AI, Biometrics and Smart Systems

Module Reflection  ·  MSc Computer Science (Conversion)  ·  Author: Orville Fernandes

What I Brought In

My interest in AI had been growing steadily since starting the programme. My part-time work as a Data Annotation Specialist evaluating LLM outputs gave me a practical, if narrow, window into how AI systems behave in practice. My Psychology background also meant I was already thinking about human factors, bias, and the gap between what a system is designed to do and how it actually affects people. I came into COM7119 curious and ready to go deeper.

The Module

The module covered the theoretical foundations of artificial intelligence, biometrics, and smart systems. Core topics included KNNs, machine learning approaches, LLMs, agents, and the application of AI to medicine. Biometrics received dedicated coverage, though only across two lectures.

The Coursework

The coursework asked us to design an AI-assistive system. I chose to design an AI-assisted retail loss prevention system that uses computer vision to detect suspicious behaviour in real time. I proposed an edge-cloud architecture to balance latency, bandwidth, and scalable analytics.

Working with surveillance, biometric identification, and behavioural profiling raises significant questions under GDPR around lawful basis, data minimisation, and algorithmic accountability which were discussed and evaluated in my report (LeCun, Bengio and Hinton, 2015; ICO, 2023). My position was that the system should support human decision-making rather than replace it, and that any deployment would need transparent governance and regular bias auditing.

The coursework planted the seed for my dissertation project, which will involve designing and implementing an AI-assisted shoplifting detection and evidence management system. See Dissertation Direction & Career Goals.

Critical Reflection

The module had some strengths but also a fundamental honesty problem. One of the stated learning outcomes was to demonstrate understanding of current AI-based security systems. The teaching did not come close to supporting this. Biometrics and AI security received two lectures between them, neither at the depth the outcome implied. The assignment brief compounded the issue by specifying an "AI-assistive system" with no mention of a security system, leaving a direct contradiction between what the brief asked for and what the learning outcome required. I found this confusion significant enough to raise a formal complaint about the module design. I have written about that experience separately.

Beyond that specific issue, the module followed the same pattern I noticed across the programme: strong conceptual framing, little practical implementation. There were two lab sessions, one exploring KNNs and another on text-to-speech, with neither going into any real depth, and between them they barely scratched the surface of what practical AI work looks like. For a Masters programme in this day and age, where AI tooling is accessible and widely used, that felt like a missed opportunity. Furthermore, the coursework asked for a system design rather than a working prototype, which meant the learning stayed largely theoretical, and a heavier focus on system design rather than AI.

Learning Outcomes

LO1 – Understanding of current AI-based security systems: Partially achieved, though not through the teaching.

LO2 – Systematic understanding of AI, Biometrics and Smart Systems: Achieved through the coursework. Designing the retail loss prevention system required engagement with CNNs, object detection, temporal analysis, and smart system architecture.

LO3 – Applying knowledge to select appropriate AI solutions: Achieved through the architectural and technical decisions in the coursework, including the selection and justification of AI models, and the edge-cloud processing model.

LO4 – Evaluating legal and ethical requirements: Achieved through the coursework. The system design required genuine engagement with GDPR, data minimisation principles, and the ethical risks of algorithmic bias and disproportionate surveillance.

References

Information Commissioner's Office (ICO) (2023) Guidance on AI and Data Protection. Available at: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/guidance-on-ai-and-data-protection/ (Accessed: 1 June 2025).

LeCun, Y., Bengio, Y. and Hinton, G. (2015) 'Deep learning', Nature, 521(7553), pp. 436–444. doi: 10.1038/nature14539.

Marks