Page 410 - Ai_V3.0_c11_flipbook
P. 410
AI Bias in Real Life
AI bias in real life means that the decisions made by AI systems are not always fair or accurate because they are
influenced by biases. Just like people, AI systems can make unfair decisions if they are trained on biased data or
programmed with biased rules.
Brainy Fact
Tay was an Artificial Intelligence Chatbot that was originally launched by Microsoft
Corporation on Twitter on March 23, 2016. When the bot started posting inflammatory
and offensive tweets through its Twitter account, it caused a subsequent controversy,
prompting Microsoft to shut down the service just 16 hours after its launch. According
to Microsoft, this was caused by trolls that "attacked" the service, as the bot's responses
were based on its interaction with people on Twitter.
Healthcare
AI model utilised for medical diagnosis can be biased if there is an over-representation of one
demographic group in the training data compared to another. For example, an AI system for melanoma
detection (a skin disease) trained primarily on fair skinned individuals may not perform well for darker
skinned ones, thus leading to misdiagnosis and delayed treatment.
Education
Drafting essays or grading standardised tests with AI can discriminate against test takers based on
culture, dialect, or the way of writing. Failure to recognise cultural expressions and languages may
unfairly disadvantage students from specific regions.
Finance
Biased training data or attributes may inadvertently lead some AI algorithms used in credit scoring to
discriminate against certain demographics. For instance, if historical loan approving data has prejudice
lending practices patterns, then AIs trained on such models might initiate them hence making it
difficult for lower-level communities to have equal access to loans.
Criminal Justice
If AI tools, trained on a system that had biases during the sentencing of inmates before their trial or
parole process, then subsequently it could also be predictive of bias. For instance, because past cases
disproportionately target people from certain races or socio-economic classes, this might result into
categorising many people from these groups as high risks thereby leading to unfairness in sentencing
procedures.
Task #Experiential Learning
Visit the website PortraitAI art generator (https://ai-art.tokyo/en/). Users upload their selfies
here, and the Artificial Intelligence uses its understanding of Baroque and Renaissance portraits
to draw their portrait in the style of a master. If a person is white, the result is very good. The
drawback is that the most popular paintings of this era were of European whites. So, the database consists
of mainly white people and an algorithm that relies on the same database tends to make you look ‘fairer’
while drawing your image.
408 Touchpad Artificial Intelligence (Ver. 3.0)-XI

