When Signals Stop Making Sense

A while ago I wrote an essay exploring what grounded theory might contribute to horizon scanning and anticipatory futures intelligence, you can read it here.

At the time I was exploring alternative and supplementary coding methods & tagging systems in an effort to make my own horizon scanning architecture and taxonomy more robust, but looking back now I think I was starting to noodle on a much deeper challenge. Initially my system taxonomy was focused around categorising signals of change so that I could cluster industry or domain-specific signals, and capture some of the potential impacts of those signals as they intersected with others in my database.

But I’ve increasingly started to wonder whether many of our current horizon scanning systems struggle because they quietly assume that meaning stabilises fast enough for our categories to keep up. We collect signals, tag them and cluster them. We synthesise them into reports and more often than not, those signals become static artefacts, fixed in the moment of capture, surviving in our systems as the world around them shifts. We might even resurface those same signals in research a year or two later, and compound that old sensemaking.

But we know by now that the systems we are trying to understand don’t behave that way. The meaning of a signal changes as new technologies emerge, systems interact, institutions adapt, narratives shift and second-order effects begin to surface. Think about the way in which AI and tech stacks are reshaping cognition or climate is reshaping economics. Let’s imagine a signal about synthetic biology that initially appears relevant to our healthcare innovation project so we tag it to that space along with adjacent tags. Later, we realise that same signal has become entangled with our emerging knowledge about biosecurity, supply chains, insurance markets, AI-enabled research, humanitarian preparedness, or even geopolitical competition. Simply adding more tags would record the connections, but it doesn’t help me to see my own understanding as it evolves.

The original scan hit we saved is unchanged, but its significance or contribution to understanding has changed over time. That shift could be from a change in the world, in our own understanding or even in the inclusion of additional sources, expert or previously excluded perspectives. The reality is that a futures researcher, humanitarian strategist or community partner might all interpret the same signal differently.

I think this understanding creates a real challenge for futures and foresight work.

Many of our tools still optimise for:

• categorisation

• storage

• retrieval

• reporting

When increasingly, I think we may need systems that optimise for:

• reinterpretation

• recursive learning

• evolving relationships

• adaptive sensemaking

Not because rigor no longer matters. If anything, I think the opposite is true. As AI becomes embedded inside scanning and synthesis workflows, the need for methodological discipline increases substantially, but I’m becoming less convinced that rigor should mean fixing meaning too early. One of the most useful ideas I encountered through Grounded Theory was the idea that coding is not simply about classification, it’s actually part of the knowledge building process itself.

Codes evolve, themes collapse and recombine and interpretation changes through repeated engagement with the material. Meaning becomes something that is worked on iteratively rather than a label imposed upfront. This feels increasingly relevant to my horizon scanning practice, particularly under conditions where AI retrieval systems can prompt me to revisit earlier interpretations.

It’s worth clarifying that in no way am I claiming that an ‘evolved’ scanning system in any way amounts to a grounded theory study. I’m borrowing practices here, because the discipline of returning to the material to see what no longer fits, is a useful one.

Many have already contributed important work to this area of futures intelligence, Toni Ahlqvist and Tuomo Uotila’s exploration of the contextual and relational character of weak signals is one pertinent example (Ahlqvist & Uotila, 2020). I’m currently working my way through a map of the key literature on horizon scanning, weak signals and emerging change and will post more on this as the research map takes shape.

This is one of the challenges I’m noodling on for my own scanning system; trying to understand how we can preserve signals whilst allowing space for interpretations to shift over time and to be able to identify the reasons why.

In these environments, forcing coherence too early can easily create a false sense of stability. I’m starting to think that one of the most important capabilities in futures intelligence may be the ability to preserve ambiguity long enough for more meaningful patterns to emerge. I’m not talking endless ambiguity or abandoning interpretation altogether, although clearly long enough needs a practical boundary.

The question is how might we might structure a research / information architecture that enables new information, new consequences or new insight to feed back into our framework of understanding?

When I think about what recursive learning looks like in a personal human context, it means not just returning to previous observations or signals of change, but also to the taxonomies and assumptions through which I initially understood them. . and importantly, allowing the impacts or consequences of previous thinking sequences to challenge both.

Recently I came across a paragraph from Indy Johar that stuck with me:

Breadth remains useful, but it does not address the degradation of the frames through which breadth operates. A person can move across many domains and still carry abstractions whose conditions of validity have already shifted. They can connect widely and still connect through categories that no longer hold. Lateral range does not, by itself, secure epistemic adequacy.

— Indy Johar · Don’t become a generalist. Become recursive.

I’ve kept returning to that as I think about horizon scanning systems.

Because I think one of the real risks in futures work is not simply missing signals. It’s becoming overly attached to interpretive structures whose usefulness may already be decaying. There’s an obvious power question here too right? In an organisational group or scanning network, who gets to decide that an interpretation has improved? In humanitarian and development work, a donor’s account and a community partner’s perspective might differ for reasons that more data won’t resolve. What would it look like to have a system that could preserve competing interpretations as well as the history of revisions?

In the spirit of borrowing practice and perspective from other disciplines, along with complexity informed sensemaking and qualitative emergence methods, I’ve long been interested in what structured intelligence tradecraft could contribute to this work. I’ve explored this before in ‘Shifting towards an explicit futures intelligence scanning methodology’ and ‘Beyond the Binary’. I’ve been exploring structured analytic techniques from the intelligence community around:

• confidence assessment

• analysis of competing hypotheses

• source reliability protocols

• chain-of-custody reasoning

• assumption testing

• analytic transparency

Not because I think foresight should become militarised or overly rigid, but because I think anticipatory systems increasingly need both emergence sensitivity and traceable reasoning. I’m interested to explore how I could surface the relationship between evidence and exploration more explicitly in my own work, how could I visualise the evolution of emerging hypotheses I’m tracking and would that even be useful? Something akin to a human chain-of-thought (CoT). A weak hypotheses can still inspire a valuable thoughtstarter or question within a project, without necessarily becoming ‘strong evidence’ for a decision gate.

Foresight research has documented persistent gaps in the traceability of scenario construction (Kosow, 2015; Weidenfeld et al., 2026), while critiques of conventional US intelligence highlight the limitations of reductionist and extrapolative approaches to emergence (Kerbel, 2019). Futures research on weak signals emphasises that their interpretations depend on observers’ perspectives and contexts (Ahlqvist & Uotila, 2020). The chewy challenge here is how to bring imaginative exploration, auditable reasoning and openness to emergence into the same practice. How do we cultivate a robust horizon scanning process and research infrastructure that enables imaginative exploration and makes sense-making auditable, whilst keeping analysis open to emergence?

Especially if AI systems are increasingly participating in synthesis, clustering, summarisation, interpretation or reasoning and decision-making. “Human-led, AI-assisted” only means something if the reasoning process remains visible. Otherwise we risk creating systems that produce polished outputs while obscuring how meaning was constructed in the first place.

At the moment, I don’t fully know what the alternative looks like. Maybe this means treating signals less like database entries and more like living semantic objects capable of being reinterpreted, re-clustered, re-contextualised and connected in new ways as new information emerges.

I’m still working through this thinking publicly and experimentally but increasingly, I suspect the future of anticipatory intelligence may depend less on building better predictive models and more on building better systems for adaptive sensemaking under conditions where meaning itself is both transparent and fluid with a strong chain of custody.

What Else

Like everyone who has been furiously experimenting with AI research architectures, I’ve been building my own futures research and intelligence system that attempts to address some of the ideas I’ve been exploring here - an AI-augmented human-led research system. I’ve been testing my model, breaking it and rebuilding it over the past 6 months and will share some of the learnings here when I get a chance.

The aim will be to share some worked examples of this process in practice; the initial sense-making and subsequent changes, what worked or failed, and what difference it made. At this stage the design intentions mentioned above are questions I’m thinking about, not challenges I’ve solved.

In the meantime, if you’re working in strategic futures intelligence or innovation-for-good somewhere in a space leaning toward better human futures - I’m all ears. I’d love to hear what you’re noodling about or what would be useful in your work.

I’m developing this with the intention of sharing a kind of open research infrastructure in the hope that it might spark something in someone else. Something you might find useful to share, reuse, test, fork, break or rebuild. Whatever moves it forward.

More soon.