đ§ Four activities provide opportunities for reflection, critique and discussion.
AI was used (but not harmed) in the generation of these training materials!
Alan Turing argued that intelligence might be understood through behaviour rather than internal processes, leading to the famous question: Can machines think?
Early AI focused on symbolic reasoning.
Intelligence was treated as a problem of logic, rules, and formal symbols.
If reasoning can be fully specified, it can be automated.
Large, structured knowledge bases emerged.
Expert knowledge was represented using rule sets, semantic networks, and frames.
Expert systems became prominent as a way to formalise human expertise.
đ§ Early AI treated thinking as rule-following, not pattern-learning.
Expert systems struggled to scale.
Human expertise had to be encoded manually, often as thousands of explicit rules for diagnosis, recommendation, or classification.
đ§ At the same time, early research into neural networks stalled following early mathematical criticism, accelerating the loss of confidence and funding.
in the 80s ML emerged when researchers realised that instead of telling computers how to solve problems, they could let them learn from examples.
Three conditions enabled this transition:
ML became the dominant paradigm because it could adapt to messy, uncertain, real-world problems.
ML is based on three main approaches:
| Approach | What It Does | Typical Use Cases | Examples |
|---|---|---|---|
| Supervised Learning | Learns from labelled examples to predict outcomes | classification, regression | regression, random forests, SVMs |
| Unsupervised Learning | Finds structure in unlabelled data | clustering, dimensionality reduction | k-means, PCA, topic modelling |
| Reinforcement Learning | Learns actions through trial-and-error with rewards | control, optimisation, agents | Q-learning, policy gradients |
In the mid 90s, research shifted toward data-driven methods, with probabilistic graphical models and support vector machines (smvs) becoming standard in scientific and industrial settings.
A landmark moment came in 1997 when IBMâs Deep Blue defeated Garry Kasparov, demonstrating the power of specialised computation and statistical optimisation.
đ§ Deep learning transformed AI performance across multiple domains by learning rich features automatically.
In the early 2000s Natural Language Processing (NLP) emerged as a field of AI that focuses on understanding and generating human language. It focused on tasks such as translation, sentiment analysis, summarisation and question answering.
graph LR
D["Input data"]
I["Input layer"]
H["Hidden layer"]
H1["Hidden layer"]
H2["Hidden layer"]
O["Output layer"]
R["Output data"]
I -->|Feed forward| O
O -->|back-propagation| I
D --> I
I --> H
H --> O
I --> H1
H1 --> O
I --> H2
H2 --> O
O --> R
Language is challenging because meaning depends on context and relationships across sentences.
Various deep learning models attempted to address the NLP problem, but these had their limitations:
đ Researchers needed models that could look at text in larger blocks (or all at once).
Transformers introduced something called self-attention, this essentially allowed models to attend to all elements of an input simultaneously. learning contextual relationships across the entire sequence.
đ This unlocked large-scale training and dramatically improved performance across language, vision and multimodal tasks.
sequenceDiagram
participant U as Prompt
participant M as LLM
U->>M: Provide prompt
M->>M: Compute attention
M->>U: Output token
loop Generation
M->>M: Update context
M->>U: Emit next token
end
By predicting the most likely next token from prior context, transformers learn statistical patterns that can be reused across tasks, enabling flexible inference beyond simple memorisation.
⨠Given a large enough training set, this allows them to generate responses the appear thougtful.
Scaling laws reshape AI research
Transformers trained through self-supervision on very large large datasets have emerged as the most common general-purpose models capable of adaptation across diverse tasks.
| Property | Description |
|---|---|
| Scale | Trained on trillions of tokens |
| Generality | Perform across domains |
| Transferability | Effective adaptation with small datasets |
đ§ What we think of as AI now are essentially just massive transformer based models.
đ Foundation models marked a transition from task-specific AI systems to general-purpose computational platforms.
AI began with rules but moved toward learning. Deep learning enabled rich representations, and sequence models advanced language processing. Transformers removed key bottlenecks, allowing the creation of foundation models and LLMs.
Big shift now is on optimisation as running large models is expensive!
đ Modern generative AI is the result of many incremental breakthroughs.
Generative AI represents a structural change in research by compressing intellectual labour associated with drafting, synthesis, translation, and technical scaffolding.
Rather than replacing expertise, it reallocates researcher effort from production to evaluation, verification, and judgement.
The key transformation is therefore epistemic rather than technical: researchers are increasingly positioned as editors of reasoning rather than sole producers of intermediate artefacts.
AI-enabled workflows can collapse traditionally sequential stages of research, reducing temporal distance between ideation, analysis, and dissemination.
graph LR A["Question framing"] --> B["Evidence synthesis"] B --> C["Analysis & modelling"] C --> D["Dissemination"] A -. "AI-assisted shortcuts" .-> C B -. "AI-assisted drafting" .-> D
This compression increases throughput but reduces natural pauses for critical reflection, peer discussion, and methodological reconsideration. Opportunities therefore emerge only when deliberate checkpoints are reintroduced.
Opportunities from AI in research are best understood as capacity multipliers
Common highâvalue gains include:
These benefits are particularly significant in environments characterised by limited time, high cognitive load, or interdisciplinary translation.
AI can increase the breadth of exploration by lowering the cost of asking âwhat if?â questions, it can:
This can improve robustness if exploration is paired with explicit selection criteria and validation.
When used carefully, AI can also support:
These gains offer both social and pedagogical impact, not merely technical.
Plenary Discussion:
For each research stages State how AI could reduce effort at that stage. Can you also identify potential risks of the approach?
AIâenabled capacity is valuable only when the researcher retains ownership of:
Where ownership is ceded, opportunity are rapidly converted into risks.
Effective use of AI in research therefore requires intentionally designed safeguards:
Opportunity only scales with control. Speed without control does not create research value.
Generative AI makes it easier for substantive contributions to be produced rapidly and invisibly:
flowchart LR A["AI-assisted contribution"] B["Research output"] C["Claim"] C1["method"] C2["interpretation"] D["Researcher accountability"] A --> B --> C --> D B --> C1 --> D B --> C2 --> D
This strains established norms around authorship, contribution, provenance, and accountability.
Most universities, including the University of Nottingham, explicitly treats unpermitted or unacknowledged AI assistance as false authorship, regardless of intent.
This includes AI contribution to:
It is your responsibility to disclose the use of AI in reasearch.
AI acceptability varies by:
The governance principle: permission must be explicit; ambiguity is risk.
The researcher must be able to:
AI can assist, but as a tool, cannot carry responsibility for the quality of research.
Traceability aligns with established research integrity practices.
| What must be transparent | Why it matters |
|---|---|
| Tool(s) and version(s) | provenance and reproducibility |
| Where AI influenced decisions | auditability of reasoning |
| What was verified (and how) | defensible claims |
| What was not verified | honest uncertainty |
A recurring problem is distinguishing language editing (clarity, structure) from substantive contribution (ideas, methods, claims, interpretation)
flowchart TB A[Surface editing] --> A1[grammar, clarity] B[Structural shaping] --> B1[outline, framing] C[Substantive contribution] --> C1[ideas, claims, methods] D[Credentialed output] --> D1[assessment/publication] style A stroke-width:3px style B stroke-width:3px style C stroke-width:3px style D stroke-width:3px
Research-relevant risks often arise from normal workflow choices:
Failing to disclose material AI contribution where required may seem minor, but can have significant ethical and governance implication.
Disclosure is not a punishment; it enables:
You make a claim? Was AI involved? If Yes, keep an audit of how its involved and disclose it!
The Following Method was produced using AI:
âData will be analysed using to identify informative variables. Variables meeting a predefined relevance threshold will be retained. Cases showing minimal contribution will be excluded and the the top 25 highest-ranked variables will be taken forward for interpretation and reporting.â
Many AI tools operate outside institutional infrastructure and governance controls.
This transforms routine academic actions (summarising notes, refining code, drafting interpretation) into potential boundary-crossing events that may bypass established safeguards.
flowchart TB B["Routine academic task"] C["External AI service"] D["Loss of control"] B --> C C --> D D --> E["Unclear retention"] D --> F["Jurisdictional uncertainty"] D --> G["Secondary use risk"] style C stroke-width:3px style D stroke-width:3px
The dominant risk is not malicious intent but loss of visibility and control once material leaves institutional systems.
In AI-assisted workflows, âdataâ includes any material that encodes research context, insight, or trajectory, not only formal datasets:
Much of this material is informally handled but formally sensitive.
| Category | Examples | Why high risk |
|---|---|---|
| Personal / sensitive | clinical, genetic, geolocation | legal and ethical obligations |
| Qualitative text | interviews, free text, notes | re-identification risk |
| Unpublished work | manuscripts, result tables | premature disclosure |
| IP / commercial | methods, pipelines, agreements | loss of competitive advantage |
| Security material | tokens, keys, internal URLs | direct compromise |
Risk frequently arises from aggregation, not single items.
Removing explicit identifiers does not eliminate disclosure risk.
Residual risk persists through rare attribute combinations, contextual cues in narrative text, small-n summaries, or linkage with external information.
The practical implication is operational rather than theoretical:
âAnonymisedâ does not mean âsafe to upload.â
A useful governance analogy is to treat external AI services as collaborators who have:
This framing clarifies why minimum necessary disclosure is a defensible default.
From a research governance perspective, unresolved questions include:
Uncertainty itself constitutes a risk factor when handling sensitive research material.
Risk is not only technical; it is behavioural.
Convenience can normalise incremental risk-taking without explicit decision points.
flowchart LR
A["Normal research task"] --> B["Convenient AI use"] --> C["Reduced deliberation"] --> D["Normalised exposure"] --> E["Elevated risk"]
style E stroke-width:3px
This form of âquiet driftâ is harder to detect than discrete policy violations.
A simple operational test:
Would I be permitted to email this material to an external collaborator without a formal agreement?
If not, it should not be shared with an external AI service. Check with your organisation directly if there are local / secure versions of these tools avalible for you to use.
AI supports research via literature search, coding, analysis and translation, but outputs may be biased, incorrect or insecure.
Key opportunities: efficiency, creativity, hypothesis generation and reduced repetitive workload.
Main risks: hallucinations, bias, plagiarism, copyright/IP issues, data breaches and loss of critical skills.
Ethical use requires transparency, accountability, fairness, privacy, and disclosureâAI is never an author.
Use secure, approved tools; avoid sensitive data sharing; follow institutional, funder and publisher policies.
