How we work with AI is being shaped in many places, from regulation and technology development to organisational choices and everyday professional practice. This post reflects on what experience from environmental management and collaborative practice might bring to thinking about how those influences connect, how we learn from what happens, and where responsibility sits along the way.
In an earlier post on designing AI for the wider world, I suggested six principles that might help us think about better AI development and use. It was part of a wider collection of essays exploring AI across practice, organisational and structural levels. The post ended by recognising that responsibility does not rest with technology developers alone. Governments, organisations, professions, communities and users all help shape how AI is introduced and what becomes normal practice. But that still leaves the question of how these different forms of influence connect, and how they add up.
Coming to this from environmental management, adaptive management and collaborative practice, parts of the situation feel familiar. In those settings, we rarely expect one intervention to determine the outcome. Regulation matters, but so do organisational decisions, professional practice, markets, communities and the choices people make in their everyday work. What happens in one place affects what becomes possible elsewhere. That experience provides one way of thinking about how AI is being shaped.
Influence is spread across the system
It is tempting to focus discussions about AI governance on regulation. Regulation clearly matters, especially where risks are significant or individual organisations and users have little ability to influence what technology providers do.
But decisions are also being made by developers, organisations buying and implementing systems, professional bodies, standards and assurance organisations, and people deciding how they will use AI in their own work.
None of this is particularly new as an observation. Work on responsible AI already recognises many of these different sites of influence, and there are mechanisms intended to connect some of them. In practice, though, those connections may be less straightforward, particularly for changes that emerge gradually and are difficult to identify as particular harms or incidents.
Experience with AI in everyday practice can influence organisational policy and procurement, while learning across organisations can inform professional standards, assurance arrangements or regulation. But that does not mean experience will necessarily travel in these ways, or remain unchanged as it does. Influence also travels in the other direction, sometimes simply as another set of rules passed down to practitioners.
Some of this we have seen before
Environmental management has had to work with problems where causes and effects are spread across many actors and over time. Water quality provides a familiar example. It has generally been easier to regulate pollution coming from an identifiable discharge than diffuse pollution accumulating through many activities across a catchment. The latter may result from thousands of individually understandable decisions about land use and management. Improving the overall outcome requires attention to how those decisions add up, alongside some combination of regulation, incentives, monitoring, changes in practice and collective action.
Something about the emerging effects of AI feels similar. Some concerns can be traced to identifiable decisions by developers, governments or organisations. Others may accumulate through thousands of everyday choices about when to use AI, what to delegate to it, which outputs to accept and what forms of judgement we continue to exercise ourselves. But those everyday choices are not made independently. They are also influenced by the tools people are given, the organisations they work in and decisions made elsewhere.
We have also learnt that recognising this does not make the problem easy to manage. Collaborative processes can improve relationships and understanding without necessarily changing environmental outcomes. More monitoring may produce better information, but that does not ensure it will affect subsequent decisions. Voluntary approaches can support innovation and build capability, but they can also delay harder decisions about limits or regulation.
These experiences also help explain the continuing importance of adaptive management and learning. When we cannot confidently predict the consequences of an intervention, we need ways to notice what is happening and adjust what we do. With diffuse environmental effects, this includes monitoring the cumulative condition of the wider system rather than looking only at individual actions.
For AI, this suggests paying attention to cumulative changes in professional judgement, participation, capability or organisational practice, and to how we might notice them.
Monitoring and evaluation can contribute, but only when the information they produce enters processes where somebody can respond to it. Different people also see different parts of a situation. Those affected by a decision may notice consequences that are largely invisible to those making it.
Creating opportunities for those perspectives to enter decision-making can improve our understanding, although that does not mean they will necessarily carry influence. We have learnt from environmental management that knowing more about what is happening does not automatically mean that those with the capacity to respond will do so.
AI within the feedback process
People using AI in their work can be important sources of information about what is happening. A facilitator may notice that an AI-generated synthesis has flattened differences between participants. In evaluation, AI might help identify patterns in a large body of qualitative material while leaving questions about what disappeared in that first synthesis. Within an organisation, apparently straightforward uses may gradually change how staff undertake tasks or exercise professional judgement.
Those observations could provide useful feedback. They might change individual practice, organisational guidance, training or procurement. Accumulated across many settings, they might also contribute to professional standards and wider governance.
AI is also becoming part of these feedback processes. It is already being used to synthesise workshop discussions, consultation responses and interview material. It can summarise evidence and help draft organisational policies and reports. It is also becoming part of how people search for and interpret information about what is happening.
Of course, human processes have always mediated information. A facilitator decides what to put on the flipchart, an evaluator which themes matter and a report author what to include or leave out. There was never an unfiltered pathway through which experience simply travelled upwards into better decisions.
The issue is partly how visible that mediation is, and whether there are opportunities to question it. When a synthesis is developed with a group, participants can ask how an interpretation was reached, challenge it and revise it. Human synthesis undertaken later and away from the group can already make that process less visible. AI can add another layer. A plausible summary may arrive without anyone being quite sure why some things have been foregrounded and others have disappeared.
Some of these issues are explored in earlier posts through the authority of the first synthesis and the possibility of creating a record of participation without the participation itself. There is a related issue when AI becomes part of the processes through which we learn how AI is affecting us.
That extends beyond formal evaluation. Organisations learn through meetings and reports, professional conversations, consultations and the accumulated experience of their staff. Governments and professional bodies depend on similar flows of information. If AI increasingly helps organise and interpret those flows, then it is entering some of the feedback processes through which its own effects may eventually be recognised and addressed.
I do not think this means that AI should be kept out of them. I use AI myself to explore ideas, test interpretations and work with material. The issue is whether we can still see enough of the interpretive process to question it, bring in other perspectives and recognise when something important may have been lost.
Connecting partial influence
There is unlikely to be one place from which the development and use of AI can be adequately shaped. Nor does recognising distributed influence mean that every actor carries the same responsibility or has the same capacity to act. A technology company making decisions about a model has access to information and resources that an individual practitioner does not. Governments can regulate in ways that professional associations cannot. Practitioners, meanwhile, may see effects in everyday work long before they become visible at those other levels.
Connecting these different forms of influence is less straightforward. Experience from everyday practice may or may not find its way into organisational choices, procurement or wider governance. Decisions made elsewhere may also become difficult to challenge or revise as they become embedded in practice.
There is a tension here. As influence becomes distributed across more actors, responsibility can become easier to pass elsewhere. The developer points to the organisation using the system. The organisation points to professional judgement. The practitioner points to the limitations of the technology they were given. Responsibility can end up closest to the point of use, even where much of the capacity to change the system lies elsewhere.
How AI develops and becomes part of our work is already being shaped in many places. Keeping connections between those places may help what is being learnt in practice reach people in a position to respond. It still leaves us needing to be clear about who has the capacity to respond, and where responsibility sits.
This post forms part of my wider work on AI and professional practice. See the AI for reflective practice, research and collaboration hub, or read the essay collection Working well in the age of AI: Judgement and responsibility in professional and collaborative practice.
[* Image by Jintana / Adobe Stock]
