Ai Generative Models

Generative Modelling Theory

Doctoral work examines the mathematical foundations of learning to produce new samples. Theory establishes what any generative approach can and cannot achieve.

Density Estimation Foundations

Research examines learning the probability structure underlying observed data. Density estimation is the formal problem generative modelling addresses.

Latent Variable Model Research

Doctoral study examines models positing unobserved factors behind observed data. Latent structure permits controllable and interpretable generation.

Variational Autoencoder Research

Research examines models learning compressed representations and reconstructions. These models provide tractable likelihoods and structured latent spaces.

Variational Inference Methods

Doctoral work examines approximating intractable probability distributions. Variational methods make otherwise impossible learning problems tractable.

Evidence Lower Bound Analysis

Research examines the objective optimised in variational generative training. Bound structure determines the trade off between reconstruction and structure.

Posterior Collapse Research

Doctoral study examines latent variables becoming uninformative during training. Collapse defeats the purpose of learning a latent representation entirely.

Generative Adversarial Network Research

Research examines generation trained against a competing discriminating model. Adversarial training produces sharp samples but trains unstably.

Adversarial Training Dynamics

Doctoral work examines the unstable equilibrium between competing networks. Dynamics understanding explains why such training frequently fails.

Mode Collapse Analysis

Research examines models producing only a narrow subset of possible outputs. Collapse produces convincing samples with catastrophically poor diversity.

Discriminator Design Research

Doctoral study examines the network judging whether samples appear genuine. Discriminator capacity governs the quality of the resulting generator.

Normalising Flow Research

Research examines models built from exactly invertible transformations. Invertibility permits exact likelihood computation that other families lack.

Invertible Architecture Design

Doctoral work designs network structures that can be exactly reversed. Invertibility constrains architecture while enabling exact inference.

Continuous Normalising Flows

Research examines flows defined through continuous time transformations. Continuous formulation connects flows with differential equation solvers.

Autoregressive Generative Models

Doctoral study examines generation producing one element at a time. Sequential generation gives exact likelihoods but slow sample production.

Sequence Generation Theory

Research examines theoretical properties of models generating ordered outputs. Sequential structure underlies language, audio and much scientific data.

Energy Based Model Research

Doctoral work examines models assigning scores rather than explicit probabilities. Energy formulations are flexible but demand costly sampling procedures.

Score Based Generative Modelling

Research examines learning the gradient of the data probability landscape. Score learning avoids estimating intractable normalising quantities entirely.

Diffusion Model Theory

Doctoral study examines generation by progressively reversing a noising process. Diffusion approaches now dominate high quality visual generation.

Denoising Process Research

Research examines the stepwise removal of noise producing a final sample. Denoising design determines both sample quality and generation speed.

Noise Schedule Design

Doctoral work examines how noise levels progress across generation steps. Schedule choice substantially affects both quality and required computation.

Sampling Method Research

Research examines procedures producing samples from a trained model. Sampling choices affect quality, diversity and computational cost together.

Accelerated Sampling Methods

Doctoral study reduces the number of steps required to produce a sample. Step reduction is the principal route to practical generation speed.

Consistency Model Research

Research examines models producing samples in very few generation steps. Few step generation makes interactive applications genuinely feasible.

Flow Matching Methods

Doctoral work examines training by matching continuous transport paths between distributions. Flow matching simplifies training relative to earlier formulations.

Stochastic Differential Equation Models

Research formulates generation through continuous stochastic processes. This formulation unifies several apparently distinct model families.

Latent Diffusion Research

Doctoral study examines diffusion performed within a compressed representation. Working in compressed space reduces computation by a large factor.

Conditional Generation Methods

Research examines generation directed by supplied conditioning information. Conditioning is what makes generative models practically controllable.

Classifier Guidance Research

Doctoral work steers generation using a separate classifying model as a guide. Guidance strengthens adherence to conditions at some cost in output diversity.

Guidance Without Classifiers

Research achieves conditional steering without any auxiliary classifier. This approach became standard across text conditioned image generation.

Discrete Diffusion Research

Doctoral study adapts diffusion to discrete rather than continuous data. Discrete formulations extend these methods to text and molecular structures.

Masked Generative Modelling

Research examines generation by progressively filling concealed positions. Masked generation permits parallel rather than sequential production.

Hybrid Model Architectures

Doctoral work combines several generative approaches within a single system. Hybrid designs seek to capture the strengths of each family simultaneously.

Model Capacity And Expressivity

Research examines what distributions a given architecture can represent. Expressivity limits determine which data a model could ever capture.

Likelihood And Quality Trade Offs

Doctoral study examines tension between statistical fit and perceived quality. Models scoring well statistically frequently produce poor samples.

Mode Coverage Analysis

Research examines whether models represent the full variety of training data. Coverage failures silently exclude entire categories from generation.

Memorisation In Generative Models

Doctoral work examines models reproducing their training examples very closely. Memorisation creates both privacy and intellectual property exposure risk.

Generalisation Theory

Research examines why generative models produce genuinely novel outputs. Distinguishing novelty from recombination remains theoretically difficult.

Manifold Learning Perspectives

Doctoral study examines data occupying low dimensional structure within high dimensions. Manifold structure explains why generation is possible at all.

Representation Disentanglement

Research examines separating independent factors within learned representations. Disentangled factors permit independent control over generated attributes.

Controllable Generation Research

Doctoral work examines precise user control over generated output properties. Control determines whether these models serve as practical creative tools.

Latent Space Manipulation

Research examines editing generated outputs by adjusting internal representations. Latent editing offers meaningful control without any retraining of the model.

Interpolation And Traversal Research

Doctoral study examines smooth transitions between generated outputs. Smooth traversal indicates that meaningful structure was actually learned.

Inversion Method Research

Research recovers the internal representation corresponding to a given example. Inversion permits editing of real rather than only generated content.

Conditioning Signal Design

Doctoral work examines how instructions and references are supplied to models. Signal design determines the precision of achievable user control.

Compositional Generation Research

Research examines combining multiple specified elements within one output. Compositional failures are among the most visible current limitations.

Out Of Distribution Generation

Doctoral study examines producing outputs unlike anything in training data. Genuine novelty is what distinguishes creation from recombination.

Extrapolation Capability Research

Research examines whether models extend beyond their training distribution. Extrapolation ability determines usefulness for genuine scientific discovery.

Scaling Behaviour Of Generative Models

Doctoral work examines how generation quality changes with model and data scale. Scaling relationships guide resource allocation before training begins.

Training Efficiency Research

Research reduces the computation required to train generative models. Training efficiency determines which organisations can build models at all.

Image Generation Research

Doctoral study examines synthesis of realistic and controllable images. Image generation is the most publicly visible generative application.

High Resolution Image Synthesis

Research examines generating images at very large pixel dimensions. Resolution demands grow computation faster than perceived quality improves.

Text To Image Generation

Doctoral work examines producing images directly from written descriptions. Text conditioning made these systems accessible to non specialist users.

Image Editing Methods

Research examines modifying existing images through generative techniques. Editing is more commercially significant than generation from nothing.

Inpainting And Completion

Doctoral study examines filling missing or removed regions within an image. Completion requires consistency with the surrounding image content.

Style Transfer Research

Research examines rendering content in the visual manner of a reference work. Style transfer raises immediate questions about creator consent and rights.

Image Restoration Generation

Doctoral work examines recovering degraded images using learned generative priors. Generative restoration may invent detail that was never actually present.

Super Resolution Methods

Research examines increasing image detail beyond the captured resolution. Added detail is plausible rather than genuinely recovered information.

Sketch And Layout Conditioning

Doctoral study examines guiding generation using spatial arrangement inputs. Spatial conditioning gives control that text alone cannot provide.

Personalised Image Generation

Research examines adapting models to specific subjects or particular visual styles. Personalisation raises distinctive consent and personal likeness questions.

Subject Consistency Research

Doctoral work examines maintaining a consistent subject across many outputs. Consistency is essential for narrative and commercial applications.

Video Generation Research

Research examines synthesis of moving image sequences from learned models. Video generation demands vastly greater computation than still image generation.

Temporal Consistency Methods

Doctoral study examines maintaining coherence across generated video frames. Temporal inconsistency is the most visible video generation failure.

Long Video Synthesis

Research examines generating extended rather than very brief sequences. Duration remains severely constrained by memory and consistency limits.

Video Editing Research

Doctoral work examines modifying existing video through generative methods. Video editing raises particularly acute authenticity and evidential concerns.

Motion Representation Research

Research examines how movement is encoded within generative video models. Motion representation determines the realism of the dynamics produced.

World Model Research

Doctoral study examines models simulating environments and their dynamics. World models connect generation with planning and control research.

Physics Consistency In Generation

Research examines whether generated content obeys basic physical plausibility. Physical violations are immediately obvious to any attentive human viewer.

Audio Generation Research

Doctoral work examines synthesis of realistic and controllable audio output. Audio generation spans speech, music and environmental sound together.

Speech Synthesis Methods

Research examines generating natural sounding spoken language from text. Speech synthesis now approaches indistinguishability from recorded human speech.

Voice Conversion Research

Doctoral study examines rendering speech in the voice of another person. This capability creates serious impersonation and consent concerns.

Music Generation Research

Research examines generation of musical audio and complete compositions. Music generation raises acute questions about creative labour and rights.

Sound Effect Synthesis

Doctoral work examines generating environmental and deliberately designed sound. Synthesised sound serves media production and simulation applications.

Multimodal Generation Research

Research examines generating several coordinated modalities simultaneously. Joint generation is considerably harder than generating each modality separately.

Cross Modal Translation

Doctoral study examines converting content between differing modalities. Translation reveals what structure is genuinely shared across modalities.

Audio Visual Synchronisation

Research examines aligning generated sound with generated moving images. Synchronisation failures are immediately perceptible to any viewer.

Three Dimensional Shape Generation

Doctoral work examines generating three dimensional object geometry. Spatial generation supports design, simulation and virtual environment creation.

Scene Generation Research

Research examines generating complete spatial environments and layouts. Scene generation requires consistency across many contained objects.

Neural Rendering Methods

Doctoral study examines producing images from learned scene representations. Neural rendering blurs the boundary between capture and generation.

Radiance Field Research

Research examines volumetric scene representations reconstructed from images. These representations permit viewing scenes from unrecorded positions.

Avatar And Character Generation

Doctoral work examines generating human and character representations. Character generation raises immediate likeness and consent questions.

Texture And Material Generation

Research examines generating surface appearance for three dimensional models. Material generation is a substantial bottleneck in content production pipelines.

Animation Generation Research

Doctoral study examines generating movement for characters and animated objects. Animation is among the most labour intensive of all production tasks.

Motion Synthesis Research

Research examines generating realistic human and animal movement sequences. Movement realism requires both physical and behavioural plausibility together.

Text Generation Research

Doctoral work examines generation of written language across contexts. Text generation underpins the most widely deployed generative applications.

Controlled Text Generation

Research examines directing textual output toward specified target properties. Control determines suitability for professional and regulated applications.

Style And Register Control

Doctoral study examines controlling tone and formality in generated text. Register control matters greatly for professional communication uses.

Long Form Generation Research

Research examines producing extended and internally coherent written works. Coherence across long outputs remains a substantial unsolved difficulty.

Structured Output Generation

Doctoral work examines producing output conforming to formal structural requirements. Structured generation makes model output reliably machine interpretable.

Code Generation Research

Research examines generating functioning computer programs from descriptions. Code generation is among the most economically significant applications.

Program Synthesis Methods

Doctoral study examines constructing programs satisfying formal specifications. Synthesis offers correctness guarantees that generation alone cannot.

Table And Data Generation

Research examines generating realistic structured tabular data records. Tabular generation supports testing wherever real data cannot be shared.

Graph Generation Research

Doctoral work examines generating networks and relational structures. Graph generation supports molecular, social and infrastructure modelling.

Molecular Generation Research

Research examines designing novel molecules with specified properties. Generative design searches spaces far larger than intuition can cover.

Protein Design Generation

Doctoral study examines generating protein sequences and their structures. Protein design has already produced functional molecules unknown in nature.

Materials Generation Research

Research examines generating candidate materials with target properties. Materials discovery is slow and generative search could accelerate it.

Chemical Structure Generation

Doctoral work examines generating chemically valid molecular structures. Validity constraints make chemical generation distinctively difficult.

Therapeutic Candidate Generation

Research examines generating molecules as candidate medicines. Generated candidates still require the full experimental development pathway.

Sequence Design In Biology

Doctoral study examines generating biological sequences for specified functions. Sequence design raises important biosecurity governance questions.

Scientific Hypothesis Generation

Research examines models proposing testable scientific explanations and hypotheses. Hypothesis generation is the least automated stage of scientific work.

Simulation Data Generation

Doctoral work examines generating data to supplement physical simulation. Generated data can accelerate simulations that are otherwise prohibitive.

Time Series Generation

Research examines generating realistic temporal data sequences. Temporal generation supports testing, forecasting and privacy protection alike.

Sensor Data Synthesis

Doctoral study examines generating realistic measurement data streams. Synthetic sensor data supports development where collection is costly.

Medical Image Generation

Research examines generating clinical imaging for research and training use. Generated imaging must never be mistaken for genuine patient records.

Synthetic Patient Data Research

Doctoral work examines generating realistic health records for research. Synthetic records permit sharing where real records cannot be released.

Geospatial Data Generation

Research examines generating realistic maps and geographic imagery. Geospatial generation supports both planning and simulation applications.

Speech Corpus Synthesis

Doctoral study examines generating speech data for training other systems. Synthetic speech addresses scarcity for underrepresented languages.

Handwriting Generation Research

Research examines generating realistic handwritten text and script. Handwriting generation supports recognition training and raises forgery concerns.

Font And Typography Generation

Doctoral work examines generating typefaces and letterform designs. Typeface generation extends design tools to underserved writing systems.

Layout And Document Generation

Research examines generating structured document and interface layouts. Layout generation must satisfy both aesthetic and functional constraints.

Generative Model Evaluation

Doctoral study develops sound assessment of generative model quality. Evaluation is unusually difficult because there is no single correct output.

Sample Quality Metrics

Research examines automated measures of generated sample quality. Widely used measures correlate only imperfectly with actual human judgement.

Diversity Measurement Methods

Doctoral work examines measuring variety across many generated outputs. Diversity measurement detects mode collapse that quality measures conceal.

Fidelity Assessment Research

Research examines how closely generated samples resemble real data. Fidelity and diversity trade against one another and must both be measured.

Human Evaluation Of Generation

Doctoral study examines assessment of generated output by human judges. Human judgement remains the reference standard yet is costly and variable.

Preference Based Evaluation

Research examines comparison based assessment between candidate outputs. Comparison judgements are more reliable than absolute quality ratings.

Benchmark Design For Generation

Doctoral work examines constructing meaningful generative model benchmarks. Poorly designed benchmarks misdirect an entire research community.

Evaluation Metric Validity

Research examines whether measures capture what they claim to capture. Several established generative measures have documented serious flaws.

Prompt Following Assessment

Doctoral study examines whether output actually matches the given instruction. Instruction adherence is frequently weaker than sample quality suggests.

Instruction Adherence Research

Research examines faithfulness to detailed and multi part user instructions. Adherence degrades sharply as instruction complexity continues increasing.

Compositional Evaluation

Doctoral work examines assessment of combined attributes within outputs. Compositional assessment reveals failures single attribute testing misses.

Factuality In Generated Content

Research examines whether generated statements correspond to reality. Factual reliability determines suitability for informational applications.

Consistency Evaluation Methods

Doctoral study examines internal coherence within and across outputs. Inconsistency undermines usefulness even where individual outputs seem good.

Robustness Of Generation

Research examines output stability under small variations in the given input. Minor phrasing changes can produce very different generated results.

Failure Mode Analysis

Doctoral work catalogues the characteristic ways generative models fail. Failure catalogues guide both model improvement and appropriate deployment.

Artefact Detection Research

Research examines automated detection of generation artefacts in output. Artefact detection supports both quality control and provenance assessment.

Preference Optimisation For Generation

Doctoral study examines aligning generation with expressed human preferences. Preference training substantially shapes what these systems actually produce.

Reward Modelling For Generation

Research examines learned models scoring the quality of generated output. Reward model flaws propagate directly into the resulting generated behaviour.

Feedback Learning In Generation

Doctoral work examines improving generation from collected human feedback. Feedback quality determines whether the resulting behaviour improves.

Fine Tuning For Generation

Research examines adapting pretrained generative models to specific uses. Adaptation converts broad general capability into practical usefulness.

Adapter Methods For Generation

Doctoral study examines lightweight modules adapting generative model behaviour. Adapters are cheap to train, to store and to combine with one another.

Personalisation Techniques

Research examines adapting models to individual users or particular subjects. Personalisation raises distinctive privacy and consent considerations.

Model Merging In Generation

Doctoral work examines combining separately adapted generative models. Merging captures multiple capabilities without any additional training.

Distillation Of Generative Models

Research transfers generative capability into smaller and much faster models. Distillation makes high quality generation practically deployable at scale.

Compression For Generation

Doctoral study reduces generative model size while preserving output quality. Compression determines what hardware can realistically host a given model.

Quantisation Of Generative Models

Research reduces numerical precision within deployed generative models. Generation quality is unusually sensitive to reduction in numerical precision.

Efficient Inference For Generation

Doctoral work reduces the computation required to produce each generated output. Generation is far more costly per output than classification tasks.

Real Time Generation Systems

Research examines generation fast enough to support interactive applications. Interactivity fundamentally changes how these creative tools are used.

On Device Generation Research

Doctoral study examines generation performed entirely on personal hardware. Local generation removes both network dependence and any data exposure.

Serving Infrastructure Research

Research examines infrastructure delivering generation at very large scale. Serving design determines both operating cost and achievable response speed.

Batching And Scheduling Methods

Doctoral work examines grouping requests to use hardware more efficiently. Scheduling determines the throughput achievable from fixed hardware resources.

Memory Optimisation In Generation

Research reduces the memory required during the generation process itself. Memory is frequently the binding constraint on available deployment options.

Hardware Acceleration Research

Doctoral study examines specialised hardware designed for generative workloads. Generation workloads differ substantially from model training workloads.

Energy Cost Of Generation

Research quantifies the energy consumed in producing generated content. Per output energy cost becomes substantial at very large usage volumes.

Environmental Impact Assessment

Doctoral work examines the full environmental burden of these systems. Water use and hardware manufacture matter alongside operational energy.

Interactive Generation Interfaces

Research examines how people direct, steer and refine generative output. Interface design determines whether these tools are genuinely usable in practice.

Human In The Loop Generation

Doctoral study examines collaboration between people and generative systems. Collaborative use produces better results than fully automatic generation.

Creative Workflow Integration

Research examines fitting generative tools into established creative practice. Tools disrupting existing workflow are abandoned regardless of capability.

Iterative Refinement Interfaces

Doctoral work examines progressive improvement of generated output over turns. Iteration is how skilled users actually achieve their intended results.

Multi Agent Generation Systems

Research examines several models cooperating to produce a single output. Cooperation permits specialisation across separate stages of a generation task.

Retrieval Conditioned Generation

Doctoral study examines generation grounded in retrieved reference material. Grounding improves factual accuracy and permits source attribution.

Tool Assisted Generation

Research examines generation supported by external computational tools. Tool access supplies precision that generation alone cannot achieve.

Structured Constraint Enforcement

Doctoral work examines guaranteeing output satisfies specified constraints. Enforcement makes generation usable where correctness is required.

Verification Of Generated Output

Research examines checking generated content against stated requirements. Verification is what converts plausible output into genuinely dependable output.

Formal Guarantees In Generation

Doctoral study examines provable properties of generated output. Formal guarantees are essential for safety critical generative applications.

Training Data For Generative Models

Research examines the corpora underlying generative model capability. Data composition determines what a model can produce more than architecture.

Data Curation For Generation

Doctoral work examines assembling and filtering generative training material. Curation choices shape aesthetic and cultural properties of output.

Data Licensing And Rights

Research examines legal permissions governing generative training material. Rights questions remain substantially unsettled across jurisdictions.

Copyright And Generative Output

Doctoral study examines ownership and infringement questions for generated content. These questions are actively litigated and remain genuinely unresolved.

Style Imitation Research

Research examines models reproducing the distinctive manner of individual creators. Style imitation is legally ambiguous and ethically contested.

Creator Consent Research

Doctoral work examines whether creators agreed to their work being used. Most generative training proceeded without seeking any creator consent at all.

Compensation Mechanism Research

Research examines arrangements rewarding those whose work enabled a model. Workable compensation mechanisms remain substantially undeveloped.

Attribution Of Generated Output

Doctoral study examines identifying influences on a particular generated work. Attribution underpins both credit and any compensation arrangement.

Training Data Attribution Methods

Research traces generated output back to specific training examples. Attribution at generative model scale remains computationally very hard.

Memorisation And Extraction Research

Doctoral work examines recovering training examples from generative models. Extraction demonstrates that training data is not fully protected.

Privacy In Generative Models

Research examines exposure of personal information through generation. Generative models can reproduce personal material from their training data.

Differential Privacy For Generation

Doctoral study applies formal privacy guarantees to generative training. Formal guarantees substantially reduce generation quality at present.

Synthetic Data Privacy Analysis

Research examines whether synthetic data genuinely protects individuals. Synthetic data is frequently assumed private without any verification.

Reidentification Risk Research

Doctoral work examines whether individuals can be recovered from synthetic data. Reidentification risk determines whether release is genuinely safe.

Watermarking Generated Content

Research examines embedding detectable markers within generated output. Watermarking supports provenance without restricting legitimate use.

Watermark Robustness Research

Doctoral study examines whether markers survive editing, compression and processing. Fragile watermarks provide very little practical provenance assurance.

Generated Content Detection

Research examines distinguishing generated from human produced material. Reliable detection remains unsolved and may be fundamentally limited.

Detection Limitation Analysis

Doctoral work examines the theoretical limits of generated content detection. Understanding limits prevents reliance on detection that cannot work.

Content Provenance Standards

Research examines standards recording how digital content was produced. Provenance infrastructure may prove more workable than detection alone.

Media Authentication Research

Doctoral study examines verifying that media is genuine and unmodified. Authentication protects the evidential value of recorded audio and imagery.

Synthetic Media Detection Research

Research examines identifying fabricated depictions of real people. Detection protects individuals and public discourse from fabricated material.

Nonconsensual Imagery Prevention

Doctoral work examines preventing generation of intimate imagery without consent. This harm falls overwhelmingly on women and demands technical response.

Likeness And Voice Protection

Research examines protecting individuals from unauthorised synthetic depiction. Existing legal protections fit these particular harms very poorly indeed.

Model Unlearning For Generation

Doctoral study examines removing specific influences from a trained model. Removal is increasingly legally required yet technically difficult to verify.

Data Removal Verification

Research examines confirming that requested removal genuinely took effect. Verification matters because removal claims cannot easily be checked.

Safety Filtering In Generation

Doctoral work examines systems preventing harmful generative model output. Filtering must be effective without blocking legitimate creative work.

Harmful Content Prevention

Research examines preventing generation of seriously harmful material. Prevention must operate across modalities, languages and cultural contexts.

Child Safety In Generative Systems

Doctoral study examines protecting children from harms these systems can cause. This is the most serious safety obligation such systems carry.

Bias In Generated Content

Research examines systematic skews within what these models produce. Generated content reflects and can amplify biases present in training data.

Representation Analysis In Generation

Doctoral work examines who and what actually appears within generated output. Representation patterns shape public perceptions at very large scale.

Stereotype Amplification Research

Research examines models producing stereotypes more strongly than their data. Amplification makes generated content worse than the material it learned from.

Fairness Mitigation In Generation

Doctoral study examines reducing unjustified skews in generated output. Mitigation attempts frequently introduce new distortions of their own.

Cultural Representation Research

Research examines how differing cultures are depicted within generated content. Depiction quality varies enormously across differing cultural contexts.

Multilingual Generation Equity

Doctoral work examines generation quality across differing languages. Most languages receive markedly poorer generation than a privileged few.

Misinformation Risk Research

Research examines generative contributions to the spread of false information. Generation substantially reduces the cost of producing convincing falsehoods.

Persuasion And Influence Research

Doctoral study examines the persuasive capability of generated content. Personalised persuasion at scale raises serious and unresolved societal concerns.

Misuse Pattern Analysis

Research examines how generative systems are actually misused in practice. Observed misuse patterns should directly guide the design of safeguards.

Access And Release Policy

Doctoral work examines decisions about who may use generative models and how. Release decisions are highly consequential and largely irreversible.

Open Model Governance

Research examines governance of publicly released generative models. Open release enables scrutiny while removing usage safeguards entirely.

Regulatory Framework Analysis

Doctoral study examines emerging regulation of generative systems. Regulatory approaches differ sharply between jurisdictions and remain unsettled.

Standards For Generative Systems

Research examines technical standards for these systems and their output. Standards translate broad principles into requirements that can be checked.

Labour Market Impact Research

Doctoral work examines how generative systems change the organisation of work. Effects fall very unevenly across differing occupations and skill groups.

Creative Industry Impact Research

Research examines consequences for those working in creative occupations. Creative workers face both displacement and unconsented use of their work.

Public Attitudes To Generative Systems

Doctoral study examines how the public views these systems and their output. Public acceptance shapes what regulation will ultimately permit.

Implementation And Adoption Research

Research examines why organisations succeed or fail deploying these systems. Adoption barriers are frequently organisational rather than technical.