In this era of remarkable technological advancement, a standout development has been the emergence of generative artificial intelligence (AI) which is able to generate new content that can be used for many purposes. The AI only requires a few words of text to generate new content.
Each instalment in this series will delve into a specific use case, illustrating how generative AI can be effectively utilized in diverse scenarios. We'll investigate the most suitable tools for each task, reveal the most effective prompts (input text), and share insider tricks to speed up the journey from input to ideal output.
Remember that when using AI tools, especially generative ones like ChatGPT, adherence to best practices is vital. Review and follow the guidelines provided in the Guide on the use of Generative AI; they ensure responsible and ethical use of such technologies.
This article covers three angles at the intersection of inclusion and generative AI:
- what is inclusive language, and how to prompt an AI tool to draft content that uses inclusive language
- how to conduct inclusive language reviews
- how to avoid outputs that reflect biases in the underlying datasets
Inclusive language using AI
Effective communication that reflects diverse backgrounds and perspectives is crucial. Using inclusive language in professional documents, such as emails or reports, not only promotes respect and clear communication but also elevates an organization’s reputation. It's worth noting, however, that crafting such messages can be tricky, as some word choices may unintentionally introduce bias.
With AI tools like large language models (LLM), a type of deep learning model designed to understand and generate human-like text based on the data it has been trained on, it's possible to draft messages that are both clear and inclusive.
While this guide primarily references ChatGPT, the methods discussed work with any LLM-based AI tool.
What is inclusive language?
Inclusive language is a form of communication that avoids exclusionary or discriminatory words, phrases, or expressions. It respects and acknowledges all individuals, regardless of their gender, race, ethnicity, disability, age, socio-economic status, or other identity factors. The Guidelines for Inclusive Writing assist the federal public service and other organizations in creating content free from biases, discriminatory slang, or expressions. This approach promotes fairness and ensures everyone feels respected and included, regardless of their unique characteristics.
Inclusive language encompasses a broad spectrum of considerations. To communicate effectively and respectfully, one should be aware of the various major categories of inclusive language.
- Gender and sexuality: Inclusive language ensures that content doesn’t assume or exclude any gender.
- Race and ethnicity: Avoiding racial or ethnic stereotypes and using appropriate terms for different groups is crucial.
- Age: Being mindful of age helps prevent ageism in communication.
- Disability and neurodiversity: Language should be respectful and not reduce a person to their disability.
- Socio-Economic status: Avoid making assumptions or using derogatory terms related to one's economic status.
- Religion and belief systems: Being respectful of various religions and belief systems, without making assumptions or stereotypes, is crucial.
- Family structure and marital status: Avoid making assumptions about someone's family structure or marital status.
- Body size and appearance: Comments or assumptions based on someone's body size or appearance can be offensive.
- Education and literacy: Avoid making assumptions or derogatory references to someone's level of education or literacy.
- Geographic and cultural references: Respect differences and avoid stereotypes related to geographic or cultural origins.
Prompts that do not discriminate
The above list, though not exhaustive and subject to change, may appear daunting and almost overwhelming.
Fortunately, many LLMs are familiar with the concept of inclusive language—the key is to direct the AI to apply it. Users can request tools such as ChatGPT to generate text with inclusive language by simply specifying the need for inclusivity in the prompt (input text). Here's an example:
Prompt:
Draft a paragraph using inclusive language about our upcoming team meeting about the 2023 GCWCC.
Since ChatGPT is a generative AI, the response will be different each time. Here is a potential sample answer.
ChatGPT response:
Certainly! For those unfamiliar, the GCWCC refers to the Government of Canada Workplace Charitable Campaign.
---
Dear Team,
I'm reaching out to bring attention to our forthcoming team gathering focused on the 2023 GCWCC. Recognizing the diversity of backgrounds, experiences, and insights within our team, we believe that every voice is paramount in shaping how we approach this year's charitable endeavours. Whether you've been closely involved in past campaigns or are new to the GCWCC, your perspective is invaluable. We aim to cultivate an environment where all feel empowered to contribute and share. As we prepare to discuss our collective participation and potential impact, please know that your input is not only welcomed but essential to the richness and success of our efforts.
---
Remember to adjust any specific details or context as needed to fit your organization's unique situation.
Note that ChatGPT is knowledgeable about GCWCC and its objectives because it was pretrained on a wide variety of sources, including books, articles, and websites that span a broad spectrum of topics. However, there may be times when the AI lacks specific knowledge, requiring users to provide more detailed information to achieve the desired output.
Need an inclusive language review? AI can do that too...
Generative AI can also review an existing document and ensure it is written with inclusive language and make edits as needed. For this example, we will use the memo created with the assistance of AI in the previous article of this series: Working with Artificial Intelligence Series: Memos in Minutes.
Prompt:
Subject: Annual National Public Service Week Picnic - Save the Date and RSVP
Dear Team,
We are delighted to announce our Annual National Public Service Week Picnic, a cherished tradition here at our organization, and warmly extend an invitation to each and every one of you. This year, we want to make this event extra special, and we would love for you to bring your family members along for a day filled with fun and camaraderie.
The picnic will be held on Wednesday, September 13, from 9AM to 2PM. We've selected the scenic Community Park as the venue, providing us with plenty of open space to enjoy a variety of activities. In keeping with our casual yet professional culture, we recommend a business casual dress code for the day.
We are making arrangements to ensure plenty of delicious food and refreshing drinks for everyone. To help us plan effectively, we kindly request that you confirm your attendance, as well as the number of family members you'll be bringing along, by [insert RSVP date here].
This event offers a wonderful opportunity to strengthen our community, enjoy some time outside of our usual work environment, and create lasting memories with our families and colleagues. We truly hope to see all of you there.
Thank you for your attention to this matter and for being a valued part of our team.
Warm Regards,
[Your Name]
[Your Position]
Review the above text for inclusive language. Please list recommended changes in bullet form, and then rewrite the text using the suggestions provided.
In the prompt (input text) above, the AI received several instructions to follow.
- Examine the text for possible inclusive language concerns.
- List these concerns for further examination in a bullet format.
LLM-based tools may not always capture the nuances or specific contexts of human language effectively. By requesting the AI to detail the issues it identifies, we can better understand the basis of its answers. This helps in assessing its accuracy, acknowledging its shortcomings, and promoting critical thinking.
ChatGPT response:
Here are a few recommendations to ensure the language used in the invitation is more inclusive:
1. "Bring your family members": The term "family" can be diverse in its structure. Some employees might have chosen families, roommates, or other important people in their lives that don't fit into the traditional definition of "family." Consider using a term like "loved ones" or "close ones."
2. "Business casual dress code": This might imply a specific type of attire that not everyone might feel comfortable or identify with. It might be better to phrase it in a way that gives a guideline but also allows flexibility, emphasizing comfort.
3. General Tone: The tone is great, but it might be worth highlighting the inclusive nature of the event, to make everyone feel welcome, regardless of their backgrounds, roles, or duration with the company.
At first glance, the original memo appeared to satisfy all categories for inclusive writing. Still, the AI found areas of improvement.
At times, particular attention may be needed for a specific text. An example would be to aid in drafting sample emails about the office dress code, using a gender-neutral and inclusive tone. Once again, the solution is to be specific in the instructions passed on to the AI via the prompt.
Prompt:
Write the main body of a memo (do not include any header, greeting or closing text) in a gender-neutral and inclusive tone about the new office dress code. Provide three versions of the main body of the text, labelled as version 1, version 2 and version 3.
Once again, the AI received several instructions to follow in the input text.
- Only write the main body for a memo on the requested topic.
- Use gender-neutral and inclusive language.
- Create three versions of the text, each clearly labelled.
ChatGPT response:
Version 1:
We understand the importance of fostering a comfortable and inclusive environment for everyone. In light of this, our new dress code policy encourages individual expression while maintaining a professional atmosphere. We request that all attire be clean, free of offensive graphics or messages, and suitable for a general workplace environment. Be conscious of any clothing items that may pose safety concerns, depending on your specific role. Please remember, while we want you to feel at ease, it's essential to maintain an image that respects all colleagues and visitors. Should you have any queries or need clarifications on the policy, please don't hesitate to reach out.
Version 2:
In our ongoing commitment to create a workspace that's inclusive and considerate of everyone's needs, we are implementing a revised dress code. This policy allows each of us to showcase our unique style, but we must ensure it remains appropriate for a professional setting. All clothing should be clean, without any graphics or slogans that may be considered disrespectful. Additionally, for safety reasons, specific roles may have additional guidelines. We're counting on each team member to uphold this balance of individuality and professionalism. Any questions or feedback are always welcome.
Version 3:
Ensuring a respectful and inclusive workplace is one of our top priorities. To this end, we're introducing a new dress code that emphasizes both personal expression and professional respect. All team members are kindly asked to wear clean attire, avoiding any potentially inappropriate or offensive graphics. Depending on your role, there might be additional safety guidelines to follow. It's our collective responsibility to create a welcoming environment, so we trust everyone to align with this balance. Should you have any concerns or need further information on the policy, we're here to assist.
A good follow-up trick is to ask the AI to evaluate its own output.
Follow-up prompt:
Review the above text for inclusive language. Please list recommended changes in bullet form, and then rewrite the text using the suggestions provided.
Asking ChatGPT to review its own work offers several benefits:
- Self-awareness: While ChatGPT doesn't possess consciousness or emotions, by prompting it to "review" its responses, you're effectively accessing its underlying mechanisms to break down information, potentially highlighting the rationale behind its outputs.
- Enhanced accuracy: By cross-checking or refining its initial response, ChatGPT can offer more accurate or better-framed information in some cases.
- Error identification: The model can occasionally provide information that is incorrect or misinterpreted. Asking it to review its work might help in spotting such errors.
ChatGPT response:
Upon reviewing the provided text, here are a few areas that could potentially benefit from more inclusive language:
Recommendations:
In Version 1, the phrase "maintain an image that respects all colleagues and visitors" could be made more inclusive by emphasizing behaviour rather than image.
In Version 2, "showcase our unique style" might be refined to avoid any potential misunderstanding or misuse of the dress code policy.
For all versions, the directive to avoid "offensive graphics or messages" could be clarified further to specify graphics or messages that are discriminatory, prejudiced, or harmful in nature.
Revised Text:
Version 1:
We understand the importance of fostering a comfortable and inclusive environment for everyone. In light of this, our new dress code policy encourages individual expression while ensuring a professional atmosphere. We request that all attire be clean, free of graphics or messages that are discriminatory or prejudiced, and suitable for a general workplace environment. Be conscious of any clothing items that may pose safety concerns, depending on your specific role. Please remember, while we want you to feel at ease, it's essential to conduct ourselves in a manner that respects all colleagues and visitors. Should you have any queries or need clarifications on the policy, please don't hesitate to reach out.
Version 2:
In our ongoing commitment to create a workspace that's inclusive and considerate of everyone's needs, we are implementing a revised dress code. This policy is designed to foster a balance between personal style and workplace appropriateness. All clothing should be clean and devoid of graphics or slogans that could be considered discriminatory or prejudiced. Additionally, for safety reasons, specific roles may have additional guidelines. We're counting on each team member to uphold this balance of individuality and professionalism. Any questions or feedback are always welcome.
Version 3:
Ensuring a respectful and inclusive workplace is one of our top priorities. To this end, we're introducing a new dress code that emphasizes both personal expression and professional respect. All team members are kindly asked to wear clean attire, steering clear of potentially discriminatory or prejudiced graphics or messages. Depending on your role, there might be additional safety guidelines to follow. It's our collective responsibility to create a welcoming environment, so we trust everyone to align with this balance. Should you have any concerns or need further information on the policy, we're here to assist.
Ensuring text is unbiased using AI
In addition to inclusive language, it's important to ensure the text isn’t inherently biased. As mentioned earlier, LLMs (such as ChatGPT) are trained on a diverse dataset, including a vast subset of internet-based text and can reflect societal biases as a result. For instance, critics of one LLM—which has been updated since—noted that prompts asking for a list of prominent scientists generated only names of men.
As with inclusive language, users can guide the AI to produce unbiased content by following a few guidelines:
- Specify your needs: clearly ask for unbiased content in your prompts.
- Consider your biases: take a moment to reflect on your own mental model for the content you’re generating.
- Ask for facts: requesting fact-based writing can help to keep opinions and biases out of the content.
Prompt:
Write a fact-based paragraph on the importance of inclusivity in the modern workplace. Include your sources.
Here we have introduced the concept of asking for sources. There are several reasons why you should ask ChatGPT (or any AI model) for its sources when generating an article:
- Verification of information: By knowing and verifying the sources, we ensure that the information is accurate and comes from reputable references.
- Avoiding plagiarism: If the content generated resembles existing material too closely without proper attribution, it could be considered plagiarism.
- Depth of research: Knowing where the information came from allows to further delve into those sources for a more in-depth understanding or to find more detailed data and context related to the topic.
- Understanding AI limitations: AI models don't "know" sources in the same way humans do. Instead, they generate responses based on patterns from the data they've been trained on. Asking for sources can provide a clearer idea of the model's basis for its statements.
ChatGPT response:
Certainly!
Inclusivity in the modern workplace is crucial for multiple reasons. Research indicates that diverse and inclusive teams can significantly outperform their counterparts. According to a study by McKinsey & Company, companies in the top quartile for gender diversity on executive teams were 21% more likely to outperform on profitability and 27% more likely to have superior value creation. Furthermore, inclusive companies are more likely to be innovative and are better at capturing new markets. A BCG study found that companies with more diverse management teams have 19% higher revenues due to innovation. Inclusivity also plays a pivotal role in attracting and retaining top talent. A Glassdoor survey revealed that 67% of job seekers consider workplace diversity an essential factor when evaluating companies and job offers. Thus, fostering inclusivity isn't just a moral imperative—it offers tangible benefits to businesses.
Sources:
1. Hunt, V., et al. (2018). "Delivering through Diversity." McKinsey & Company.
2. Rocío Lorenzo, et al. (2018). "How Diverse Leadership Teams Boost Innovation." BCG.
3. Glassdoor (2014). "Glassdoor Survey Finds More Than Two-Thirds of People Say Diversity Important When Deciding Where to Work." Glassdoor.
The AI-generated text can now be cross-referenced with online search engines to confirm the accuracy of the generated text.
The last step involves refining the output text to make it more suitable and appealing to the target audience.
Follow-up Prompt:
Can you rewrite the above text for a general audience, removing all mentions of research and studies, while focusing on the importance of inclusivity in the workplace?
ChatGPT response:
Of course!
Inclusivity in the workplace is essential for fostering creativity and driving success. When a team is diverse, bringing together various backgrounds, experiences, and perspectives, it tends to perform better and come up with innovative solutions. Diverse companies often see better financial results and are more likely to capture new markets. Moreover, a workplace that champions inclusivity is more appealing to potential employees. People are drawn to companies that value and respect the uniqueness of every individual. So, promoting inclusivity is not only the right thing to do; it also makes good business sense.
Inclusive and unbiased language is an essential aspect of modern communication. Leveraging ChatGPT and other such tools can be an effective way to ensure our emails, memos, and reports are inclusive and respectful of all audiences. Remember, the key lies in framing your input text (prompts) correctly and always being vigilant about reviewing the generated content. As recommended by the Guide on the use of Generative AI, always align with the FASTER principles (Fair, Accountable, Secure, Transparent, Educated, and Relevant) to ensure the responsible use of generative AI tools.
Resources
Enjoying this article series? Read this article next: Working with Artificial Intelligence Series: Memos in Minutes
- Article | OpenAI's ChatGPT Explained
- Article | Demystifying Artificial Intelligence
- Article | Using Large Language Models (like ChatGPT) in the Federal Public Service
- Article | Dialing Down the Heat with Machine Learning
- Course | Discover Artificial Intelligence (DDN210)
- Course | Getting Started with Machine Learning (DDN220)
- Guide on the use of Generative AI - Canada.ca
