Augmentative intelligence, not artificial intelligence
AI stands for augmentative intelligence in this context. Same acronym, completely different philosophy. The distinction is not semantic. It determines whether research systems serve the practitioner or replace them.
The same acronym, two completely different ideas
Artificial intelligence, as the phrase is commonly used, describes systems that perform tasks previously requiring human cognition. The framing is substitutive: the machine does what the person used to do.
Augmentative intelligence describes something else entirely. Systems that extend what a person can do, not by replacing their judgment, but by removing the constraints that prevent that judgment from operating at its full capacity. The framing is additive: the practitioner does more, better, with greater reach and precision, because the system handles what does not require their expertise.
The acronym is the same. The philosophy is not.
This distinction is not primarily semantic. It determines how systems are designed, where they are deployed in a pipeline, what they are asked to do, and what kind of trust relationship is possible between the practitioner and the tool. A system designed around artificial intelligence logic tries to produce the final answer. A system designed around augmentative intelligence logic produces material that a practitioner can interrogate, evaluate, and act on.
What augmentation actually means in a research pipeline
A research practitioner brings things to a project that no system can replicate: the framing of the right question, the reading of a client’s real concern beneath their stated brief, the judgment about what a finding means in a specific organisational context, the professional accountability for conclusions that go into a board presentation. These are not tasks that are too difficult for current AI. They are tasks that are structurally dependent on being a person with a stake in the outcome, a relationship with the client, and a professional history that informs every decision.
What a practitioner does not need to do personally is handle volume. Source retrieval across dozens of platforms. Deduplication across hundreds of near-identical posts. Initial filtering of content that fails basic quality criteria before it ever reaches analysis. Theme clustering across a dataset large enough that manual review would take days. These tasks are time-consuming, fatigue-prone, and do not benefit from the practitioner’s specific expertise. They benefit from speed, consistency, and auditability.
Augmentative intelligence puts AI at the volume stage and keeps the practitioner at the judgment stage. Not because the practitioner cannot do volume work, but because doing so is an expensive misallocation of what makes them valuable.
The sequencing matters. Systems that apply AI before data is cleaned are not augmenting the practitioner; they are asking the model to do tasks it is not suited for, and then handing a practitioner output they cannot fully evaluate or defend. The failure modes are consistent: noise treated as signal, model priors filling gaps in retrieved data, prevalence claims that cannot be verified, methodological opacity that accumulates as a liability. For a detailed account of each, see what goes wrong when AI runs too early in the research pipeline.
Augmentation works when the sequence is right: deterministic collection and filtering first, AI for synthesis and pattern recognition on clean source-linked data, practitioner judgment at both ends and throughout. AI belongs after the data is clean, not before makes the engineering case for this sequence in detail.
The accountability test
There is a simple test for whether a system is augmenting the practitioner or replacing them: can the practitioner stand behind every finding?
Not in the sense of being able to point at the tool that produced it. In the sense of being able to explain, to a sceptical stakeholder, the basis for each claim. Which sources. Filtered by which criteria. Leading to which pattern. Interpreted through which professional judgment. Producing this specific conclusion.
A system that produces confident outputs the practitioner cannot trace does not augment the practitioner. It exposes them. The finding is out in the world with their name on it, but the path from data to conclusion is inside a model they cannot open. When the finding is challenged, and findings are challenged, the practitioner has no ground to stand on except “the tool said so.”
An augmentative system works in the opposite direction. It makes the practitioner’s position stronger by making their reasoning visible. Every source link is a brick in the argument. Every filtering decision that can be stated and defended is a demonstration of methodology. Every theme that can be traced back to specific content is a claim that can be examined rather than simply accepted or rejected on the basis of confidence.
The practitioner who uses augmentative tools goes into every stakeholder conversation with more evidence, more clearly organised, more readily retrievable than would be possible without the system. That is amplification. The practitioner who uses substitutive tools goes into the same conversation with findings they cannot fully explain. That is exposure. For a fuller treatment of why accountability is non-negotiable in professional research, see why the researcher has to be able to defend every finding.
Why the industry framing gets this backwards
Most AI tool marketing in the research space emphasises what the tool produces. Faster insights. Automated analysis. Research at scale. The implicit argument is that the tool produces the outcome and the practitioner receives it.
This framing is commercially convenient and analytically wrong. It conflates the generation of output with the production of research. Output is text, numbers, themes, summaries. Research is defensible knowledge that was produced by an accountable process and can be used to inform consequential decisions.
The gap between those two things is where professional judgment lives. It is not a narrow gap. It is the entire space in which a practitioner adds value that a model cannot: the question framing that determines what is worth collecting, the contextual knowledge that distinguishes a significant pattern from an expected one, the institutional memory that explains why the last set of findings was ignored, the communication judgment that determines how a conclusion needs to be framed for this specific client at this specific moment.
A research tool that positions itself as producing research is overstating what it does. A research tool that positions itself as augmenting the practitioner who produces research is being precise about what it does.
The distinction matters to practitioners who want tools they can actually stand behind. It also matters to the organisations that commission research and need to trust the findings that come back. Trust in research ultimately rests on trust in the practitioner, not trust in the tool. A tool that makes the practitioner more capable, more evidence-grounded, and more methodologically transparent earns trust by extension. A tool that obscures the practitioner’s role undermines the very basis on which the research is credible.
What this means for how systems should be built
Augmentative intelligence as a design philosophy produces different architectural decisions at every level.
Data collection is deterministic and auditable. What gets collected is defined by explicit configuration: sources, query terms, date ranges. Not by a model deciding what seems relevant. The practitioner knows exactly what the system looked for and where.
Filtering is rule-based and transparent. Criteria for inclusion and exclusion can be stated in plain language and examined. Any piece of content that was filtered can be retrieved and the filtering reason can be explained. There are no probability distributions involved in the decision. See why we do not let AI run the show for what this looks like in practice.
AI operates on the clean, filtered dataset. It is doing what it does well: identifying patterns across volume, characterising themes, synthesising signal from structured inputs. It is not being asked to decide what is worth including, which is a task it is not suited for.
Every output is linked to its source. The practitioner can trace from conclusion back to specific content. They can assess prevalence, examine the evidence anchoring any given theme, and form their own view of whether the pattern is significant. The system surfaces the material; the practitioner makes the judgment.
The practitioner owns the conclusion. Everything the system produces is input to the practitioner’s analysis, not a replacement for it. The finding that goes into the client deliverable is the practitioner’s finding, informed by systematically collected and processed evidence.
This is augmentative intelligence in practice. The same acronym as artificial intelligence. A completely different relationship between the system and the person using it.
Mimir is built on augmentative intelligence principles: deterministic collection and filtering first, AI for synthesis on clean source-linked data, with every output traceable back to the specific conversations it came from. The practitioner stays in the driver’s seat throughout. Start for free.
Subscribe for news updates.
CSRD and CSDDD do not directly regulate most non-EU suppliers. They regulate the EU companies that buy from them. But through reporting obligations, due diligence requirements, and contractual mechanisms, the practical burden lands on suppliers who are not in scope at all. This article explains how that flow-down works and what it means in practice.