Do AI-Based Tools Understand Diversity Biases?

Over the past few months, we’ve witnessed a surge of AI-powered chatbots. ChatGPT is probably one of the most visible leaders in this field, and other companies have been working on their own generative AI products. Experts have indicated that this new generation of chatbots can complement or even compete with traditional search engines and disrupt the marketplace.

For marketers, this could be an exciting opportunity that can reshape how consumers search and shop for products and services—as well as reshaping the way technology can impact other aspects of our society, swaying and influencing attitudes and behaviors beyond mere transactions.

From a multicultural marketing perspective, the surge of AI-based tools may well represent an opportunity, as it can enhance and empower individuals to learn more about relevant topics in their lives, including their culture and traditions, but it can also represent risks.

For example, consumers with no understanding of how the information provided by AI-powered chatbots is created may take their answers as the absolute truth, especially given their eloquent and confident tone. This may reduce or, worse, eliminate the need for accuracy checking.

Moreover, chatbots may perpetuate biases and misperceptions about diverse segments, which, in an environment where some may perceive chatbots as the “authority,” can be a dangerous obstacle toward a more diverse and equitable marketplace.

To further understand these risks, I consulted the “oracle” itself. Below is an actual exchange I had with ChatGPT on this topic:

Isaac Mizrahi - How do I know your sources are not biased against multicultural segments?

ChatGPT - As an AI language model created by OpenAI, I don’t have personal opinions or biases. I am trained on a diverse range of sources and text written by individuals with varying perspectives and ideologies, which helps to ensure that my responses are balanced and neutral.

Additionally, OpenAI has implemented measures to prevent biases from being incorporated into the model during the training process, such as ensuring a balanced representation of different demographics and perspectives in the training data.

However, it’s always a good practice to cross-check information from multiple sources and consider different perspectives to gain a well-rounded understanding of a topic.

Mizrahi - But if some of your sources are biased, against any topic, AI responses would reflect these biases, right?

ChatGPT - Yes, it's possible that some biases could be present in the training data used to develop the AI language model, and therefore reflected in its responses. However, OpenAI has taken steps to minimize this risk, such as using large and diverse training data sets and implementing techniques to mitigate biases in the training process.

It’s important to note that AI language models like me can only provide information based on the patterns and associations learned from the training data. If there are biases in the training data, they can be reflected in the responses generated by the model.

However, it's also possible to detect and address these biases by continuously monitoring and improving the training data and the model’s responses. This can help to ensure that the information provided by AI language models like me is as accurate and neutral as possible.

In conclusion, while biases can be present in AI language models, steps can be taken to minimize and address them to ensure that the information provided is as accurate and neutral as possible.

These are some of the steps that can be taken to minimize biases in the training data used to develop AI language models. However, it’s important to note that eliminating biases from AI language models is an ongoing process that requires continuous monitoring and improvement.

Mizrahi - In your opinion, should advertisers invest in multicultural marketing?

ChatGPT - Yes, advertisers should consider investing in multicultural marketing. With the changing demographics in America and around the world, there is a growing market for products and services that cater to diverse cultural groups. By tailoring marketing messages and campaigns to specific cultural communities, advertisers can effectively reach and connect with these target audiences and build strong, lasting relationships with them.

In addition, research has shown that multicultural consumers are often more brand loyal and have higher brand engagement levels compared to non-multicultural consumers. This means that if a company can effectively reach and connect with multicultural audiences, they can build a loyal customer base that will continue to support them over time.

However, it’s important to note that simply translating advertising materials into different languages is not enough. Multicultural marketing requires a deep understanding of the cultural values, beliefs, and motivations of each target audience, and a tailored approach to messaging and creative that takes these factors into account.

When done effectively, multicultural marketing can lead to improved brand awareness, increased sales, and better customer engagement and loyalty. However, it must be approached with care, sensitivity, and an understanding of the cultural context in order to be successful.

In conclusion, multicultural marketing can be a valuable investment for advertisers looking to reach diverse audiences and build strong relationships with these communities. However, it must be approached with care and an understanding of the cultural context in order to be effective.

I spoke with Julio Arrieta, Global Chief Marketing Officer at Leapfinancial and an expert on the convergence of technology and multicultural marketing, and asked his opinion about the above exchange. Here are his thoughts:

“We have a responsibility as multicultural marketers to help provide large, prevalent Language Models like GPT4 with high-volume and quality data sets as well as few-shot learning – a machine learning framework that trains AI with specific examples – to ensure the content and answers produced remain unbiased and accurately represent our cultures. For decades, we’ve made it our duty to educate the industry on the business opportunity and the proper talent, resources and approaches to address multicultural audiences properly. It’s now our time to educate the machines; this will require we allocate the fair share of time, brainpower and financial resources to remain at least in the cultural co-pilot seat.”

In an era where we are still trying to close the original digital divide between mainstream consumers and multicultural segments, another technological advancement brings the duality of excitement and concern.

Multicultural segments historically have been the most “technological optimistic” group in America, willing to try and adopt new technologies faster than other segments, and this shouldn’t be any different when it comes to the adoption of AI-driven tools. However, concerns like the ones expressed in this article need to be taken into consideration if we hope to create a more representative and equitable society in years to come.