CS 370 3-1 Hidden Bias in AI discussion post
.docx
keyboard_arrow_up
School
Southern New Hampshire University *
*We aren’t endorsed by this school
Course
370
Subject
Communications
Date
May 25, 2024
Type
docx
Pages
2
Uploaded by LieutenantHamster2518
Hello Class, After reading the three articles, It appears to me that AI bias isn’t too far removed from the bias of the sectors in which the AI is being trained to emulate, Two examples mentioned by Rebecca Heilweil in her article with Vox entitled “Why algorithms can be racist or sexist, and continued upon by George Denison of Prolific in his article “4 shocking AI Bias examples”, are Amazon’s development of an AI system to rate the match percentage of candidates based upon the job qualifications and the resume submitted, which showed an underlying bias against female applicants, and a system called COMPAS (Correctional Offender Management
Profiling for Alternative Sanctions) used by the DOJ (Department of Justice) to estimate the likelihood of reoffence if an inmate were to be released, inherently build and influenced by data skewed by centuries of systemic racism and abuse. In both scenarios, the training data itself is inherently misleading the AI technology during its training to develop biases based on the trends of past data. More specifically, the major prominence of men within the technology sector, and the overwhelming number of arrests and more severe sentencing of people of color versus their white counterparts. In my opinion, it appears both algorithmic biases stem from innate human bias. With the recognition of our own biases, and the biases within data, one reality and effort that is needed to make AI algorithms less biased would be to better vet training data samples, regular audits identifying equity within varying and diverse data sources helps train the system (Silberg and Manyika).This idea is further touched upon by Rosaria Silipo at Customer Think with her article entitled, “How to keep bias out of your AI models”, where she goes on to list her experience in training a model to use free text language given a certain number of characters to start with, She trained the algorithm using two sets of data, one using rap songs, the other using Shakspearian works. While using the training data of the rap songs, the AI would generate text laden with profanities and vulgarities, which are common
tropes within rap music, while maintaining the rhythmic and poetic patterns of speech that are exuded within rap as a genre. While using the Shakespearian works,
the algorithm used a manner of speech emulating its source data. While Shakespearian works aren’t free of profanities or vulgarities, the way in which they are expressed aren’t as impactful or even as forward as those expressed within rap, leading to inherent bias to what the AI thinks is appropriate to communicate to the user. This idea of intrinsic bias within training data isn’t new, it has been an ongoing issue, and one that has to be addressed in order to help deal with innate bias expressed within the AI algorithmic decision-making process. References:
Denison, George. “4 Shocking AI Bias Examples.” Prolific, 24 Oct. 2023, www.prolific.com/resources/shocking-ai-bias.Figure Eight Federal. “Overcome and Prevent Bias in AI.” Figure Eight Federal, 15 Jan. 2021, f8federal.com/overcome-and-prevent-ai-bias/.
Heilweil, Rebecca. “Why Algorithms Can Be Racist and Sexist.” Vox, 2020, www.vox.com/recode/2020/2/18/21121286/algorithms-bias-discrimination-
facial-recognition-transparency
.
Silberg, Jake, and James Manyika. “Tackling Bias in Artificial Intelligence (and in Humans).” McKinsey & Company, 6 June 2019, www.mckinsey.com/featured-
insights/artificial-intelligence/tackling-bias-in-artificial-intelligence-and-in-
humans
.
Silipo, Rosaria. “How to Keep Bias out of Your AI Models | CustomerThink.” Customerthink.com, 3 Mar. 2020, customerthink.com/how-to-keep-bias-out-
of-your-ai-models/.
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
- Access to all documents
- Unlimited textbook solutions
- 24/7 expert homework help