Human-centered systems versus the machine

“We have a choice. Either we follow the techno-centric route and we automate people or we pursue the human-centred route where we informate people.”

- Mike Cooley (2007, p. 393)

“We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter.”

- Sam Altman (2026, March 11)

In a recent editorial, Gill (2025) recounted the founding of this journal and with it the “AI For Society Club, with the objective of applying the new technologies of AI to the alleviation of problems in society” (p. 258).^{Footnote 1} He noted a division between those with and without expert knowledge, such that communication of the transformative potential of information technologies was difficult to communicate. AI was unknown in the 1980s and 1990s, being “those bits of Computer Science which do not quite work yet” (p. 257). Since that time, however, basic research has blossomed into big business, derivative technologies have become commonplace, and transformative results obvious.

If founded today, Gill (2025) reflects, the mission of the AI For Society Club (my emphasis) might be more easily fulfilled. People are generally more aware of the potentials associated with information technologies, many negative, to bring closer attention to prosocial possibilities. At the same time however, through its commodification, AI research has been steered away from prosocial roots. Gill writes: “The atmosphere has changed, from exploration to exploitation. The technology has now left the laboratory. The genie is out of the bottle.” (2025, p. 257) AI research is no longer about solving fundamental problems to potentiate a better future through knowledge. It is about money and power, not the most prosocial animus for a freed genie. Redirecting that animus will require communication to coordinate constructive ways forward, meaning that the mission of the AI For Society club is more important now than ever.

AI & Society follows a human-centered philosophy. Human-centered technology systems, as a field of study, have a long history, with human-centered design in the context of AI receiving increasing attention. One idea in human-centered design in AI is to focus on augmenting or enhancing rather than replacing human capabilities (cf. Shneiderman 2020). Another aspect involves human–computer interaction facilitated through human-like AI interfaces including robots. Researchers at the National Institute of Standards and Technology (NIST) in the USA reported on a small survey of 20 members of the general public and 25 experts about perception of AI, noting that humanness—representing distinguishing features of human intelligence—was a consistent feature (Jensen et al. 2024).

Human-centered design is in general a problem-solving approach involving human beings at every step from ideation, design, development, and deployment. The NIST describes human-centered design in terms of a four-step recursive process in this way (https://digital.gov/guides/hcd/introduction). The lead editor of the Open Forum for this journal, and lead author of this editorial, teaches a similar human-centered process using his ACTWith model, with positive results; the steps correlate. In practice, the focus is on the client with a problem case in need of resolution, and/or user as participatory or co-designer, incorporating feedback during design and prototyping prior to production. The human-centered design approach is intuitive and effective in structuring diverse projects in technology, but it does not represent the level of society, at least not explicitly. Such issues must be addressed through additional course materials.

Human-centered systems, as a study, focus on the social and deeply integrated and embedded aspects of humans, as well as on more immediate conditions. These aspects of human beings, their sociality and historicity, may be under-represented in product development feedback loops framed in terms of client, user or consumer, rather than more holistically. The main idea in this editorial is that we are seeing a similar dynamic playing out at a large scale, now, that this is a mistake, and that resources for repurposing present missteps into something more socially desirable are accessible from the human-centered tradition.

One way to think about technology systems is in terms of clients, consumers, and users. There is another way from the human-centered paradigm.

In 1996, Karamjit Gill wrote about human-centered systems, and what characterizes the approach in the context of information communication technologies (ICTs). He opened that paper with some background on the human-centered movement in Britain, writing:

“The contemporary Human-Centred Movement within the British context has its origins in the historical LUCAS Plan of the early 1970s. The two central planks of the LUCAS Plan, "Socially Useful Technology" and "the user as both a producer and a user of technology" are rooted in the British social movement which sees the individual worker an integral member of the wider society rather than as a Fordist component of the organisation separated from the society. The worker (including researchers, scientists etc.) should not just be concerned with the production of technology but also with its use in society which includes social and ethical responsibilities of technology design and its applications.” (Gill 1996, p. 109).

Channeling Weizenbaum, Gill (1996) broke those two planks into two simple questions; can a technology be designed, and should it be designed “if it does not serve societal purpose” (p. 110). The first he classed a necessary condition, and the second sufficient conditions for technology development:

“From the perspective of the British human centred tradition, both these questions corresponding to the necessary and the sufficient conditions have to be satisfied for a technology or a system to be considered human centred. In other words the technology design is concerned not just with the technical feasibility of design but also with the social desirability of technology.” (p. 110).

Satisfying the client/user/consumer is not enough. Broader societal implications must be considered, and not only by the people at the top, directing the development of the technology, but by anyone affected; this is what it takes to determine “social desirability” on this more holistic human-centered approach. Consider the difference between asking others if they want what you are making, and others asking you to make what they want. The former is tech-driven, machine-centered, the latter responsive to human conditions. The latter approach is present in the human-centered tradition, but deficient in practice, with less than desirable consequences. Twenty years ago, Mike Cooley warned: “It is essential to understand that if we just proceed on this machine centred basis, the kind of mistakes we make will be truly profound.” (Cooley 2007, p. 394) Twenty years later, profound mistakes are being made; this is the point.

Human-centered systems thinking in a holistic way can be compared with machine or technology-centered systems thinking, to disambiguate more and less holistically human-centered approaches. In the latter, researchers develop products and applications for social deployment, for markets, for money, with or without user or consumer input, and then, results are assessed on the affected population (consider agri-chemicals as discussed in Sect. 4). Technological development, in this way, is “in-house” in a cultural sense, in the sense that potentials are constrained by invested interests in proprietary technologies and their popular adoption, i.e., researchers spend lifetimes to succeed, do what they need to do to see that success, and usually expect to be paid. This is a sort of framing through goal-directed activity, after which further development is constrained by interests decided when initial careers, patents, and industrial processes began.

This is where we find ourselves, today, having been funneled through visions into what had seemed necessary, without first asking why beyond the technology-centered question, surrounded by industries solving 50 year-old problems still trying to grow, to dominate, to constrain possibilities, and so determine the future for most everyone else. It is these industries driving social innovation, not the other way around. Technologies are not developed in response to human needs, but rather needs are exploited to market technologies. This is the animus of Gill’s genie flying the bottle. To let it loose is a mistake.

There is vested interest in making a technology work, when how well one does in life depends on that technology working, first. For instance, consider vision. AI exploded with convolutional neural networks and image processing, and this advance in basic science fed surveillance. A lot of people got rich making ID systems and software picking one face from a crowd, now morphed into real-time tracking with civilian targeting integrated into military decision-making. This is a textbook example of technology driven; nobody asked for cameras at every corner, tracking your every move in case you become a target, any more than anyone asked for half the cost of a car to be computers that do the same and let the CIA remotely control it into a tree. In response, people are removing telemetry systems from modern cars. And, in the USA, people are destroying Flock cameras installed by the government for surveillance and enforcement. Computer vision is one thing; ubiquitous surveillance is another.

At the same time, some parts of the world have been mismanaged enough to make people desperate enough to make such technology systems seem necessary. The San Francisco Standard recently published a report on the embrace of Flock cameras by local leaders; “San Francisco once led the fight against police surveillance. Now it’s a laboratory for it” (https://sfstandard.com/2026/03/25/sf-surveillance-state-crime-drones-billionaires/). Conditions might seem to mandate AI for real-time tracking, with robot police dogs and armed drones, to find criminals or follow a single father through a park. Only the guilty have reason to complain. The buzz of a hovering turret should be comforting. If conditions are poor enough, dangerous enough, with people desperate enough, then it might.

Conditions can be created to make a proposed machine-centered solution appear necessary, but when things could be done better, when the outcome of the technology on society as a persistent determinant of the condition of that society is not best for that society, conditions for such development can never be sufficient. People can be leveled down to the machine, for processing as if it were social progress. However, it is progress only from a state of neglect and poor leadership, by all appearances purposeful, and from this same class now, the answer is face-scanning drones with guns. If developed in this direction, target social problems will not be solved; they will be perpetuated through technology infrastructure that cements them in machine automation. Future generations will inherit this infrastructure as material consequences of present intentions, with the same leadership still on top. This is a profound mistake.

With human-centered thinking, the problem comes first, and it is a problem identified not by vested interests, but by regular people with more or less common concerns. These are often social, such as involving corrupted political ecologies (e.g., the “swamp”) and historical injustices as was the case with Occupy Wall Street (e.g., the Federal Reserve Act of 1913) and in general solving problems inherited due to inadequate leadership, in the past. AI might be designed to solve such large-scale social, political, and economic problems, to undo damage done by poor decision-making and systematic application of bad policy over 5 or 20 generations. However, these are big problems representing long-term concerns, fixes for which are typically not the foci of technology researchers. For one thing, technology researchers may not have time to focus on everything required to understand, to anticipate, and then prepare solutions for such problems. They tend to work within the envelop of technology in delivering the tech-centered future that benefits the people who pay their salaries. Tech-workers, especially with residence conditional on employment, might feel more comfortable with cameras on every corner and armed robots at every door than most everyone else; at least they know something about how they work. It is up to others, scholars, philosophers, and researchers, to formalize social dynamics in some clear way, to show how things might be done better. Bringing these visions together is not guaranteed, which returns us to the mission of the AI For Society club.

Ethical responsibility to produce socially desirable technology systems is emphasized in the holistic human-centered tradition, but disappears when commercial interests stand in for social interests, and visions so constrained guide technology development, instead. On a human-centered systems approach, technology development is guided by targets determined outside of any personal investment in a research or policy pipeline. Technology is not at the center of this tradition, people are; technology is a tool for culture, communication, and for social transformation in a more or less cooperative way. The focus is not on creating conditions to multiply customers, to maintain power, or on statistics measuring a step away from insufficient, to better, but worse. The focus is on the situation that we want to create, as integral members of whole and healthy societies, as workers, researchers, scientists, and so on. Communicating this message and creating the conditions to make consistent visions reality, this is the human-centered approach, and the perspective missing on the tech-driven client/user/consumer commercial model.

Thirty years ago, Gill (1996) stressed the need to empower people to drive development of information communication technologies in prosocial directions, bottom–up; “people have to be capable to participate, they need information and knowledge to propose and test their solutions” (p. 122). He saw the potential for AI and associated technologies to facilitate this participation, to foster something like an open democracy, responsive to social needs with associated views guiding technology development. One key to the approach is to think of human beings as subsidiaries of communities rather than as clients/users/consumers, and of information technologies as tools to make this relationship effective in steering technology development to improve these communities and the situations of the subsidiary individuals so embedded.

Gill (1996) stressed the potential for ICTs as tools for communication, to steer social innovation away from technology-driven constraints. However, he also recognized that information technologies are only one tool, ineffective with institutions obscuring and impeding participation rather than supporting subsidiarity relationships.

“… current social innovations, be they technological, about economic regeneration, and educational, are managed and coordinated by mechanisms which are not accessible to ordinary citizens, who are excluded from the formal mechanisms and processes. Those outside the formal innovation processes have to cope with layers of mediation at the local, regional, national and European levels. The issue of subsidiarity is how to use ICTs to bring people nearer to strategic policy makers by removing the competitive mediation hoops.” (Gill 1996, p. 123).

So, he countered that “we need human institutions and processes in place to exploit the potential of ICTs for participatory and consultative activities” (p. 122), proposing:

“… new network alliances between universities (including other centres of scientific knowledge bases) and the civic society to integrate the scientific knowledge base of the university and its expertise into the broader social base of society.” (p. 124).

Ultimately, Gill’s (1996) message is that “technological innovations cannot be divorced from the social dimension”, that “change takes place in the real world of work, living and learning” and further that “[i]t is this dynamic interdependence of the technical and the social which is at the heart of the debate on human centredness and inclusive information society.” (p. 123) When this interdependence is neglected, conditions for development of a technology are insufficient.

Fast-forward 30 years, and the contemporary information society is not so inclusive of subsidiary human interests. The holistic vision of AI for society, that is bottom–up while recursively informing everyday people to motivate research and development, is more or less absent. Popular discussion frames problems accordingly with established technology-centered expectations, with derivative analyses focused through tech-goggled predictive programming. The why question has already been answered, because we can, to exploit needs to market technologies to do more of the same. This is where we find ourselves, today. To move further in this direction is not progress; it is a profound mistake.

Relative neglect of the holistic vision dear to this journal and its editors is investigated in the following section. These are not the concerns of people pumping and dumping data centers to anchor their superyachts off of private Islands, lobbying for lack of oversight and winning in the swamp of politics. What might be done with AI to correct such problems, from within the human-centered tradition.

Section 9 of the annual Stanford Report on AI development summarizes results from different resources polling expert and public opinion (https://hai.stanford.edu/ai-index/2026-ai-index-report/public-opinion). Some information comes from the Longitudinal Expert AI Panel (LEAP), an initiative of the Forecasting Research Institute (FRI; https://leap.forecastingresearch.org/). LEAP surveys are performed monthly, each focused on different issues, drawing on the same pool of respondents in each case. The Stanford report summarizes results from the first polling “waves” focused on broad societal impacts, including for science, and popular and commercial adoption of AI. Discussion below concerns wave 1 of the LEAP surveys, comparing expert and non-expert predictions for AI development over two timespans, which appears in subsection 9.2 of the Stanford report (see Fig. 9.2.3).

LEAP survey expert responses were weighted equally across four categories; computer scientists, AI industry employees, economists, and AI policy experts (see Table 3, p 20, of the whitepaper, accessible here; https://static1.squarespace.com/static/635693acf15a3e2a14a56a4a/t/69e7784f32d1b603f4d3ee23/1776777302811/the-longitudinal-expert-ai-panel.pdf). All groups represent heavy—lifelong, personal, professional, and disciplinary—investment in the future of AI technology development and deployment. Of industry experts, more than half were from the big five, trained top-20 models by compute, top-30 by funding in 2025, or were identified as contributors to top-15 LLMs as reported in 2024. Median citation count was over 9000. Policy respondents were almost double other categories, with 75 of 119 focusing on AI policy for think tanks and AI policy institutes. Of 76 computer scientists polled, more than half were professors and most of those from top-20 schools. Ten percent were among the top 200 cited in AI (relying on OpenAlex.org). Median citations were over 7000. Fifty-four of 68 polled economists were also professors, with 20 from top-50 institutions, and a median citation count of 2200. Such a demographic appears to reflect Gill’s (1996) vision “to integrate the scientific knowledge base of the university” into broader society. This is a good thing.

Non-expert respondents of the LEAP panel include “highly engaged” participants from prior projects, with additional invitations extended to under-represented groups including over-50s who identify as Republican in the USA (to offset noted big-tech political biases), and people stopping with high school education (rather than engineers and scientists doing the tech development. teaching or studying policy) to represent general public opinion. Respondents also include “superforecasters” identified by a private company for having consistently accurate predictions of geopolitical events over multiple years (summary at https://leap.forecastingresearch.org/panel).

One striking disparity between group predictions about AI impacts over two timespans, by 2030 and by 2040, concerned anticipated progress in mathematics, with experts + 25% over non-experts for a major discovery before 2030. The greatest disparity, however, presented in estimates of AI impact overall, on a “technological Richter scale” of disruption delivering “a transformative technological event” before 2040. Experts estimated a high likelihood over non-experts, + 30%. Though the “knowledge base” of universities is represented, here, disparities with non-experts concerning what to expect from the relatively near future of AI technologies are troubling.

Stanford’s report notes that the divide in public and expert expectations is important in the context of rapidly accelerating AI model performance. While an expert may be biased in favor of expecting transformative impacts, experts are also aware of developments that potentiate such impacts, e.g., models linking text with video and multi-media to generate realistic output without prior case training. Non-experts are unlikely to be aware of such models and their potentials, to temper their predictions. The divide is most troubling, in the context of this editorial, not because it signals a difference in knowledge or understanding of the technologies and their implications, but because it represents a barrier to communication. And, world-shaking, Richter-scale transformative events are exactly when open communication facilitating coordination and cooperation between subsidiary members of affected communities is most critical.

Had problem issues been framed from a bottom–up, subsidiarity perspective, with a focus on recursive instruction of prosocial technology design since 1996 (reaching back to Gill) or 2007 (reaching back to Cooley), such a divide in expectations about social impacts due to current technology development would be obviated. More directly, the idea here is that divided expectations present as a consequence of the systematic application of a technology-driven, machine-centered philosophy, and moreover that this philosophy, once identified, is evident at the highest levels of public policy and administration serving money and power over constituent communities. Without mediation, divided expectations confounding informed solutions to coordination problems should be expected to grow, along with the indentured and entitled class divide. This is worrisome as AI model capabilities accelerate toward Richter-scale change. If people are to transition through large-scale institutional rearrangements associated with tectonic social transformation, divided expectations confounding communication and coordination are to be avoided.

Aligning public and expert forecasts, coordinating expectations, requires informing people about potential impacts through public-facing scientific communication, but it also requires setting out—openly, and in accessible terms—sufficient conditions to warrant the development of technologies as they are currently planned. As things stand, human-centered AI design might do what is necessary to make things for people, the way that those people say that they want those things made, but at the same time being tech-driven is a bit like asking someone how they like their pizza. Some people don’t like pizza, and there is no way to convince them that they should. This is a big problem when that pizza is someone’s life and home, negatively affected by AI.

Consider US Presidential Executive Order 1431, Accelerating Federal Permitting of Data Center Infrastructure (July 23, 2025; https://www.whitehouse.gov/presidential-actions/2025/07/accelerating-federal-permitting-of-data-center-infrastructure/), committing Federal lands and resources to the construction of AI data centers “consistent with the land’s intended purpose—to be used in service of the prosperity and security of the American people.” Pew research recently polled more than 8500 people about data centers. Almost 40% responded negatively concerning data center impacts on energy costs and environment, while another 30% noted negative impacts on the quality of life for people living nearby (https://www.pewresearch.org/short-reads/2026/03/12/how-americans-view-data-centers-impact-in-key-areas-from-the-environment-to-jobs/). If the Pew report is accurate, the largest portion of respondents, by far, do not associate data centers with the best interests of their communities. In contrast, only between 4 and 6% of respondents held favorable views of data centers in regards to energy costs, quality of life and impacts on the environment, i.e., someone’s life and home. It is not clear whose prosperity and security the Executive Order is representing, but few appeared to participate in this poll.

One area which has been receiving a lot of recent attention is robotics. LEAP wave 7 concerns Robotics and Physical AI, reporting on responses from 214 experts (panelists as described above), in addition to 55 superforecasters and 630 members of the public, between March 9th and March 31, 2026 (https://leap.forecastingresearch.org/reports/wave7). Within expert respondents, 48 computer scientists, 44 industry professionals, 52 economists, and 70 research staff from policy think tanks and other institutions participated. Experts expect that a robot will be able to make coffee in an unstructured environment, outside of the lab as in a family home, by 2034. Superforecasters expect this two years earlier, by 2032, and the public by 2039. Unassisted surgery is expected by experts by 2040, superforecasters by 2035, and the public again by 2039. As for work, experts expect steadily declining labor participation with a correlate increase in robots.

Though robots in industry are expected to more than triple by 2040, experts reported that their predictions were tempered by tolerable rates of social change. The idea is that people can only support so much labor displacement along with other social impacts from AI and robotics in a given span of time. Where robots appear most efficient, operating in structured, routine environments, experts project more rapid deployment. Assessed in terms of sales in US dollars per employee at Amazon, increases are projected from roughly + 15% by 2030 to roughly + 80% by 2040, with superforecasters projecting more than + 100%. Warehouses are paradigm-case robot-friendly environments; flat, predictable, climate controlled, handling mostly boxes and pallets with regular dimensions in concert with other machines. Warehouses can also be grueling and dangerous places for human beings; robotics here would seem to be a good application, for workers as well as for clients/users/consumers. But, what about on the more holistic view? Though the customer, here the people running Amazon, may be happy, does this satisfy sufficient conditions to inflate extant, global technology infrastructure in this way? Or, are people doing it because they can, because they can pay for it, and shape the future for everyone else according to present vested interests instead?

Where Gill warns of the commercialization of AI, consider the capital centralization and wealth disparity with associated interests driving contemporary technology-centered systems, funding their development. Robots are expensive to make, to maintain, requiring rare-earth minerals and most often exploitative mining, destroying—permanently poisoning—affected areas, such as in the Congo (cf., Kara 2023). Take away capital interests; consider healthy communities. Where, in that picture, do we see sprawling logistics centers pumping small packages around the world, direct to door via-drone copter from semi-autonomous delivery truck at retail, ending up two years later e-waste dumped in Africa? (see for instance Ssebugere et al. (2025) summarizing critical ecological problems due to lax regulation, including forever chemical pollution due to imported e-wastes). Further development in this direction might be great for the people developing current technologies, but is it best for the communities and their subsidiaries, the human beings who live there, and not only live there, but care about the people who live there?

Consider farming. Farming may be expected to be a production process improved through robotics and AI. Machines identify invasive plants, destroy them with lasers, fertilize and water from planting to harvest, and then prepare the fields for winter. Human beings do not need to be exposed to pesticides, herbicides, or other chemical additives, at least not at the point of application. This would seem to be a good thing. The machine-centered model is effective for large tracts of mono-cropped industrial grains, corn and potatoes, so exactly those applications demanding most excessive use of pesticides, herbicides, and other chemicals, to generate revenues more than to grow healthy food. After all, some of these chemicals do not go away, they reside in the skins of fruits and seep into the soil. Damages to microbiome and beneficial insect and other animal populations persist long after application. The National Biosolids Data Project in the USA reported that roughly one-sixth of farmland acres in the USA are licensed for use of wastewater treatment sludge which often contains PFAS (and other “forever chemicals”; https://www.biosolidsdata.org/national-summary). More than half wastewater sludge is dumped, somewhere in the USA. It is difficult to imagine a family farmer, providing produce for locals, treating the land in this way; who wants to raise their children in a toxic environment, let alone participate in their toxification? A robot, on the other hand, may be less bothered.

Consider the meat industry side of factory farming. If robots are doing surgeries by 2034, when should we expect them to grow, butcher and package whole animals like chickens, pigs and cows for human consumption? Sooner? About the same time? Again, great for technology-driven investors looking to corner future meat supply, but given what is known about factory farms, the pollution, the misery and the stench, the disease from close packed despair in filth under artificial lights, suffering hearts racing from the ractopamine,^{Footnote 2} the toxic run off and antibiotic injections, is this really what we want to see? What about the people 30 km downwind of about a million hyper-medicated pigs and cows, millions of caged chickens with beaks clipped and tubes stuffed down their throats? Should they feel better, if it is a robot doing the caging?

Being informed of what happens with and to many or most of the food animals in factory farms motivates people to look for a world as it may have been developed, built from technology solving a different set of problems. The Social Change Lab surveyed almost 4000 people in the US and Britain finding that the more that people knew about factory farms, the more they want them to stop. In the UK, at least 73% opposed factory farms for chickens, for pigs, or for cows. US respondents were slightly more tolerant of factory farming, but overall and in both countries, those for a ban far exceeded those opposed to a ban of factory farming. More than half of British and just less than half, 45% of US respondents, would ban factory farms completely (https://www.socialchangelab.org/post/the-more-people-know-about-factory-farming-the-more-they-oppose-it). It is difficult to see where robo-butchers delivering fresh-cuts by flying refrigerator fit in to this picture. Does such an industry represent a prosperous and secure community? Enriched with hormones and forever chemicals, the meat is cheap. By this metric, lower prices are better; too cheap to matter might be best. By another metric, however, it is much worse than what might have been, and could be. On this view, to cement factory farming in further automation is not socially desirable. There must be a better way.

Here, the divide between expert and the interests of subsidiary members of affected communities appears sufficient to leverage a redress of the contemporary malaise. An alternative may be set out clearly enough to coordinate action. How might AI be best developed to transition to a more just and equitable, healthy use of land and liberty than to mono-crop it in huge tracts with robo-tractors and agri-chemicals including Teflon fed to chickens through tubes clamped to cages just a bit bigger than a shoebox? Consider this divide: with more than half of people asking for an outright ban, this is not a difference in expectations but a hard fork, at least when it comes to food. And, as food is an expected application for AI and robotics, the point here is to illustrate potential for miscalculation in other domains, as well. A holistically human-centered systems approach avoids such infelicities.

Repetitive manual routine performance can be outsourced to machines, from box shifting to plot plowing to corpse dissection. Clients/users/consumers, capital holding companies and insurance underwriters, can be satisfied. But is this sufficient? Are these developments desirable for human beings integrally constrained to resulting societies? How far down do we level people along with the rest of the planet for information processing, before the socially desirable disappears, completely?

To put a heavy line under this point, any rationale for monstrous energy-hogging water-boiling data centers hurriedly assembled to help teenagers around the world to cheat on their homework is insufficient; this much is clear. Why build it out if this is how it is going to get used? If it is for police state surveillance standing in for healthy communities and their subsidiaries, also, insufficient. But then again, ‘why’ is not really the question, for experts invested in outcomes.

The preceding section raises questions about sustainability and commerce more generally. Are massive warehouses involving robots for global trade more socially desirable than support for local economies. Are factory farms more socially desirable than small farms without so much technological dependence, organic, without persistent soil depletion and industrial poisons, that are more efficient and also more rewarding in terms of quality of life for subsidiaries of local communities, more resilient with risks and returns distributed, with better nutrition, less packaging, minimal transport and refrigeration? Local farm markets get by with paper bags and boxes when Amazon will pack a half-kilo steak in styrofoam with dry ice and inflated plastic, say from Argentina, by truck, by train, by ship and plane, finally by delivery drone direct to your door. On one view, for fresh healthy food and sustainable production, none of this is necessary, or at least very little. On another view, there is more money to be made with robots.

In this light, critical care must be taken to see that questions, such as from LEAP surveys discussed in the preceding section, often frame anticipations as positive about AI development, i.e., as “When?” not “if” or “should” or “wouldn’t it be better, if”. Consider, for example: “In what year will a robot first successfully perform an appendectomy—surgery to remove an appendix—on a live human, without physical human help?” (https://leap.forecastingresearch.org/reports/wave7) That this is to become a fact is a given; when it becomes a fact, this is up for debate. The question remains, however, from a human-centered perspective, why? Who is pushing for this? Should technologists anticipate a better solution, rather than automate the world around current practice, instead? Instead of automating appendectomies, why not obviate them? For instance, tiny robots that enter through a small needle and exit with infected tissue while replanting the appendix with beneficial microflora, rather than wait for disease and then remove the whole thing? Still invasive, still robotics, not an appendectomy. Or, stable relationships with local physicians and other experts, facilitated also through ICTs, in free and open communication concerning symptoms as an aspect of overall health, to prevent appendicitis through direct care? Conservatively, perhaps pursue the more holistically human-centered practice, until more socially desirable technology solutions present through ongoing research? This is one suggestion.

Practically, why are robot appendectomies necessary? Human beings have been doing appendectomy surgeries since at least the first recorded instance, in 1735. Of all things to automate, and then to populate treatment centers with rows of machines, why this? If conditions are treated early and injury avoided, then all that may be needed is antibiotics, and even here there is much room for improvement. No robots necessary. Moreover, if injured, and traumatized, the last thing that a human being may want to hear is a robot with a knife. In what ways are robot surgeons superior to their human counterparts, when it comes to caring for a human patient suffering a dangerous condition? No matter how much humanness an AI represents, none will burst an appendix, or be able to empathize with someone who has.

Or, is the idea that robot surgery machines will result in better doctors? A recent study found that physicians suffered a 6% decline in unassisted cancer detection rates after 6 months of relying on AI to do associated diagnostics for them (Budzyń et al. 2025). Is this how we care, for future people, to train robots to help them, rather than direct those energies to communicate knowledge to help them to help each other? With everyone a client/user/consumer, what sort of society is left? Alternatively, are we to one another as subsidiaries of mutually supporting communities, communicating useful ideas, because we care? Rather than appendectomy machines, how about mechanical arms stabilizing human surgeons performing operations aided with image processing to deliver nanometer precision? Help people to care for people, rather than replace them with machines that do not care, at all; wouldn’t it be better, if?

Recall from Gill (1996) that the ���worker (including researchers, scientists etc.) should not just be concerned with the production of technology but also with its use in society which includes social and ethical responsibilities of technology design and its applications.” (Gill 1996, p. 109). Imagine a purposive future built from AI for society rather than the other way around. Imagine a future built from and with technologies not because we can, but because we should, because they are socially desirable; “questions corresponding to the necessary and the sufficient conditions have to be satisfied for a technology or a system to be considered human centred.” (p 110) What development of AI and robotics might satisfy such conditions?

Ideally, longitudinal surveys would be conducted to ask, in an open-ended way, with regular people in recurrent discussion with experts to feel-out potentials for technological solutions to local and long-term problems at the same time, e.g., economic stability, meaningful work, health, and well-being without poisons in the food and Teflon in the soil. Questions might be framed in terms of “would it be better if” with the object overall the desirability, not of the technology, but of the society that results from the systematic application of technology, through daily routines and diets of information so constrained. Nobody wants to live in a factory farm, but this is what we risk making if we do not find a way to repurpose contemporary investments into the planetary panopticon into something that might deliver us to a situation in which such systems are more obviously unnecessary. Through open discourse, AI might be used to help.

AI and associated technologies have potential for compiling solutions to large-scale coordination problems, and moreover communicating those solutions, in natural energetic terms and graphically, illustratively, patiently, to facilitate long-term inter-generational transitions away from contemporary domination of tech-bros and anyone opposed to AI accelerationism. One problem, however, will be how to transform infrastructure produced in this acceleration and that will be inherited by future generations—that freed genie of the digital super-state commodified into national security data centers and AI powered bioweapons labs all the way down to robot baristas by 2034—into something socially desirable.

Though beyond this editorial to fully address, one obvious entry point to reconciliation may be that hesitation of experts to expect over-much from people, in adopting robots to replace them in daily life. Here, there is an interface that may be useful in bringing disparate concerns and their divergent interests together. The question to ask might be what can be done with that inherited infrastructure, to satisfy sufficient conditions for the original construction and outlay of resources, i.e., it would have been better, if.

One way forward might be to turn ubiquitous surveillance systems on their heads, to invert the panopticon into a system that informs everyday people of implications of policy initiatives, in terms of energy and resource costs, environmental impacts, and quality of life for everyone affected. Use information technologies to expedite subsidiarity relationships to inform technological development, driving innovation for socially desirable futures; this is the basic idea. Infrastructure tuned to client/user/consumer tracking and predictive policing via ubiquitous surveillance might be repurposed to simulate possible futures resulting from different public policies and compare them. Given the specificity of individual profiles from accumulated data, how and where a person might end up through different social and institutional transformations may be illustrated. Different people might see an AI play out different possible futures, and ask for options. “Don’t like factory farms? Let’s investigate the alternatives and tell me what you like, better. Then, I will output a graphical representation of the institutional and social transitions necessary to arrive at your preferred social conditions that also illustrates impacts on your personal situation.” Such predictive simulations might help to both inform people, generally, about expected changes, and alert them to potential tectonic social transformations, facilitating coordination of near and long-term plans toward community-wide optimal solutions in an open and accessible way. Such an effort might repurpose existing and near-term planned infrastructure around a holistic human-centered solution to problems exacerbated by historically poor leadership and the machine-centered approach to technology development. It would be better, if.

By enabling subsidiary relationships of integral members of shared and overlapping communities and societies, contemporary and near-term planned AI infrastructure might be repurposed to realize something like the original prosocial promise of the basic research that made it possible. A better world through communication, achieved without the coercion and violence typical of Richter-scale social transformations throughout history—AI can help to make that possible, if this is what people want to do with it.

The technology-centered philosophy guiding AI development is a mistake, a position championed by editors and contributors to this journal since its conception. Some most recent representative works are briefly surveyed below.

Online-first articles scheduled to appear in forthcoming volumes of this journal reinforce the main themes of the preceding discussion. For example, Niclas Rautenberg writes: “Technology is never purely scientific nor technical; it is an inherently political matter. As such, it also requires a political response—even from academics.” (Rautenberg 2026) This message resonates deeply with the founding mission of this journal, as echoed also in Gill’s (2025) clarion call: “For many years, philosophers have been interpreting the world. It is time to change it. … we cannot stand on the sidelines: we are actors in the ongoing process of change.” (p. 258).

Recall the implicit positive framing of anticipated technologies found in LEAP questions. Brauner et al. (2026) measured perceived likelihood, overall value, and perceived risks and benefits of 71 AI scenarios, in order to identify shared and divergent expectations between experts and non-experts. They reported that experts anticipate realization of associated technologies as more probable than non-experts, with lower risks and greater benefits, similar to divergent expectations highlighted in the Stanford AI Report, and conclude:

“If AI and its applications are primarily designed by academic AI experts and practitioners trained within these benefit-oriented mental models, the resulting systems may unintentionally tilt toward technical affordances rather than reflecting broader societal values.”

They call this unintended outcome “procrustean AI”.

“… drawing on the Greek myth of Procrustes, who forced his guests to fit a standard bed by stretching or cutting their limbs. In a modern socio-technical context, this refers to a potential bias in the prioritization of research and the implementation of applications that implicitly privilege expert-centric perspectives. Without deliberate efforts toward participatory alignment, we risk creating systems that require the public to conform to rigid technological frameworks and expert assumptions, rather than developing technologies that are just, inclusive, and responsive to diverse human needs.” (Brauner et al. 2026)

In the end, Brauner et al. (2026) emphasize human-centered AI development through regulation, and advance the mission of the AI For Society club, arguing that “public education on AI could play a role in bridging the identified perception gap”.

On the theme of selfish interests manufacturing the future for personal enrichment, Brian Kelters (2026) describes systematic alignment decay (SAD) as a set of conditions arising through misdirection of evolved propensities counter to long-term human interests. Kelter argues forcefully that:

“… it is only by rigorously curtailing AI proliferation, by sufficiently curtailing AI use … that SAD outcomes may be reliably averted. Making tolerable futures attainable, therefore, requires curtailing the AI industry, through suppressing its capacity to proliferate and advance its defining technology.” (Kelters 2026, Sect. 8; online first preprint page 14 of 17).

Consider the leveling-down of human beings for machine processing. David Jacob Harrison (2026) argues that a “reverse alignment problem” exists in which humans are manipulated to become computable and more machine-like, rather than machines made to align with human values, instead. This message mirrors Cooley’s dilemma, with which this editorial began, and represents the holistic human-centered tradition, reviewed through subsequent discussion. As an example of AI for society, Phung (2026) uses AI to correlate the correction of student drop-out rates with expected institutional burden. Such efforts might inform comprehensive simulations integrating individual choices as mediated by public policy into long-term collective outcomes, using AI as proposed in the preceding section of this editorial. Lucas Scripter (2026) emphasizes the “thickness” of human lives especially through meaningful work, underscoring the subsidiary relationship of the human being with embedding communities discussed in this editorial, the sociality and historicity often neglected in practice, reflecting also social limits on rapid social change, such as with labor replacement by AI and robotics.

Finally, Grandinetti (2026) considers “the various technical and sociotechnical strata at which algorithms exist” and how different models answer different questions, differently, in an effort “to demystify the opacity […] of AI by assessing the code of commonly used models, and, in doing so, avoid the discursive trap of opacity that reifies the mystique of unknowable and all-powerful AI.” This work also directly aligns with the founding mission of this journal, to facilitate clear communication between expert and non-expert, in this case in the context of foundational principles and model limitations. Black boxes are less black, when you know what the inside of a box looks like. This sort of scientific communication is prosocial in the tradition of human-centered systems: better informed people can help better informed people develop better technologies that make future societies better. This is the basic idea.

AI&Society has pursued a human-centered tradition with a focus on social impacts of AI and associated technologies since the start. Emphasizing the subsidiary relationship between each human being and shared, overlapping, contemporary, and historical communities affords a view on the humanness of human-centered AI and design in terms that are essentially social. Understanding AI and technology, as industries, as sub-communities of human beings who are also subsidiary to larger communities, and thus to distributed and distant human beings both in time and space, is the first step in inverting the bottle on the commercial genie, trapping the exploitative animus to redirect those potentials to socially desirable ends. This journal has supported critical research works preparing the knowledge base of institutions including universities and think tanks to meet this challenge for forty years. It is time, as Gill (2025) writes, to act. Contributions advancing the development of AI and associated technologies, including repurposing of available and anticipated technologies in alleviating social problems according with the holistic human-centered tradition described in this editorial, are especially welcome.