1 of 6

Bias in the Myth of “Artificial Intelligence”

Adam Kareem El-Ramly

The Durham School of the Arts

2 of 6

Research Question:

How have Machine Learning (AI) algorithms been subject to unintentional bias against a wide range of underrepresented minorities?

Thesis Statement:

Machine Learning “AI” algorithms have grown significantly in use over the past decade. However, an unintended consequence of their growth is discriminatory bias towards marginalized minorities due to a lack of inclusion. This bias is a result of programmers and cloud companies not being open-minded when designing algorithms and perceiving “AI” to the public as more than it’s capable of.

3 of 6

Data

Machine Learning Algorithms used to detect diabetes are 13% less accurate towards people of color/
Diversified ML algorithms proved to be 95% accurate once tested on a wide variety of users.
NYPD identity algorithms contain 42000 gang affiliates with no requirements to prove gang activity.
Majority of ML algorithms have a 75% accuracy as a threshold.

4 of 6

Key Findings

“Artificial Intelligence” algorithms are truly binary lines of code, which are best represented by the name Machine Learning.
“Artificial Intelligence” has no critical thinking or incentive, making “Intelligence” a deceiving name.
Machine Learning algorithms directly reflect biases of those who program them.
Having diversified data and a proper testing process is what leads to more diversification in algorithms.
Corporations have little financial motivation to diversify algorithms.

5 of 6

Proposed Solution

It is vital that algorithm users are aware of the capabilities of “Artificial Intelligence”.

If users are aware of the functionality of Artificial Intelligence, then they can hold corporations accountable for potential bias.

Even though Machine Learning algorithms with diversification are more of an investment to produce, they can reach a broader scope of consumers if diversified.

6 of 6

Thank You

adam@elramly.com