Page 272 - Touhpad Ai

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.

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
BRAINY account, it caused a subsequent controversy, prompting Microsoft to shut
FACT 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.

AI TASK 21 st
Century #Experiential Learning
Visit the website PortraitAI art generator (AI Gahaku) at https://ai-art.tokyo/en/ Skills
or scan the QR code. 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.

270 Touchpad Artificial Intelligence - XI