Tech Fixx · Video
Why Bad Data Breaks AI Models: Bias, Noise & Label Errors
Watch this 3 min 31 sec video from the Tech Fixx channel, published .
Check that the device, software and version shown match your setup. Use the player controls for captions, playback speed and full screen where available.
Loading the player connects to YouTube. If playback is unavailable here, watch on YouTube ↗.
Captions and text transcripts
After loading the player, use its CC control to turn on captions when available. To look for a text transcript, use the Watch on YouTube link above, expand the video description and select “Show transcript” if offered.
Caption and transcript availability varies by video, and automatic captions may contain mistakes. The original channel description is separate from a transcript. Caption accuracy, audio description and text alternatives have not been verified for every video in this library.
From the video description
Original channel description, recorded . Product, software and service references in it may have changed since publication.
An AI model can learn the wrong lesson when its training data is inaccurate, unrepresentative or poorly labelled. This explainer shows how data problems travel through the machine-learning pipeline and appear as unreliable outputs.
You’ll learn: • Incorrect and inconsistent labels • Missing, duplicated and noisy records • Sampling and representation bias • Data leakage and misleading shortcuts • Distribution shift between training and real use • Why overall accuracy can hide failures • Practical checks before deployment
Real-world performance should be monitored after launch because data and user behaviour can change.
ICO material on AI accuracy and fairness: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/guidance-on-ai-and-data-protection/
#DataQuality #AIBias #MachineLearning