Black founders are building a customized ChatGPT for a more personalized experience

At first, John Passmore was excited about ChatGPT.

The serial founder has been working in the field of artificial intelligence since at least 2008. He recalls the days when experts said it would take decades for the world to see something like ChatGPT. Fast forward – that day is now here.

But there is a catch.

ChatGPT, one of the world’s most powerful artificial intelligence tools, struggles with cultural nuances. For a Black man like Passmore, this is understandably annoying. In fact, this oversight has sparked resentment from many Black people who already don’t see themselves properly represented in the algorithms that are asked to one day save the world. The current ChatGPT provides answers that are too generalized for specific questions that cater to certain communities, as its training seems Eurocentric and Western in its bias. This isn’t unique — most AI models aren’t built with people of color in mind. But many Black founders are adamant about not being left behind.

A number of Black-owned chatbots and ChatGPT variants that specifically serve Black and brown communities have emerged in the past year, as Black founders like Passmore seek to take advantage of OpenAI’s cultural decline.

“If you ask the model who the most important artists in our culture are, it will give you the names of Leonardo da Vinci and Michelangelo,” Passmore said of ChatGPT. “It’s not going to say anything about India or China, Africa or even African Americans, because its bias is focused on the European trajectory of history.”

So Passmore launched Latimer.ai, a language model that gives customized answers to reflect the experiences of Black and Brown people. Erin Reddick started ChatBlackGPT, a chatbot focused on Black and Brown communities. Globally, there’s Canada-based Spark Plug, which is essentially a ChatGPT for Black and Brown students. Africa is also seeing massive innovation in this area, with language models emerging to cater to the more than 2,000 languages ​​and dialects spoken on the continent that Western AI models still ignore.

“We are the keepers of our own stories and experiences,” Spark Plug founder Tamar Huggins told TechCrunch. “We need to build systems and infrastructure that we own and control to ensure our data stays ours.”

Personalized AI is here

Generalized AI models cannot easily capture the African American experience because many aspects of that culture are not online. Current algorithms use the internet for sources, but many traditions and dialects within African American culture are passed down orally or directly, leaving a gap between what AI models understand about the community and the nuances of what actually happens.

That’s why Passmore tried to use sources like the Amsterdam News, one of the oldest Black newspapers in the US, when building Latimer.AI, focusing on accuracy rather than training on user-generated data scraped from the internet. By doing this, he began to see differences between his model and ChatGPT.

He recalled that people once asked ChatGPT about the Underground Railroad, the route used by enslaved Black Americans to escape slavery to northern states. ChatGPT’s model would refer to runaway slaves, while Latimer.AI adjusted the wording, referring to “slaves” or “people seeking freedom,” which has become more socially favorable when discussing formerly enslaved people.

“There are some subtle differences in the language the model uses because of the training data, and the model itself only thinks about black and brown people,” Passmore said.

Meanwhile, Erin Reddick’s ChatBlackGPT is still in beta mode and plans to launch on Juneteenth. Her product works just as it sounds: a chatbot where one can ask questions and receive customized answers about Black culture. “The core of what we’re doing is really community-driven,” she said.

She is in the process of building the tool, asking users what they want it to look like and how they want it to work. She is also working closely with education institutions like Historically Black Colleges and Universities (HBCUs) to work with students to teach them and help them train her algorithms. She said she wants to create “a well-rounded learning opportunity for black and brown people so they have a safe space to explore AI.”

“The algorithm prioritizes Black information sources, so that it’s talking to a source of knowledge that’s more immediately relevant than your average experience,” he told TechCrunch, adding that, like Passmore’s product, technically anyone can use it.

Tamar Huggins built Spark Plug to provide a more customized experience to Black and brown communities. Her platform translates educational content into African American Vernacular English (AAVE), the ethnic language associated with Black American communities. That dialect is traditionally passed down orally and directly, rather than studied and written down like Standard English, which means an AI model (or person) learning it from the internet alone will have decreased accuracy. Accurately capturing AAVE is important, not only because the chatbot will respond using it, but also so students can more easily type in prompts that will return the results the AI ​​needs to them.

“By creating content that connects with Black students, we ensure they see themselves in education, which is critical for high engagement and academic success,” Huggins said. “When given the opportunity, Big Tech will almost always prioritize profits over people. So we created our own path within the AI ​​space.”

Huggins trained her algorithm on Black writers of the Harlem Renaissance, Black writers in education, and even her teenage daughter’s vocabulary to capture the essence of AAVE. Huggins also works with educators, linguists, and cultural experts to review and validate Spark Plug’s output. Her product is also not built on top of ChatGPT. It’s its own model, meaning users control their data.

Passmore also plans to build a separate underlying model for his Latimer.AI. Right now, he is working on expanding his company into schools, especially HBCUs, as more and more students turn to ChatGPT to get their work done every day.

“It’s a better AI companion for most of the tasks that black and brown kids are given,” he said.

Uniting the Diaspora Community

Africa is finding itself overlooked in the current AI movement. For example, according to the 2023 Artificial Intelligence Index report, only 0.77% of the world’s total AI journals come from sub-Saharan Africa, while East Asia and North America account for 47.1% and 11.6% respectively. Population-wise, Africa is home to about 17% of the world’s population, while North America has only 7%. When it comes time to bring in AI knowledge and experts, research from sub-Saharan Africa is significantly less likely to be used, which could impact the development of global AI tools.

While great progress is being made in Africa toward building more inclusive language models that can better serve the Black diaspora, right now existing AI models from ChatGPT to Gemini cannot fully support the over 2,000 languages ​​spoken in Africa.

Yinka Iyinolakan created CDIAL.AI to solve this problem. CDIAL.AI is a chatbot that can speak and understand almost all African languages ​​and dialects, with a special focus on speech patterns rather than text.

Iyinolakan reiterated to TechCrunch the same sentiment expressed by many Black Americans — that basic AI models are primarily scraped from internet data and the most commonly spoken languages. Much like its African American offspring culture, many African languages ​​and traditions are absent from the internet, as it is a culture that has historically been communicated verbally rather than in written form. This means AI models do not have enough information about African cultures to train themselves, thus creating a knowledge gap.

For CDIAL.AI, Iyinolakan brought together more than 1,200 native speakers and linguists from across Africa to gather knowledge and insights to create what he calls “the world’s first multilingual voice-first large language model.” The company plans to add more languages ​​over the next 12 months and build models that support text, voice, and images.

He’s not alone here. Google recently awarded Kenya-based Jacaranda Health a $1.4 million grant to develop its machine learning services so it can work in more African languages, and Intron Health recently raised several million dollars to enhance its clinical speech recognition for more than 200 accents spoken across Africa.

“Silicon Valley wants to believe it has it all when it comes to artificial intelligence,” Iyinolakan said. “But to ‘get’ artificial intelligence, which is a goal for all companies, they need to incorporate a third of the world’s knowledge.”

making progress

Working on AI chatbots isn’t the only innovation Black founders are trying to tackle.

Steve Jones started the Pokestock company to create stock images of people of color, because for decades stock imaging has lacked those representing minorities. That’s one reason why models today are primarily creating pictures of white people when users ask them to create pictures of anything from doctors to pop singers.

“All platforms and tools must be trained with complete, racially inclusive, and culturally accurate data, otherwise we [perpetuate] “Our larger society currently has bias issues,” Jones told TechCrunch. To address this, Pokstock has spent the past five years collecting diversity data and building its own visual tagging system that the database uses to help businesses train their AI models so it can produce more inclusive imaging.

However, some improvements are happening. Jones said he’s noticed that the big stock imaging companies that source AI companies are taking more steps in increasing the diversity of their content. Passmore also sees the future as bright, saying that personalized AI is the future anyway and that the more an AI model interacts with its users, the more it will understand the wants and needs of a specific individual, “which, I think, eliminates a lot of bias.”

There may be room for even more culturally-specific AI models in the future, especially as more and more Black-owned options continue to emerge. After all, the world is much bigger and more nuanced — trying to fit it into a black box serves no purpose.

“My hope is that as soon as possible in this next economic boom, more founders of color will be involved in developing their own AI platforms or creating new AI-related jobs,” Jones said. “AI is going to create billionaires, and I would love to see people of color as producers, not just consumers.”

This article has been updated to reflect what Spark Plug was trained with and that it and Latimer.ai have their own baseline models.