Episode 41: How does AI work?
- Embedded IT
- Jun 2, 2025
- 2 min read
Updated: Jan 16
Artificial intelligence can feel complicated, but at its core it works in a surprisingly simple way. This episode breaks down how AI systems operate by using an everyday example to show how pattern-spotting drives the answers they produce.
This builds on our earlier overview of what AI is, including how different types of AI fit together.
How AI relies on data
AI is often described as intelligent, but it is not thinking for itself. It works by analysing huge amounts of data and recognising patterns. The more data it has, the better those patterns become. It does not reason, form opinions, or understand context. It predicts.
This is why data quality and quantity matter so much. AI is only as good as the information it has been trained on.
A simple example of pattern-spotting
To show how this works, a quick demonstration is used. Someone is asked to answer a series of simple one-word questions. When asked what M-O-S-T spells, the response is 'most'. Again, what do the letters C-O-A-S-T spell? The answer 'coast'. Their brain begins to link the pattern. Next, what do the letters B-O-A-S-T spell? The answer is 'boast'.
Then, when asked what a hostess does, the expected continuation of the pattern would be “hosts”.
But when asked what goes in a toaster, the automatic response is “toast”, even though the correct answer is bread. This moment is used to show how hallucinations occur in AI. When a system is too fixated on the pattern, it predicts an answer that sounds right rather than being factually right.
Why AI sometimes gets things wrong
Large language models like ChatGPT work in the same way. They are trained on vast amounts of internet content and learn the relationships between words. When asked a question, they generate the most likely next word based on all the patterns they have seen.
This also explains why AI tools can produce biased or incorrect outputs. If the training data contains biased or low-quality information, the model reflects it. AI becomes, as some say, “the average of the internet”, which is not always a reliable standard.
Understanding AI in a procurement context
For procurement teams assessing AI tools, this simple explanation matters. If the data is poor or the assumptions are flawed, the output will be unreliable. AI can be powerful, but it must be understood for what it is: pattern recognition on a massive scale.
For organisations exploring AI tools and needing clarity around data quality, risk, and procurement, get in touch.
