ChatGPT: New AI tool, old racism and bias?

Every time a new application of AI is announced, I get a brief burst of excitement. Soon followed by a knot on my stomach. This is because we know that technology is often not designed with fairness in mind.

One system, ChatGPT has arrived 100 million unique users(opens in new tab) Only two months after its release. This text-based tool engages users in interactive, friendly interactions generated by AI. This chatbot is developed to speak authoritatively on any subject you are prompted to address.

In an interview with Michael Barbaro daily podcast(opens in new tab) from new york times, technical reporter Kevin Roose said an app similar to ChatGPT, Bing’s AI chatbot, built on OpenAI’s GPT-3 language model, will serve as a side dish to French onion soup for Valentine’s Day dinner. I explained how I responded to a request for suggestions. his wife. Bing not only answered my question with salad recommendations, but also showed me where to find the ingredients in the supermarket and how much I needed to make the recipe for two. Valentine’s Day — You can also add a heart emoji.

The precision, specificity, and even charm of this interaction speaks to the accuracy and depth of knowledge needed to advance technology. Who wouldn’t believe a bot like this?

Bing analyzed the keywords in Roose’s prompts (especially “French onion soup” and “sides”) and used a matching algorithm to create responses that were most likely to answer Roose’s query. provided this information. The algorithm is trained to respond to the user’s prompts using a large language model developed by engineers working at OpenAI.

In 2020, members of the OpenAI team published an academic paper stating that the language model is the largest ever created. 175 billion parameters(opens in new tab) Behind that function. Having such a large language model means ChatGPT can speak anything, right?

Unfortunately, it doesn’t. Models of this scale require input from people around the world, but inherently reflect the biases of their creators. This means that the contributions of marginalized women, children, and others in human history will be underestimated, and this bias is reflected in her ChatGPT features.

AI Bias, Bessie and Beyoncé: Can ChatGPT erase the legacy of Black Excellence?

I made a guest appearance this year as well. The Karen Hunter Show(opens in new tab)and she noted that at the time she was unable to answer her specific question when asked if ChatGPT was an artist. bessie smith(opens in new tab) It inspired gospel singer Mahalia Jackson, but without additional prompts to introduce new information.

The bot was able to provide biographical information for each woman, but could not reliably discuss the relationship between the two. Bessie Smith was not only an influence on Jackson, but by musicologists, Laid the foundations of popular music in America(opens in new tab)She is said to have inspired hundreds of artists. Elvis Presley, Billie Holiday, Janis Joplin(opens in new tab)However, ChatGPT failed to provide this context regarding Smith’s influence.

This is because one of the ways racism and sexism manifest themselves in American society is through the erasure of contributions made by black women. and culture in general. “White Supremacist, Capitalist, Patriarchal” Values(opens in new tab) that shaped the United States.

Smith’s contribution is therefore minimal.As a result, when OpenAI engineers were training his ChatGPT model, Smith’s influence on contemporary American music. This was made evident by ChatGPT’s failure to give Hunter an adequate response.In doing so, failure reinforces the minimization of contribution by black women as the norm in the music industry.

For a more modern example of exploring the potential effects of prejudice, consider the fact that Beyoncé, despite being the most famous Grammy Award winner in history, has never won Record of the Year. prize. why?

One Grammy voter. variety As a “70s music business veteran,” he called Beyoncé’s renaissance The fanfare surrounding its release “Too weird.”(opens in new tab) The influence of this opinion has contributed to artists continuing to lose recognition for Record of the Year, regardless of the quality of the album itself.

Looking ahead from a technical perspective, imagine an engineer developing a training dataset for the most successful music artists of the early 21st century. If her status as a Record of the Year Grammy Award winner was weighted as an important factor, Beyoncé might not appear in this dataset, which is ridiculous.

Underestimated by society, underestimated by AI

Oversights of this nature infuriate me, as new technological developments allegedly advance our society. If you are middle class, cisgender, heterosexual white, they are. However, if you are a black woman, these applications reinforce Malcolm X’s claim. Black women are the most neglected people in America(opens in new tab).

This devaluation of what Black women contribute to the wider society affects my perception of the tech industry. For context, I am widely regarded as an expert on the racial impact of advanced technological systems and am regularly asked to serve on advisory boards and support product teams across the tech industry. At each of the venues, I attend conferences during which people are amazed at my expertise.

This despite the fact that I lead the team Algorithm Accountability Law(opens in new tab) To the U.S. House of Representatives in 2019, and then again 2022(opens in new tab)and the wording on impact assessment: US Data Privacy Act of 2022(opens in new tab)The non-profit organization I lead has been asked to help shape the UN’s thinking on algorithmic bias. And despite the fact that he held fellowships at Harvard, Stanford, and the University of Notre Dame, where he examined these issues.

Despite this wealth of experience, my presence is full of surprises because black women are still seen as diverse employers and are not qualified for leadership roles.

ChatGPT’s inability to recognize the impact of racialized sexism may not be a problem for some. But when we think about it, it becomes a concern for all of us. Microsoft’s ChatGPT integration plans(opens in new tab) To the online search experience through Bing. Many people rely on search engines to provide accurate, objective, and unbiased information, which is impossible — not just because of the bias in the training data, but because the algorithms that drive ChatGPT provide information. It’s designed to be predictive rather than fact-checking.

This has already led to some notable mistakes(opens in new tab).

The question arises why use ChatGPT.

Bet on this movie accident(opens in new tab) Low, but consider the following: Colombian judges are already using ChatGPT in their sentencing(opens in new tab) — is a big concern for black people.

We have already seen Correctional Offenders Administration Profiling (COMPAS) Algorithm for Alternative Sanctions(opens in new tab) used in the United States predicts that black defendants will reoffend at a higher rate than white defendants. Imagine a ruling written by ChatGPT using arrest data from New York City’s “Stop and Frisk” era. 90% of black and brown men arrested by law enforcement were innocent.(opens in new tab)

Seize opportunities for inclusion in AI

Once we recognize the existence and importance of these issues, rectifying the omission of the voices of black women and other marginalized groups is within reach.

For example, developers may engage third-party validators or independent experts to conduct impact assessments on how historically marginalized groups of people use their technology to ensure that training data defects can be identified and addressed.

Releasing new technology in beta to trusted users, like OpenAI did, could also improve representation. That is, if the pool of “trusted users” is comprehensive.

In addition, if legislation like the Algorithm Accountability Act, which was resubmitted to Congress in 2022, is passed, it will increase the rights of American citizens, including requirements for impact assessments and transparency about when and how technology is used. Federal guidelines are established to protect Safeguard.

I sincerely hope that technological innovation will give rise to new ways of thinking about society. The rapid adoption of new resources like ChatGPT has allowed us to quickly enter a new era of AI-supported access to knowledge. However, using biased training data carries the legacy of suppression into the future.

Mashable Voices’ columns and analysis reflect the opinions of the writers.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *