AI for Science
Bridging AI to the sciences.
Collaborating with researchers and practitioners across diverse fields to pursue interdisciplinary inquiries through AI.
Purpose
AI for Science brings AI in as a new instrument of observation, replacing knowledge production that depended on experience and judgement with procedures that can be measured and repeated. Affectosphere Group works with researchers across education, political science, ecology and law, carrying over the techniques built in CAI for treating the variance in human judgement as data.
Goal
One structure recurs across fields: an LLM produces the plausible average at high quality, and systematically drops the tails — the rare argument, the minority state, the long tail of naming, the domain-specific rule. AI for Science is less the application of a method than the work of deciding, field by field, what may be delegated and what may not.
Looking for partners
We are looking for those who need AI collaboration
Whether you would like to incorporate AI into your research, analyze data together, or are simply wondering "could AI be useful in my field?" - we are eager to work with researchers and practitioners like you.
Field does not matter. We have collaborated with partners across a wide range of disciplines, spanning both the sciences and the humanities. From method selection, data design, and model evaluation to ethical considerations, Affectosphere Group walks alongside you with the techniques we have cultivated.
To begin, please use the form below to share your research theme or interests freely.
Initial meeting
Free (30 min)
A session to hear about your research theme and challenges, and to consider together whether the question is one AI can address, and if so, what approaches might be possible. Please feel free to reach out as a place to align directions.
Ongoing collaboration
Research and personnel fees
After the initial meeting, if we move forward with a full collaboration, we charge research and personnel fees on an actual-cost basis, depending on the theme, difficulty, duration, and deliverables. An estimated quote will be provided after the initial meeting, once the scope is organized.
Inquiry for collaboration and AI support
Your message will reach Affectosphere Group. We will follow up to arrange the initial meeting.
Case studies
Case studies
A brief introduction to the AI-driven collaborative research that Affectosphere Group has carried out with researchers from other fields.
Using AI as the analyser creates a new validity problem: the bias of the analyser itself. So in AI for Science the methodology of evaluation becomes as much the object of study as the result.
Education × AI
Generative AI lifts the quality of student writing sharply, while the rarity of its claims sits at about a fifth of a human's and it strips roughly seventy per cent of the variance out of how arguments are joined. Writing assessment should move from “was this written well” to “is this thought anyone else's”.
Political science × AI
From sentiment analysis over 200,000 Japanese tweets, and from measuring how far predictions invert when the country names are swapped, we quantified the national bias sitting inside LLMs. Scanning the large pre-training corpora shows that the skew is already present in the training data.
Ecology × AI
Across 48,464 spider species, we evaluated automatic classification of the etymology behind their scientific names. Names derived from morphology or geography approach human accuracy; names derived from behaviour or culture fall away sharply — and the boundary of what can be delegated shows up right there.
Law × AI
Given ten years of the Japanese real-estate transaction licence exam, no model reached the pass mark. Not a stage of full substitution, then, but promising as support for a learner or an assistant to a legal team.
It is where diagnostic technique built in CAI is carried out to questions that matter socially — and, through the exchange with domain experts, where CAI's own methodology gets reworked.