The Fast Company Impact Council is an invitation-only membership community of leaders, experts, executives, and entrepreneurs who share their insights with our audience. Members pay annual dues for access to peer learning, thought leadership opportunities, events and more.
Since ChatGPT’s launch in 2022, it feels like artificial intelligence is finally going mainstream. From Fortune 500 board rooms to dinner tables, everyone is talking about AI, its applications, and its promise. With more than $500 billion flowing into AI infrastructure investments, many investors predict the AI wave is just gaining momentum.
Those investors are right, AI still has a long way to go before it is truly ubiquitous. But more importantly, we have to tread carefully when we talk about AI going mainstream. The reality is that while many reading this article are already using AI in our daily lives, there are billions of people around the world who are a long way away from feeling AI’s impacts and opportunities.
So how do we truly change the world with AI? The opportunity isn’t just about reach, but about the underlying data and infrastructure that will be needed to make AI a truly global technology revolution.
Lessons from mobile phone adoption
We can learn a lot about the promises and pitfalls of technology revolutions by looking to the past. Today, 70.5% of the world’s population uses a cellphone. Yet, it’s taken nearly 50 years for cellphones to gain worldwide adoption since the first mobile phone call was made in 1973 by Martin Cooper, a Motorola executive, using a prototype mobile phone.
While mobile phone technology has improved significantly, with phones getting smaller and smarter over the years, the real power of mobile phones took hold with the cellular network’s evolution. The 2G cellular network introduction in 2000 catapulted mobile phone usage forward and made it possible for companies like Apple to imagine the first iPhone, launched in 2007.
Without significant investment and expansion in global cellular networks—the foundational infrastructure required to bring cell phone technology to every corner of the world—it’s possible that cell phones would never have gained popularity or market share.
Biases and blind spots
So, what hurdle does AI need to overcome to truly become a global technology? While many investors are looking towards power and chips—the critical GPUs that allow AI to perform—they are missing a much more important foundation: data.
Large language models (LLMs)—the backbone of today’s AI—are only as good as the data they are trained on. Unfortunately, data often comes with built-in biases and blind spots.
Consider for a moment that many of the most popular LLMs have been built by U.S. companies and are trained on large, publicly available datasets using online sources like literature, news, social media, and Wikipedia. While expansive, this data is inherently influenced by Western cultural norms, political ideologies, and historical viewpoints. This is a problem if the AI product is meant to be used globally.
It’s a simple truth: Online data tends to reflect wealthier, tech-savvy populations that represent a very small percentage of the world population. As a result, the LLMs powering the most exciting AI are only relevant and working for English-speaking users with regular internet access, but are failing to account for the experiences and realities of the global majority.
The path forward
One solution is stronger AI governance—implementing policies and procedures that actively mitigate biases in AI models and the underlying data they depend on. This has become a growing focus for policymakers and industry leaders alike, aiming to make training data more inclusive and models more reflective of diverse perspectives. Auditing systems for algorithmic fairness is one way to address this.
However, relying on a handful of AI companies to self-regulate has its limitations. Arriving at an industry standard consensus can be difficult, policy adoption can be slow, and enforcement is often inconsistent. We need a broader approach.
Another way forward is for companies to take matters into their own hands by pairing the depth of their own proprietary datasets and domain expertise with the breadth and processing power of existing AI models. By making a commitment to their own data management, companies across industries and regions present a huge opportunity to help improve and expand available data sets. Leveraging new, alternative sources of customer data is core to my company Tala’s thesis on reaching true global scale—and has enabled Tala to efficiently implement AI in its financial infrastructure.
A truly global revolution
One thing is clear: AI is here to stay, and its pace of development will only accelerate. But if we do not address its biases and blind spots now, we risk leaving billions of people out of the equation.
There is hope that the AI industry—from incumbents to disruptors—will recognize the global opportunity to implement AI. Companies must take proactive steps by adopting forward-thinking AI governance, while also leveraging proprietary data to fill in the gaps of the first generation of LLMs. The opportunity starts with global data and infrastructure. We are early enough in the lifecycle of AI to make sure we are building products to revolutionize the entire world, not just parts of it.
Shivani Siroya is founder and CEO of Tala.
