IASSIST 2017 keynote questions that still shape research data

The archived IASSIST 2017 website records a conference held at the University of Kansas in Lawrence from 23 to 26 May 2017. Its theme, “Data in the Middle: The Common Language of Research,” connected big data, deep learning, digital agriculture and global food security. The program also preserves plenary details, presentations, check-in guidance, transport information and local recommendations, making the event useful as a snapshot of research priorities and practical conference design.

The IASSIST 2017 Keynote Q&A: Audience Questions and Themes is best read as a map of concerns rather than a transcript of one exchange. Taken with the archived plenaries, it shows researchers asking how spatial information science can bridge research domains. The common issue was not simply access to data. It was whether evidence could be discovered, compared, explained and trusted by people working with very different vocabularies, methods and interests.

That question travels well to Australia. Here, national agencies cluster around Canberra while universities, farms, technology firms and commodity businesses operate across a much larger and more varied market. Australian research discussion often favours direct, plain language; a measured “fair enough” can acknowledge a concern without shutting it down. The conference archive offers a useful standard for that kind of productive, sceptical conversation.

Questions that connected the plenaries

The keynote questions linked technical scale to social purpose. Big data raised questions about volume, speed and infrastructure; deep learning raised questions about evidence and interpretation; digital agriculture and food security raised questions about whose decisions would improve. Treating these as separate topics would miss the central theme. Research data sat between them, and audiences wanted to know how that middle layer could support shared language.

The practical conference information strengthens that reading. Check-in, transport and local recommendations may appear routine beside research sessions, yet they reveal an important principle: knowledge sharing fails when access is inconvenient or context is ignored. The same applies to datasets. A file may be downloadable but remain hard to reuse if its licence, provenance, quality or intended purpose is unclear. The Q&A framed these everyday obstacles as research issues.

Interoperability as a working language

Interoperability was more than technical compatibility. Participants were concerned with metadata, controlled vocabularies, identifiers, persistent links and APIs that allowed a person or machine to move from one dataset to another without losing meaning. The “common language” metaphor was useful because it exposed poor translation. Two projects could use the same words for different measurements, or different words for the same feature.

Australia’s data environment shows why shared standards need institutional work as well as technical work. The Australian Bureau of Statistics and agencies in Canberra publish official statistics, while Data.gov.au aggregates government datasets and state and territory portals add jurisdictional collections. Their catalogues have improved, but differences in formats, update cycles and licence wording still complicate cross-border analysis. Federal, state and local datasets often need careful reconciliation before they can support a national spatial question.

The market is similarly mixed. Public agencies provide much of the authoritative data, while universities, CSIRO, Data61, agronomy platforms, consultancies and farm-technology vendors hold specialised models and operational records. Cooperative Research Centres have supported collaboration by bringing public and industry participants into shared projects. A common metadata profile, clear data ownership rules and agreed pathways for publication could help those partnerships create durable research assets rather than one-off file transfers.

Deep learning, explanation and trust

Deep learning attracted enthusiasm because large geospatial datasets can support image recognition, prediction and pattern detection at scales that manual analysis cannot match. Audience questions pushed beyond accuracy. People wanted to know which evidence supported a prediction, how uncertainty was communicated, whether a model would work in another landscape and whether users could inspect training data and code. A precise forecast can still be misleading if its limits remain invisible.

Australian conditions make those questions concrete. Satellite and farm data must support decisions across the wheat–sheep belt, the sugar and cotton districts of Queensland, and the cattle country of northern Australia. Sparse populations, variable connectivity and regional differences in land use mean that a model trained in one place may fail elsewhere. Teams need external validation across regions, not only a high score on a standard test set.

The practical lesson from 2017 remains relevant to climate, drought, fire and production planning. A neural network should be compared with a simpler baseline, tested against seasonal shifts and accompanied by plain-language explanations of uncertainty. The Bureau of Meteorology, research institutions and agricultural agencies can then use models as decision support without presenting probabilistic output as certainty. Reproducible code, versioned datasets and documented preprocessing are part of trust, not optional extras.

Digital agriculture and global food security

Digital agriculture connected data from field sensors, machinery, weather services, markets and supply chains. The question was whether richer information would improve productivity, resilience and access to food, or simply give large operators another advantage. Global food security made the stakes wider. Research audiences wanted common measures that could connect a farm-level observation with regional planning and international food-system evidence without erasing local knowledge.

Australia offers a distinctive test because the country combines export-scale commodity production with diverse regional conditions. Data from broadacre grain and sheep operations, irrigated cotton and cane, and northern pastoral businesses cannot be treated as one homogeneous environment. The commercial digital-agriculture market spans machinery manufacturers, precision-farming services, brokers, software suppliers and processors, with data rights often linked to contracts or equipment ownership. Technical integration therefore depends on commercial relationships as well as technical infrastructure.

Questions about access were equally important. A platform that reduces yields for a large farm may still be unaffordable or unusable for a smaller producer, while poor connectivity can exclude remote communities. Food-security research must examine affordability, skills, language and ownership alongside model performance. The useful common language is one that helps a grower, policy officer, researcher and community organisation interpret the same evidence without requiring every participant to become a specialist data scientist.

Ownership, consent and public value

The audience repeatedly framed data as a relationship rather than a commodity. Who could reuse a dataset, under what terms and for whose benefit? Did consent cover commercial applications, model training or onward publication? Could a public dataset become a private product without negotiation? These were practical governance questions, and they showed why a technical schema alone could not establish a legitimate research commons.

Australian privacy expectations and the Australian Privacy Principles are relevant when health, worker, household or farm records are linked. Public-sector data is not automatically neutral, especially when groups have historically faced weak control over information about them. Indigenous data sovereignty in Australia calls for authority to remain with communities and for cultural protocols to shape collection, storage and publication. The “fair enough” test is whether affected people can see and influence what happens to evidence attributed to them.

Commercial reuse creates further tension. Sydney and Melbourne technology markets often move quickly from pilot to platform, while agricultural vendors may treat sensor records as commercially sensitive. Licensing should distinguish access from ownership, identify derived data and permit responsible verification. A common language for research must also make room for different forms of knowledge, including community expertise, rather than allowing a dominant platform vocabulary to present itself as universal.

A working agenda for Australian research teams

The archived Q&A points to habits that can be applied directly: document data meaning before sharing, test systems beyond their original setting, and treat governance as part of the method. For Australian projects, these actions need to connect institutions, industry and communities while reflecting differences in scale, distance and market structure.

• Adopt a shared metadata profile that records definitions, licence, provenance, update frequency, quality limits and the responsible contact for every reusable research dataset.
• Publish code, versioned inputs and processing decisions so that results can be reproduced, audited and safely extended by another university, agency or industry partner.
• Pair every predictive model with a baseline, external regional testing and a plain-English account of uncertainty, especially when outputs guide farm or public-resource decisions.
• Set consent, licensing and Indigenous governance before collection begins, with explicit clauses for commercial reuse, model training, derived data and publication.
• Test portability across at least two Australian settings and one international dataset, recording differences in accuracy, cost, connectivity and community impact.

Applied together, these practices make the conference theme more than a slogan. Data becomes a common language when people can trace it, question it, apply it in different places and negotiate its use. The enduring contribution of the IASSIST 2017 questions is a reminder that interoperability, ethical governance and public value must be developed together, not added after the analysis is complete.