Page 77 - CT_AI_Class-7
P. 77
Algorithmic bias: Algorithmic bias happens when the design of the AI system itself causes Facial Recognition Disparities
unfair decisions. This can happen if the AI system focuses too much on one factor and ignores Researchers found that some facial recognition systems are better at recognising the faces of
others. For example, in a video game, an AI system might reward players who make quick light-skinned people than those with darker skin. In one study, the system made mistakes less
decisions, but not those who take their time to think carefully. This would be unfair to players than 1% of the time when recognising light-skinned men, but it made mistakes 30-35% of the time
who play thoughtfully. Algorithmic bias happens when the system favours one style of play for darker-skinned women. This happened because the system was trained with data that didn’t
over another. include enough examples of people from all groups, especially those with darker skin.
Human bias: AI systems are created by people and if the people designing the AI have their It’s important to use diverse and balanced data when training AI. This ensures that the system
own biases, those biases can affect the AI. For example, if someone who loves rock music works fairly for everyone, not just for one group of people.
creates an AI that suggests songs, it might suggest rock music to everyone, ignoring other
types like pop, classical or jazz. This is human bias, where the personal preferences of the Hiring Algorithm Bias
creator influence the AI’s decisions. A company developed an AI system to help review job applications by learning from resumes
21 st
Century #Critical Thinking submitted over the past ten years. However, most of the resumes the system reviewed were
ai in action Skills
from men and over time, the AI started favouring patterns commonly seen in male resumes.
The AI Bias Game shows how bias can affect decision-making. In this game, you work for As a result, when reviewing resumes from women, the AI gave them lower ratings, making unfair
the World Sports Board and need to pick lucky fans to win free tickets to a game. You will recommendations.
receive test results from different people who claim to be sports fans. Visit the given link
https://ai-bias.sustainablelivinglab.org/ or scan the OR code to play the game): Once the company realised that the AI system was biased, they stopped using it and changed the
way it worked to ensure fairer decisions.
Healthcare Risk Prediction Algorithm
How Bias Affects AI
A healthcare system developed an AI algorithm to predict how sick a patient might become based
When AI systems are biased, they can make unfair decisions. This can cause harm, especially in on the amount of money they had spent on medical treatment in the past. However, the system
important areas like healthcare, education and jobs. Here are some examples of biased decisions overlooked the fact that spending less on healthcare doesn't always mean a person is healthier.
in AI:
Some patients spend less because they cannot afford treatment or lack access to healthcare
Biased decisions when giving loans: An AI system could unfairly deny someone a loan services. As a result, the system sometimes predicted a lower health risk for these patients, even
because it was trained on biased data. when their condition was just as serious as others who had higher medical costs.
Unfair hiring or recruitment decisions: If AI is used to decide who gets a job, it might be unfair Once the healthcare system realised this flaw, they made changes to the algorithm to consider
if it’s trained on data that favours certain groups of people over others.
other factors like access to care and medical history, ensuring a more accurate prediction.
Wrong predictions about crime risks: If AI is used to predict crime, it might unfairly target
certain communities based on biased data. Lessons Learned from AI Bias
Incorrect medical or health advice: If AI systems are used to give health advice, they might To make sure AI systems are used fairly and responsibly, it’s important to remember a few key
make wrong suggestions if the data is biased or incomplete. lessons:
Mistakes in identifying people: Facial recognition systems can make mistakes, especially if Data can reflect biases in society: The data used to train AI isn’t always neutral. If it shows
they haven’t been trained on diverse data, leading to wrong identifications. biased patterns from the past, the AI will likely repeat them.
AI systems should be checked regularly: To ensure AI is working fairly, it should be checked
CASE EXAMPLES OF AI BIAS
often for any bias.
Artificial intelligence (AI) is being used in many areas of lives, from hiring people to helping doctors. Balanced and varied data helps reduce bias: The data used to train AI should come from
But sometimes, AI systems can be unfair. This happens when the data used to train these systems different groups of people and situations. This ensures the AI makes fair decisions.
has hidden biases or when the system is not designed properly. These biases can cause the AI to
make unfair decisions, which can affect certain groups of people more than others. Let us explore AI needs regular monitoring: Even after the system is running, it must be consistently checked
few examples of AI bias. to ensure it continues to work fairly.
Ethics and AI Bias Awareness 75
