S1E3 Limitations of AI

What this lesson is about

AI models can write answers that sound clear, detailed, and sure of themselves, but they do not check facts the way a careful researcher does. They predict the most likely next words from patterns they learned in training, so they sometimes produce claims that sound believable and are simply wrong, or even entirely made up. In this lesson you will learn to tell the difference between fluent language and accurate information, and you will practise catching mistakes that are dressed up to look correct.

What you'll be able to do

• I can analyse AI-generated information by identifying unreliable claims.
• I can recognise bias and apply critical understanding of how digital technologies generate content.
• I can evaluate the difference between fluent language and accurate information.

A little background

You will meet real examples of AI getting things wrong, from a chatbot that promised a customer a brand-new car for one pound, to AI inventing court evidence that landed real lawyers in trouble. Some experts prefer the word confabulation to hallucination, because AI has no senses to misread, it simply fills gaps with plausible-sounding detail. To use AI responsibly, you have to question its answers and check them against trusted sources.