Collective memory, collective attention, collective reasoning: the clearest account of collective intelligence on offer, and a fair description of what these tools do. All three act on contributions that already exist. None of them builds the understanding that was not there before.
The best available account of collective intelligence names three components. Gitcoin's 2026 research note on protocols for thinking together sets them out precisely: collective memory, the group's ability to store, retrieve and share information across its members; collective attention, its ability to synchronise focus on the most relevant problems; and collective reasoning, its ability to process information, evaluate options and reach decisions that account for diverse perspectives.
This is a serious framing and it deserves to be taken seriously. It is not marketing language. It has a lineage that runs from Condorcet's jury theorem in 1785 through Pierre Lévy's account of distributed intelligence in 1994 to the founding of MIT's Center for Collective Intelligence in 2006. The tools built on it work. Pol.is clusters opinion at a scale no facilitator could handle. Prediction markets aggregate dispersed information into a price more reliably than most expert panels. Anyone who dismisses this is not paying attention.
But read the three definitions again and notice what they share. Memory stores what has been contributed. Attention selects among what has been contributed. Reasoning evaluates options drawn from what has been contributed. Every one of the three operates on material that already exists. The architecture is complete as an account of how a group handles its inputs, and silent on where the inputs come from.
That silence is the whole problem, because the questions organisations get wrong are the ones where nobody has contributed the right thing yet.
Collective memory stores what was said, not how anyone knew it
A record of contributions is not the same as a record you can reason from a year later. Two people write the same sentence about a market shift. One is reporting a figure from a report. The other has spent a decade with the customers in question and is describing something they have watched happen. Stored as text, those contributions are identical. Retrieved six months later, they are indistinguishable. And the difference between them is exactly the difference that mattered.
Hunome records what Hunome calls the knowtype: the type of knowledge behind a contribution, declared by the person making it. That sits alongside a wider set of Ignite characterisations covering relevance, human context, innovation signal, change orientation and impact. The point is not the taxonomy. The point is that a contribution carries the conditions under which it was made and received by others, so that the group can tell lived experience from inference from measurement when it goes back to the material, and can see when an entire line of thinking rests on one kind of knowing.
Collective memory as normally built has no field for this. It preserves the sentence and loses the ground it stood on.
Attention selects between contributions. It cannot create the one nobody made.
Collective attention is described as synchronising focus on what matters most. It is a filtering function, and filtering is only as good as the pool it filters.
Pol.is illustrates the limit cleanly, and it is worth being precise because Pol.is is genuinely good at what it does. Participants vote on short statements and can submit new ones; the system clusters people into opinion groups and surfaces the statements that bridge them. Deployed through vTaiwan from 2015, it was used on 26 national technology issues, of which vTaiwan's own tally reports around 80 per cent led to government action. That is a real record.
It is also a design that deliberately forbids replying to another participant's statement, because replies produce flame wars. The consequence is that no participant can build on what another participant said. The pool of statements is the sum of what individuals arrived already holding. The clustering is superb; it is clustering of positions people brought with them.
Memory holds what was said. Attention picks what to look at. Reasoning chooses between the options on the table. Understanding is what puts something new on it.
What deliberation adds is the move that this architecture rules out: someone reads a contribution from a part of the organisation they never speak to, and it changes what they now have to say. The thought that follows existed in nobody's head beforehand. Hunome calls the conditions for this assisted serendipity, and it is not a nicety. On the questions that decide whether a strategy survives contact with the world, the useful thought is almost never one that any individual walked in with.
Reasoning toward a decision is not the same as understanding the question
The third component is where the evidence is strongest and the substitution is easiest to miss.
Anita Woolley and colleagues, publishing in Science in 2010, found a collective intelligence factor, the c-factor, that accounted for around 43 per cent of the variance in group performance across a battery of tasks. It is a landmark result. The replication picture since is more measured: Riedl and colleagues, in PNAS in 2021, pooled 22 samples covering 1,356 groups and confirmed the factor exists, while meta-analytic work reported a moderate correlation of roughly r = 0.26 between the c-factor and separate group performance criteria, and Credé and Howardson argued in 2017 that statistical artefacts weaken the case for the construct altogether.
Take the finding at its strongest and it still measures one thing: how well groups perform on tasks that have an answer. That is collective reasoning working properly: options in, decision out, quality assessable against a criterion. It tells you nothing about whether the group understood the situation it was deciding about, because the situation was supplied by the researchers.
Organisations do not get handed the options. Producing them is the work. A group that reasons flawlessly over a badly framed set of choices arrives efficiently at the wrong place, and the reasoning quality of that process will look excellent in any audit you run on it.
If your organisation is making a decision where the options themselves are the uncertain part, the reasoning layer is not the layer to invest in. [Talk to us about how deliberation produces the options in the first place.](https://hunome.com/contact)
The missing component is a living structure of how the understanding formed
The fourth thing is not a fourth box to bolt onto the architecture. It changes what the other three have to work with.
In Hunome, a group's thinking on a question forms a SparkMap: a living structure of Sparks, the individual contributions, that connect, build on and challenge each other, each carrying its characterisation. Because the structure holds the connections rather than a list of items, it shows how thinking moved: which contribution prompted which, where a line of reasoning gathered force, where two parts of the organisation are using the same word for different things. The Shared Understanding Index measures the quality of the understanding being built: whether the conditions are in place for new, credible, actionable understanding to emerge, read across eight dimensions from scale and momentum through clarity, coherence and refinement to emergence. It is deliberately not a consensus score, and it is not a reading of the content. A high score can hold live contestation. It is the quality foundation the other indices rest on, so when it is poor, nothing read off the deliberation above it can be trusted. The Lens raises the analysis worth attending to out of the deliberation, and lets the people running it navigate the same material in different ways: by where the thinking is dense, by what kinds of knowing are carrying a line of argument, by where a challenge went unanswered.
That full output is what Hunome means by deliberative intelligence. Set it against the three components and the difference is structural rather than a matter of degree. Memory becomes a record with epistemic ground attached, so it can be reasoned from later. Attention has something worth selecting between, because the pool now contains thinking that formed during the deliberation rather than only positions that predated it. Reasoning inherits what it works on already framed, and that is rarely just a question: it is as often a challenge to be tested, a claim to be validated, or a trend or a risk that has to be analysed properly rather than assumed.
Collective sensemaking is the process that does this. Deliberative intelligence is what it produces. Neither is a synonym for collective intelligence, and the substitution is not harmless: a leadership team that believes it has covered the ground because it has a memory system, a filter and a decision mechanism has bought three components of an architecture and skipped the one its hardest questions depend on.
What to ask before you believe the claim
There is a single question that separates the two, and it survives any amount of product language: can a contribution in this system change another contribution?
If contributions are stored, ranked, clustered, voted on and summarised but never built upon, the system is doing collective intelligence, and doing it is worth doing. If contributions build on each other, carry how their authors know what they know, and leave a structure that shows how the group's understanding formed, the system is doing something else, and it is the something else that answers questions nobody has answered yet.
The three components describe a group that handles its knowledge well. They do not describe a group that creates any. Most organisations already have the first. What they are short of is the second, and no amount of memory, attention or reasoning will produce it.
Questions people ask
What are the components of collective intelligence? The standard account names three: collective memory, the group's ability to store, retrieve and share information across its members; collective attention, its ability to synchronise focus on the most relevant problems and information; and collective reasoning, its ability to process information, evaluate options and reach decisions that account for diverse perspectives.
Is collective intelligence the same as the wisdom of crowds? No. The wisdom of crowds is one mechanism within it: the finding that the median estimate of a large group often beats any individual expert. Collective intelligence as a field covers a wider set of mechanisms, including prediction markets, opinion clustering and reputation weighting. All of them aggregate judgments that individuals arrived already holding.
What is the difference between collective intelligence and collective sensemaking? Collective intelligence handles contributions that already exist: it stores them, filters them, and reasons over them to reach a decision. Collective sensemaking is the process through which a group builds understanding that none of its members held beforehand, preserved as a structure that shows how the understanding formed. The first tells you what the group chose. The second tells you what the group understands, and on what grounds.
Can collective intelligence tools produce shared understanding? They can reveal where agreement already exists, which is valuable and often invisible without them. They cannot produce understanding that formed during the process, because most are designed so that contributions cannot build on each other. That constraint is deliberate, and it is what keeps large-scale opinion tools usable, but it sets a ceiling on what they can produce.
How do you measure whether a group understands something together? Not by measuring agreement, which can be produced by pressure, fatigue or a well-framed vote. Hunome's Shared Understanding Index measures the quality of the understanding the deliberation is building: whether the conditions are in place for new, credible, actionable understanding to emerge, read across eight dimensions including scale and momentum, multidimensionality, presence, clarity, coherence, refinement, aliveness and emergence. It coaches rather than judges, and it is deliberately not a consensus score, so a high reading can hold live contestation. A group can score high on agreement and low on shared understanding, and that gap is usually where the execution failure comes from later.