5.3 Big Idea Computing Bias Student Lesson • 17 min read

Big Idea 5.3 Computing Bias- Tanay

Overview/Definition: Computing Biasses are the numerous Biasses in application that are based on human prefrences.

• Computing innovations can reflect existing human biases because of biases written into the algorithms or biases in the data used by the innovation
• Programmers should take action to reduce bias in algorithms used for computing innovations as a way of combating existing human biases
• Biases can be embedded at all levels of software development

Types of Computing Bias - Tarun

• Data Bias: The data does not accurently represent the values of the real world ex. If data is taken from a sample size that doesn’t reflect the actual population - If you wanted data to represent the population in America but your sample that is being surveyed is from a Texas. The population in Texas does not accurately reflect the entire population of America.
• Human Bias: Those who make programs may be influenced by their own biases ex. If a development team are experts in using a certain language and their algorithm demonstrates that language, they will feel that people who specialize in that language are qualified and better. This is essentially bringing in their personal biases and applying to a larger amount of people.

Explicit data vs Implicit data: - Pranavi

Explicit data:

• takes the data that you give
• When watching a video, and it asks “are you enjoying this?”, and you respond with either a thumbs up or down, you are giving them explicit data

Implicit data:

• When you watch or search up certain things, data can be deduced on what is the “norm” for the person

Example: Netflix

• When browsing through Netflix, they show Netflix exclusives, they do this because they want your subscriptions
• showing the netflix exclusives is the bias in this scenario

Popcorn Hack:

In what other applications could have intential bias? Applications such as news stories likely intentionally display sources that are biased towards a view, others like shooting games probably bias towards a certain demographic like Battlefield appealing to teenage boys and men, and more.

Intentional Bias vs Unintentional Bias - Tanvi

Example 1: Hypothetical Loan company

• Suppose a software was created to assist loan officers, and certain trends of successful loans were taken
• If people are rejected of those who don’t fit in their trends of either age, gender, race, etc.
• This software is biased in the way that it only chooses candidates who will have higher chances in successful loans

Example 2: Candy Crush vs Call of Duty

• Call of Duty is geared towards the teenage boy demographic, 18-24, with more grunge type of music
• Candy Crush is more visually appearing to younger audience as it includes pictures of candies and playful music
• This is biased as the games include aspects and characteristics that will seem appealing to a specific audience

Popcorn Hack:

How is there unintentional bias in apps such as TikTok or Instagram or other social media apps? There’s unintentional bias in apps when the algorithm or recommendations use previous searches to curate content to display, biasing them towards a certain topic or viewpoint.

Mitigation Strategies - Shubhay

• Utilize data from various sources
• Pre-Processing: A way to check the inputs for bias before it is being used as data
• In-processing: This algorithm changes the data during analysis of the data to keep the data consistent
• Post-Processing: # step check to make sure the model is fair and accurate
  • Input Correction: This strategy makes adjustments to the data to make the data more comparable
  • Classifier Correction: polishing and adjusting the algorithm after it has been trained to reduce the biases
  • Output Correction: The predictions made by the model is modified to eliminate biases

Homework:

1. Is bias enhancing or intentionally excluding? This question can go either way depending on the intent of the program and where the bias stems from. For example, the bias can be completely unintential if the development team of a program did not predict the bias or had an overall beneficial goal but somehow it received biased information, such as with Microsoft’s attempt at an AI that failed. Furthermore, there will always be a bit of bias in any program since the programmer’s values and intents will be subconsciously implemented into the program. However, a program can also be intentionally biased (which it shouldn’t be) if the programmer wishes it to be, purposely overrepresenting a group of people and underrepresenting another. An example could be training an AI purely on a news show that is notoriously right-winged, resulting in the AI being biased towards conservative policies.
2. Is bias intentionally harmful/hateful? Again, this depends on the intention of the programmer and purpose of the project. Usually, the answer is no since programs generally want to accurately represent the whole population and not overrepresent or underrepresent certain groups. Programmers do their best to make their program not biased by feeding it a variety of data, gathering various opinions and critiques, and limiting their personal beliefs present in the program. In such cases, bias is not intentionally harmful. However, if the programmer is trying to commit malicious acts, then they could intentionally make the program biased toward certain opinions.
3. During software development are your receiving feedback from a wide variety of people? Yes, during development of our project, we are and will try to get feedback from many people. We have all as a group given our individual opinions, asked Mr. Lopez for feedback, talked with other groups, and collaborated with others in different periods of CSP.
4. What are the different biases you can find in an application such as Youtube Kids? Applications such as YouTube kids have lots of bias present within it. Some examples include intentional bias, as they (understandably) expect their demographic to be young children, thus curating the videos allowed to be tailored to young children and be not harmful or hate inducing and presenting their site to be more youth friendly. Two other forms of bias present are explicit and implicit bias, as YouTube asks for feedback on the video at times (which is explicit bias) and implicit bias as it uses the searches to provide recommendations that are similar to the searches.

Answer in complete sentences, due Sunday 11:59 pm