This is a guest post by Dr Stephen Anning, Visiting Researcher in the Department of Web Science at the University of Southampton and online tutor for the MA in Artificial Intelligence.
This blog post shares insights from student discussions in the ‘Introduction to AI’ module on the University of Southampton’s online MA in Artificial Intelligence, exploring the key technologies that underpin how we artificially generate intelligence.
The course is designed to open up the world of artificial intelligence to non-STEM students. If you have never touched code or have no great love for maths but want to understand how AI works and how it is shaping society, this course is for you.
Deterministic vs probabilistic artificial intelligence: understanding the two approaches to AI
In a previous webinar, students made the distinction between artificial intelligence - the philosophical and sci-fi exploration of whether machines can think - and artificially generating intelligence - the use of machines to process data and create knowledge.
This webinar built on the latter by exploring the four families of technology that underpin how we artificially generate intelligence.
Rule-based AI systems: how logic, decision trees, and early chatbots like ELIZA work
We began our journey in the 1960s, the origin of artificially generating intelligence. This family is built on "if-then-else" statements, a deterministic world of logical flows and decision trees. We discussed ELIZA, the 1966 therapy bot that mimicked a therapist by using pre-defined questions from the DOCTOR's script.
What’s fascinating is that even though users knew ELIZA was a machine, they developed deep emotional attachments because they felt “heard" for the first time. Eliza - a rudimentary technology from the 1960s based on logic, rules and concepts - actually predicts many of the risks we see today with the most advanced technologies.
While logic, rules and concepts might seem historic, we agreed that this family of technologies are very much alive. You see it in enterprise-level chatbots, medical triaging systems, and expert systems that capture specialist knowledge.
We also touched on "knowledge graphs" and "ontologies", which are about using prior knowledge to infer new relationships, like knowing that if two people work for the same organisation, they are likely colleagues. Knowledge graphs may not be as fashionable as machine learning models, but they remain a robust, human-centric way to structure information. Many examples of knowledge graphs exist, such as the Google Graph, DBPedia and the GDELT project.
What is machine learning? Understanding how AI models learn from data
We defined Machine Learning as the mathematical modelling of data to make predictions. Unlike the rigid rules of logic, this is a probabilistic family of technologies. As we discussed, the machine identifies patterns in data on a scale humans simply cannot match, or often even fathom. In supervised learning, a model learns from training data, which we have modelled using inherent features and annotations.
For example, predicting a car's price based on its engine size (feature 1) and manufacturer (feature 2). Machine learning is the basis for most other AI technologies (depending on how you define AI - still contested by many!)
A key takeaway from our session was the "black box" nature of machine learning models. Because the underlying algorithms derive their predictions from pattern identification, we sometimes lose the ability to see the reasoning behind their outputs.
We also discussed the risks of feedback loops; if a government recruitment AI assistant only looks for specific keywords, it might filter out "disruptors" or diverse thinkers, leading to an organisation where everyone thinks identically.
Natural language processing and large language models: how machines model human language
As a subset of machine learning, NLP is the mathematical modelling of language to make predictions. Vendors often claim NLP gives computers the ability "communicate" with and "understand" humans. We talked about Large Language Models (LLMs) and how they predict the next word in a sentence based on a mathematical model of all the previous words.
We had a profound discussion about the "Chinese Room" thought experiment. Does a machine truly understand Serbo-Croatian or Arabic, or is it just following a sophisticated script to produce an output?
One student made a brilliant point: perhaps the "understanding" doesn't happen in the machine at all, but within us, as we interpret the machine's output. This distinction between human and machine understanding highlights the ethical importance of human-centricity.
We must be careful not to attribute human emotions or intent to a system that is ultimately just calculating probabilities.
What is Media AI? How artificial intelligence analyses images, video, and audio
Finally, we looked at Media AI, which is the mathematical modelling of images, video, and audio to make predictions. This family of technologies uses principles similar to those of other machine learning families: mathematically modelling digitised data to deconstruct images or audio to make predictions from the underlying patterns. Whether the machine is predicting if an image contains a dog or a cat, or even inferring a car's value just by looking at a photo, it’s all about pattern recognition.
These technologies are tools we can use in professional and social contexts. The real "intelligence" lies in how we manage and critically evaluate the information they provide. Too many people put the cart before the horse: "AI's the answer, what's the question?" As we discussed, Eliza and decision trees in chatbots can achieve the same outcomes as advanced technologies.
Find out more about the MA in Artificial Intelligence
Artificial intelligence includes a range of technologies, from rule based systems and machine learning to natural language processing and Media AI. Understanding how these systems work, and how they shape decisions in society, is increasingly important across many sectors. The University of Southampton’s online MA in Artificial Intelligence is a conversion course designed for professionals who want to understand and apply AI responsibly, with no prior coding experience required.
Explore the course
