End of the Story First:
AI carries a strong Western bias, and it is likely to become more Western-centric as technology becomes ubiquitous. LLMs/Chatbots are massive consumers of public data and content, with the internet being the largest single source of their training material. That will naturally overweight US and Western Europe-oriented data from ChatGPT and the like. For international decision-makers, this creates blind spots and inaccurate data when we try to use AI to get even basic information about developing economies. Due to the way both AI and Wall Street operate, Western bias is likely to get much more severe in the near future.
AI Bias is Double-Trouble for American Managers.
There are two problems with this Western bias for international decision-makers. First, you’re getting incorrect, outdated, or incomplete data on developing economies and on overseas counterparties. Chatbots are much better at SOUNDING credible than actually verifying the quality, relevance, or timeliness of their responses. When I use Chatbot4 to research US or global business issues, the results are generally reliable and sometimes downright insightful. But when the analysis deals with Mexican or Vietnamese business issues, the quality of the information drops to the point that I will not use the data with clients.
What might be even more dangerous is that your local competitors have better data on you than you have on them. One of the most exciting parts of AI is the ability to lower expenses, so the adoption of this new tech will naturally accelerate the trend away from pricey expat postings.
When the United States and Western Europe were generations ahead of developing markets in terms of technology and management processes, this wasn’t that serious. Nowadays, however, developing economies have access to the same MBA programs (face-to-face and/or online) as professionals in NY or London. Years of outsourcing have trained an overseas corps of highly trained technicians and the management teams to oversee them. State-directed manufacturers from major emerging markets (i.e., China) tend to send more managers overseas when they expand – giving them better access to real-time data and on-the-ground experiences. Americans coming to Mexico ask about cartel violence. Chinese coming to Mexico ask about customs procedures and regulations.
Western Bias in AI – Roots and Branches.
The way AI is developing, Western bias is going to get deeper and broader. First, look at how LMMs are built. They have to be initialized with a body of public data and then trained & and updated with user-generated data.
Root of Bias: The public data that makes up an LLM training dataset – which comprises 60% of an LLM’s “knowledge” – is from the internet, books, and articles. They use sites like Wikipedia (blocked in China, restricted access in Russia), Twitter (banned in China, Russia, and Iran), Project Gutenberg, Reddit (banned in China, and Indonesia), and government statistics, in addition to social media, books and papers, and news outlets.
New Branches of Bias: 30% of LLM “knowledge” is user generated. Often referred to as the “AI feedback loop,” this means that chatbots are learning from users – their queries, follow-up questions, and reactions. While this may SEEM to favor greater internationalization of AI’s knowledge base, one has to factor in censorship & and restrictions, and national efforts to establish domestic systems. Chatbots are being trained on the internet, and fine-tuned by YOU and other users who tend to be from the US and Europe. This reinforces the Western-centric character of AI responses.
Challenges for International Managers
Unjustified Confidence in AI Results is Growing
The quality of AI’s responses is improving for queries involving the US or Europe. That is giving users a false sense of confidence in the reliability of data overall. This isn’t just a matter of personal behavior, but also company policy. Once AI finds its way into the company budget as a line-item expense, other types of research (company visits, conferences, reports, consultants) will be cut. Decision-makers will come to rely on LLM tools as much as they rely on internet research now, even as blind spots and credibility gaps from non-Western economies grow. If you thought “mainstream media” and fake news were a problem, then buckle up for the brave new world of opaque, mysterious, unaccountable corporate chatbots.
Cultural Bias in AI
The US and parts of Western Europe view free speech and open exchange of information as social and economic virtues. But that is far from universal, and the new wave of AI-powered info tools are going to reflect that.
• China believes in censoring information, and that is going to be reflected in the dataset available about China.
• Russia has a long history of propaganda and seems to feel it is quite effective. Data from Russia will reflect this.
• Mexico and other parts of Latin America have weak protection for journalists, who avoid publishing stories on corruption, narcos, or politicians out of well-founded fears for their safety.
• Europe tends to take a stakeholder model of corporate governance, which is often at odds with global tech firms’ standard operating procedures.
• The US still operates on a shareholder model, which puts profits and benefits to owners above all else. “Big Tech” social media and data policies have been widely criticized, but these existing platforms are far more transparent and accessible than LLMs.
Final Word
If you’re doing cross-border deals, then the Western bias of AI and the general asymmetry of available data is a challenge that you are going to deal with. AI bias is either going to bury you or give you a competitive advantage as you negotiate across borders. That’s because you have an opportunity to get ahead of the pack in several ways.
First, you can train your own chatbots and AI tools with better datasets. This will not be cheap or easy, but it will contribute to a sustainable competitive advantage for your firm. We like to talk about data as the “lifeblood of an enterprise” – but traditional chatbots are full of toxins for international analysis.
Second, you will need a more nuanced approach to determining your data sources. We all love profits and hate high expenses, but information is a lousy place to try to cut corners. If analyst A can come up with a credible answer or response in 3 minutes, it’s going to be very difficult for analyst B to take a week to come up with a 20% better answer. International decision-makers are going to be forced to make some hard decisions about how they get data on new markets or overseas counterparties and competitors.
The challenge – and opportunity – for Western managers, especially US managers, is that Globalism 2.0 is weakening US dominance of global markets, and we have new competitors coming from China, Southeast Asia, India, and other developing economies. They have cutting-edge information about you, but your data on them is at best incomplete and possibly very inaccurate.
Next: The people who fear sudden human extinction at the hands of AI are the optimists.