Week 5 - BALT 4362 - AI Bias

-

 Chapter 8 of Automation with AI introduces some concerns with AI that I would not have considered. Since it is artificial intelligence, I thought it would not be possible for it to exhibit bias or discriminate. I decided to go straight to the source and ask ChatGPT how it could have these traits. It agreed with what the book was saying that training data was the root of this problem. 

"ChatGPT is trained on massive amounts of text data from the internet, books, and other sources. Because those sources can contain stereotypes, biases, or discriminatory views, the model can sometimes reflect or amplify them. OpenAI uses techniques like reinforcement learning with human feedback (RLHF), safety filters, and careful dataset curation to reduce these risks, but bias can still slip through. There are two main ways bias might appear: Content bias – If certain perspectives or groups are underrepresented or misrepresented in training data, the model’s responses may skew in a particular direction. Interaction bias – Depending on how you phrase a question, ChatGPT may frame its response differently, sometimes unintentionally favoring one view or group." 

Because their sources were written with bias, the response cannot help but have bias. OpenAI has acknowledged that bias can never be fully removed. They can continue to improve their algorithms, and they will need to, as the number of people relying on AI to summarize their content is increasing rapidly. One of the best examples I can provide as evidence of this is the social media app X. X has its own AI called "Grok". If you tag Grok in the comments of a post, it will come up with a response. On almost every post that deals with a world event, you will see hundreds to thousands of people tagging Grok, asking the AI to summarize, confirm, or prove the post wrong. 



Comments

Post a Comment

Popular posts from this blog

Jude's Introduction Post 4361-4364

-
Image
Welcome to my blog!  My name is Jude Sullivan, and this will be my third Fall here at Benedictine. I am originally from Paris, Illinois but made the decision to move across the country and continue my baseball career in warmer weather, at the Mesa campus. I played on the baseball team my first two years, but I have decided to focus on school/work for this Fall semester. I will be graduating in December with an Accounting degree, and I am looking forward to learning the skills that this class will require.
Read more

Week 8 - BALT 4362 - AI Image and Video Generation

-
Image
AI’s ability to create realistic images and artwork has exploded in the last few years, and in my opinion, it is both amazing and terrifying. With the correct prompt, anyone can make professional-quality pictures, no art degree required. Businesses can now use tools like DALL·E, Midjourney, and Canva’s AI generator to design posters and logos in seconds. This levels the playing field for small businesses that do not have the marketing budget to compete with bigger corporations. This can also be applied to students. You can make visuals for class projects, creative writing, or social media without needing advanced design skills. I know I have been able to use it to create presentation visuals I never would have been able to come up with on my own. It does a great job bridging the gap for people who might lack creativity, like me.  But as cool as it is, there are some real downsides. Artists worry about AI copying their work or taking credit for their style, and some AI-generated im...
Read more

Week 2 - BALT 4363 - CH 2

-
Image
Chapter Two of Data Toolkit,  and more specifically the exercises within the chapter,   helped me understand the basics of how Python can control and organize what a program does. Control structures like loops are like giving a set of instructions that the computer can follow repeatedly or only under certain conditions. The examples using “for” and “while” loops made it clearer how repetition works. An example of this would be repeating something a certain number of times or until a condition changes. I liked the comparison to a remote-control car, because it helped me see how code can “steer” what happens in a program. Seeing the examples with “if” and “if-else” statements helped me see how Python can make decisions, like checking whether a number is even or odd. The guessing game example at the end was especially interesting because it combined all the ideas, loops, conditions, and user input into one project. It made Python feel a little less intimidating, since I could se...
Read more