Academic field 1.2 · 14 ideas
Computer and information sciences
research ideas.
Qualified university research directions mapped to Computer and information sciences. They are starting points for further scoping, literature review and supervision. They are not completed proposals or claims of novelty.
Ideas in this field
Search and filter these ideas →Decision-Making Frameworks for AI Moral Patienthood Under Conditions of Epistemic Uncertainty
This research explores how to establish ethical frameworks for the potential moral status of artificial intelligence when empirical evidence regarding machine consciousness remains fundamentally inconclusive. It seeks to define institutional and design-based approaches to precautionary action in the face of irreducible uncertainty.
Integrating Heritage Science and Machine Learning for Object Provenance and Restitution
This research explores the synthesis of material analysis, isotopic data, and high-resolution imaging with machine learning to reconstruct the historical biographies of cultural artifacts. The project aims to develop computational frameworks that can identify potential illicit provenance in digital marketplaces and facilitate evidence-based restitution processes.
Does AI screening mark down research summaries written in a second language?
Large language models increasingly help screen and rate research writing, and they have been shown to prefer text written by other language models. It is unclear whether that preference partly penalises people writing in an additional language. The idea tests whether AI ratings of research abstracts differ between first-language and additional-language South African writers, and whether AI editing closes any gap once underlying quality is held constant.
Do multilingual AI tutors narrow learning gaps without flattening local pedagogy?
A comparative study of AI-supported tutoring in multilingual classrooms, focusing on learning outcomes as well as the survival of local examples, idioms and teaching practices.
Do AI quality scores agree with South African experts on South African research?
Language-model scores of research quality correlate only weakly with expert judgement at the level of single articles, and the strength of that link varies by field. Evidence from African research systems is almost absent. The idea compares AI quality scores for South African articles with blinded expert ratings and tests whether prompting for national relevance changes the agreement.
When does retrieval grounding actually reduce factual errors in language-model answers?
Retrieval-augmented generation is widely adopted on the assumption that grounding a language model in retrieved documents reduces fabricated or incorrect statements. This direction asks when, and by how much, retrieval actually reduces factual errors in a specialised domain, and how the quality of grounding itself should be measured rather than assumed.
How far do general image-segmentation foundation models transfer to under-represented domains without fine-tuning?
General-purpose image-segmentation foundation models perform strongly on common benchmarks, but their behaviour when applied without fine-tuning to niche or low-resource imaging domains is poorly understood. This direction characterises where zero-shot transfer succeeds or fails and what features of a domain predict failure.
Which data-scarcity strategies actually generalise across low-data problems?
Many techniques promise to rescue deep-learning performance when training data is scarce, including data augmentation, transfer learning and synthetic data. This direction asks which of these reliably help across different low-data problems, and under what conditions each strategy fails.
How reproducible are reported gains on machine-learning benchmarks under independent re-evaluation?
Progress in machine learning is tracked largely through benchmark leaderboards, but independent, standardised reproduction of reported improvements is rare. This direction asks how often reported gains hold when re-evaluated under matched conditions, and what factors explain irreproducible results.
Epistemic Frameworks and Auditability in Autonomous Scientific Discovery Agents
This research explores the development of closed-loop artificial intelligence systems capable of generating, testing, and falsifying scientific hypotheses. It focuses on establishing robust epistemological standards, transparent audit trails, and error-attribution mechanisms to ensure the reliability of machine-generated scientific outputs.
Verification Protocols for Quantum Computational Advantage in Fault-Tolerant Systems
This research explores the development of quantum algorithms designed to demonstrate computational superiority over classical simulation within the 100–200 logical qubit regime. It emphasizes the necessity of robust, independent classical verification methods to mitigate the risk of premature or inaccurate claims of quantum advantage.
Automated Decomposition and Formalization of Mathematical Proofs in Lean
This research explores the development and benchmarking of computational systems capable of decomposing complex mathematical conjectures into verifiable lemmas. It further investigates the autoformalization of informal mathematical proofs into the Lean theorem prover to address current bottlenecks in proof verification.
The hidden water footprint of small urban data centres
Hyperscale facilities are studied closely, but small edge and enterprise data centres may create a dispersed water burden that is poorly measured and rarely visible to city planners.
Automated Color Grading Techniques for Cross-Scene Visual Consistency in Studio Production
This research explores the application of artificial intelligence to the challenge of color matching across disparate shots in high-end film and video production. It seeks to identify and evaluate computational techniques that can maintain visual continuity and studio-quality aesthetic standards.