The next speakers in this session at the SEASON 2025 conference are Jutta Haider and Malte Rödl, whose interests are in what they call ignorance logics in search, with a particular focus on environmental issues. Algorithmic systems are involved in the shaping of knowledge, and of what is knowable and can be known; and societal responses to ecological crises, in particular, are now failing not so much because of a lack of knowledge but because of the possibility of ignorance. This is due in part also because of the curation of information by algorithmic black boxes.
Which worlds are made less likely, plausible, and visible by search engine algorithms, then, and why? What forms of ignorance are made possible by them – from a range of definitions of ignorance as a virtue, as selective choice, as passive construct, or as active construct? On, in, and through multi-sided platforms, such ignorances tend to result from the sociomaterial configuration of the ignorances of different other actors and groups.
In other words, what are the ignorance logics of multi-sided platforms? How do they differ between conventional search, retrieval-augmented generation (RAG) and generative artificial intelligence (GenAI) contexts? In search, semantic analysis of source texts results in search rankings, and the user retains some authority over source selection; in GenAI, prompts are extruded through training data and potential ignorances result from word similarities that are not surfaced. These issues combine in RAG: prompts are transformed to queries via semantic similarity; ignorance results from then illusion of certainty through sourced texts; and it also results from the absence of source rankings. RAGs thus combine the worst of both worlds.
This can be scoped especially in environmental search contexts; but these observations may also be transferable to other topics where extreme views are normalised or imbalanced power relations are perpetuated. For instance, querying Microsoft Copilot (in both ‘quick’ and ‘deep’ versions) for the term ‘climate alarmists’ results in limited detail that loses context and enables certain ignorances. Querying Google for “summer clothes” results in a consumerist response offering ads and shopping opportunities; ChatGPT (in its earlier versions) offered an encyclopedia definition, but this might have moved more towards shopping opportunities in recent time.
Such patterns are also strongly influenced by whether there is a Wikipedia article on the topic being queried: GenAI responses in particular frequently base their responses on such source materials, and diverge notably from such authoritative information where they do not exist. The concept of ignorance logics is therefore a very useful concept and method for thinking through and exploring the effects of search infrastructures on the information they provide.
