems are built and governed.

The result is a compliance culture extended to human expression itself —to how we write, think, and create— with the terms of thought dictated by the machine. In LLMs, this logic reaches its conclusion in language, reasoning, and creativity: the last domains we assumed were irreducibly human. The question is not whether LLMs are prescriptive, but whether they could be used differently toward holistic ends.

Earthworm Theory of Social Change

Franklin's answer to the question of how we should build technology is simple and pragmatic: she puts the common good above profit motive. To counterbalance the logic of maximum gain, she proposes a governing principle of minimizing disaster. While maximum gain optimizes for the largest possible return, minimizing disaster designs against the worst possible outcomes — placing the burden of proof on those introducing the risk, not those opposing it. The consequences of a technology often only become apparent once embedded within the complexity of social life, but by then, the damage is already done. We can, however, build in safeguards and preserve the capacity to course-correct.

Franklin offers a "civic checklist" which we should be able to answer when making any technological decisions:

• Does it operate with reciprocity?
• Does it put people before machines?
• Does it benefit the many, not just the few?
• Does it prioritize conservation over waste?
• Does it favour reversibility over irreversibility?
• Does it promote justice?

These are not abstract ideals. They are the baseline for any team, organization, or institution when building or adopting new tools. This is what Franklin calls her earthworm theory of social change: the patient, ground-level work of building a culture of communication and reciprocity — a commitment to open debate over principles rather than deferring to the "move fast and break things" momentum of progress. What she calls redemptive technologies account for more than financial return. They are measured by their social and ecological impact, and by how effectively they bring marginal and disenfranchised communities into the decisions that disproportionately shape their lives.

"Social change will come through seeds growing in well prepared soil — and it is we, like the earthworms, who prepare the soil. We also seed thoughts and knowledge and concern. We realize there are no guarantees as to what will come up. Yet we do know that without the seeds and the prepared soil nothing will grow at all. I am convinced that we are indeed already in a period in which this movement from below is becoming more and more articulate, but what is needed is a lot more earthworming." — Ursula Franklin ^[1]

Earthworming in Practice: Responsible Tech

Franklin's checklist is not a rigid formula; it is a set of questions anyone building or deploying technology should return to at every stage. These questions grow only more urgent as prescriptive tech becomes "smarter" and more ubiquitous. In practice, we should be asking: Who benefits from what we are building, and who is excluded? Are the people most affected by our decisions involved in making them? Are we measuring the true cost of what we ship — not only in compute and capital, but in social and ecological impact? Are we designing for reversibility and avoiding lock-in?

When using intelligent tools, are we reviewing their output critically or accepting it wholesale? When deploying agentic systems that act autonomously, who is accountable for the outcome? Before defaulting to the most convenient platform, have we asked whether a locally run, open-source, or privacy-first alternative would serve just as well?

These are not questions the industry habitually asks, or only asks superficially as optics and marketing. What may feel like compromise or sacrifice is, in Franklin's framework, the opposite: a principled choice that prioritizes data sovereignty, resists dependency, and contributes to a healthier work culture and a more sustainable technological ecosystem.

None of this is straightforward in workplaces where LLM adoption is being driven from above, often without meaningful consultation. The reciprocity Franklin calls for is frequently absent. In this context, earthworming means what it has always meant: advocating within teams for transparency and accountability, refusing use cases that erode judgment or concentrate power, supporting open alternatives where possible, and insisting that the question of how these tools are used belongs in open, ongoing discourse.

These principles point toward a different kind of technical practice, one that moves from compliance to possibility. For corporate leaders, AI promises productivity without what Silicon Valley has cynically called "the tax of human labor." Their short-term productivity gains may prove real. But systems built on extraction eventually erode the trust and cooperation they depend on. Reciprocity is also a competitive advantage: trustworthy, interoperable ecosystems outlast extractive ones. The goal is not efficiency at any cost, but technology that makes meaningful work possible.

We are downstream of decisions made without our input. Franklin's warnings have gone largely unheeded. How do we, as makers in the real world of technology, stay principled, let alone relevant? When these systems offer no reciprocity by design, what room do we have to act?

Before navigating the specifics of redemptive design, we must first commit to shared principles. We cannot build alternatives if we have already accepted the logic of our own displacement. Reciprocity does not emerge from compliant systems — it must be cultivated. This is precisely what prescriptive technologies are, by design, making increasingly difficult. Systems built around compliance, extraction, and the ideology of techno-solutionism leave little room for alternatives to take root. That does not exempt us from the obligation to prepare the soil. We need to demand that AI infrastructure be approached as a public utility, not a private asset.

The AI industry now speaks of "growing" agents — agents trained on our content, our language, our thought. The earthworm's work is to cultivate the culture they grow from, so that what emerges is reciprocal rather than prescribed.

We have built machines of artificial intelligence. But intelligence was never scarce. Wisdom, unfortunately, is in short supply.

"We must protest until there is change in the structures and practices of the real world of technology, for only then can we hope to survive as a global community. If such basic changes cannot be accomplished, the house that technology built will be nothing more than an unlivable techno-dump." — Ursula Franklin ^[1]

[1] Franklin, Ursula. "CBC Massey Lectures: The Real World of Technology". CBC Enterprises, 1990. ↩¹ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷