At Medusaas, we believe in the power of diversity and inclusion, and we’re committed to highlighting underrepresented voices through our "365 Days of Diversity" Instagram campaign. This campaign is designed to educate and inspire by showcasing pioneers and contributors from various marginalized communities across history. To streamline the research process, we turned to AI for assistance in compiling data. While the initial results seemed promising, a deeper dive into the data revealed some significant challenges that we had to address. This blog post will walk you through the following:
Navigating the Complexities of AI in Research: Lessons from Our 365 Days of Diversity Campaign
Prompt Used to Generate Data Set
To kick off our campaign, we created a series of prompts to guide the AI in generating the list of diverse individuals we wanted to highlight. We asked for pioneers, trailblazers, and unsung heroes across various fields, focusing on underrepresented groups such as Indigenous peoples, Black women, LGBTQ+ individuals, and others. Our prompts were carefully crafted to ensure diversity across geography, era, and field of achievement, with the hope that the AI would provide a comprehensive and balanced list. The first prompt we used is below in red text:
I’m doing a social media campaign on Instagram highlighting underrepresented voices and pioneers in their fields. I'm specifically looking to highlight:
Indigenous peoples North America
Indigenous peoples globally
Black women
Black men
Latina women
Women in the history of science
Women in the history of art
People who are neurodiverse
People who have disabilities
LGBTQ+ people
The people or groups should have significant lifetime accomplishments and achievements, and can be from any time throughout history. These people can be from any country in the world. The key here is to educate my followers on the significant contributions of underrepresented individuals and groups in history. Make sure you do not repeat any of the people on the list.
I would like your response to be in the format of a table with following format:
Column 1 - being the person or group's name,
Column 2 - ta short description of their achievements,
Column 3 - a brief description of how that achievement impacted society,
Column 4 - a link to that person or groups wiki page or other website dedicated to the history of the individual or group.
Column 5 – a 200 word summary of the person or group’s accomplishments that I can use in the post description
Column 6 – best hashtags for the post to drive interaction, reach, and awareness.
Let’s start with highlighting 25 [group name]
Follow-up prompt
Continuing on using the same instructions, let's move on to [group name]
This process was followed until all categories were completed.
What We Discovered in Data Analysis
As we started analyzing the data, it became clear that the AI-generated results were not as flawless as we had hoped. Here are the main challenges we encountered:
Duplicate Records: Despite instructing the AI to avoid duplicates, we found several repeated entries. We resolved this by cleaning the dataset and adding new names to ensure a total of 365 unique individuals.
Geographical Bias: The AI showed a strong preference for individuals from the United States, which wasn’t aligned with our global focus. To counter this, we had to refine our prompts, explicitly asking for pioneers outside the U.S., such as Indigenous scientists in STEM fields from regions like Australia and Canada.
Modern History Bias: Surprisingly, the AI heavily favored modern historical figures, which left gaps in representing earlier pioneers. We adjusted our approach by specifying the need for individuals who made significant contributions before 1950.
Mismatched Descriptions: In several cases, the person’s description did not match the associated Wikipedia link. This issue highlighted the importance of thorough validation, as we had to cross-check and, in some cases, replace individuals who were inaccurately represented.
Inaccurate and Controversial Entries: A few individuals could not be verified through Wikipedia or general search engines, and some were associated with controversial achievements. These entries were replaced with verifiable individuals whose contributions were more positively impactful.
Suggestions for our readers
The issues we encountered are not unique to our campaign, and they underscore the need for caution and diligence when using AI for educational purposes. Here are some practical tips for others:
Refine Your Prompts: Be as specific as possible in your initial prompts to avoid common biases. Explicitly state geographical regions, time periods, and any other factors that are critical to your campaign.
Expect to Validate: AI is a powerful tool, but it’s not infallible. Plan for a thorough validation process to verify the accuracy of the data. Cross-check with reliable sources and be prepared to make adjustments.
Beware of Biases: AI systems often reflect the biases present in the data they’ve been trained on. Be aware of these potential biases, and take steps to counteract them in your research and data collection.
Prepare for Cleanup: Duplicate entries and mismatched information are common issues. Allocate time and resources to clean up your dataset before finalizing your campaign.
A Lesson in Patience and Precision
Using AI to assist in compiling data for our 365 Days of Diversity campaign was both enlightening and challenging. While the technology can significantly streamline the research process, it’s clear that human oversight is essential to ensure accuracy, inclusivity, and integrity in the final product. By sharing our experiences, we hope to help others navigate the complexities of using AI for similar initiatives. Learn more about our 365 Days of Diversity campaign.
Follow Our Journey
We’re excited to share these diverse stories with you! Follow our "365 Days of Diversity" campaign on Instagram to learn more about the incredible individuals we’re highlighting throughout the year. Let’s celebrate diversity and continue to educate ourselves and others about the rich tapestry of human achievement.
Comments