Algorithms are ubiquitous in our lives, governing everything from Netflix movie choices to credit decisions to policing. While these algorithms can be very useful and powerful, they often have biases that can lead to injustice or discrimination in society. These problems have had and will continue to have a serious impact on marginalized groups (Source).
Disadvantages of algorithms
The video discusses the drawbacks of algorithms in detail, and the speaker emphasizes the dangers of algorithms to society and human life.
One example of how algorithms can hurt people is Twitter’s moderation algorithm, which suspends users without thinking about the situation or the big picture. In addition, Activision Blizzard’s and Electronic Arts’ proprietary algorithms that encourage players to get more out of their gaming experiences are negatively impacted.
Other problems related to algorithmic harm have also been cited, such as:
- Facial recognition systems.
- Fealth care systems.
- Recruitment algorithms.
- Deep fakes that threaten digital appearances.
Some argue that the consequences can be far-reaching, even devastating, with little accountability (Source).
Understand the impact of algorithmic bias on society.
Algorithm bias occurs when the preferences and values that are coded into the output of an algorithm consistently harm members of certain social groups. This is a serious problem because it can lead to discrimination in areas such as education, employment, housing, and health care.
It is critical to understand both the intentional and unintentional ways that bias can be built into algorithms to better regulate their use in our lives. By identifying and correcting these problems, we can begin to prevent the inequality, injustice, and other societal ills that result.
Analyze how automation perpetuates existing biases and prejudices.
When algorithms are built using biased data sets, they can directly reinforce pre-existing biases and prejudices.
When machine learning algorithms are trained on data that has been corrupted by bias, the result is a lower quality of results for those underestimated by the algorithm. These incorrect results can further entrench existing social inequality by excluding underrepresented populations from resources.
Experts suggest using rigorous fairness methods, such as counterfactual fairness, that check for inconsistencies in automated decisions to prevent this from happening.
Work to correct unfair treatment of social groups in algorithms.
Social groups such as women and minorities are often discriminated against by algorithms used in automated decision making. But this doesn’t have to be the case.
Organizations should work to correct unfair treatment of social groups in algorithms by incorporating counterfactual fairness methods that examine discrepancies in automated decision outcomes. In addition, organizations must actively work to eliminate algorithmic bias by ensuring that their data sets are non-discriminatory and implementing process controls to verify the accuracy of an algorithm’s predictions against a range of data points from multiple sources.
Commit to diversity, fairness, and transparency in algorithms.
It’s imperative to ensure that people affected by algorithms have access to data about the decisions being made so they can question and challenge them. At the same time, companies should work to ensure that their algorithms are trained on diverse data sets. This will reflect the diversity of their customers or those affected by their algorithmic decisions.
Ensuring fair representation in data sets helps prevent algorithmic bias and ensures that everyone gets a fair shake. In addition, organizations should evaluate whether algorithms are effectively working toward established goals and requirements.
Explore alternatives to algorithmic models to reduce inequality.
To ensure that people affected by algorithms aren’t discriminated against, it’s important to consider alternative models whenever possible. For example, research has shown that manual oversight can produce better results than models based on algorithms alone. This is especially true when bias is an issue. In addition to reducing disparities, manual oversight can help organizations identify areas of their operations that may need deeper scrutiny and investigation. This could lead to improvements in the services they provide.