What is Personal Data?
Does your personal data include images?
– Bias is built into AI, which is based on the description of people and places that were used when it was developed.
– AI has not been trained on many resources from outside the white ‘developed’ world. This means that representations of cultures are going to be problematic, particularly if the AI is asked to generate images.
– AI is prone to generate hallucinations – these are basically things that the AI has invented, because it doesn’t really understand all the information it has been using to generate an output.
There are three main types of hallucination.
– Incorrect predictions: AI may predict an event that is unlikely to happen. e.g. weather predictions will suggest conditions that wouldn’t happen
– False positives: AI may identify something as being a threat when it is not. e.g. AI identifying a person as a potential shoplifter who is not
– False negatives: AI may fail to identify something as being a threat when it is. e.g. if used for medical purposes it may fail to diagnose a disease it is being asked to check for
Activity: Play the video on the Council of Europe Guidleines on facial recognition
If there is no filter on AI inputs, there will be toxicity in the output if the input is toxic. Toxic AI refers to AI systems that exhibit harmful or undesirable behaviour
Activity: Consider the possible issues with relying on facial recognition and AI to identify potential criminals in urban spaces?