While I’m not exactly “Team AI Overlord”, I do acknowledge that they raise some good points. If bots’ ascendancy is assured, then I’m willing to go along to get along.
But early adopters relying on AI for business intelligence in Mexico need to be aware of 3 built-in biases that threaten to undermine their analysis. The biases are: The Gap, The Lag, and The Just Plain Crazy. I imagine that the same challenges exist in other developing economies.
The tech world is abuzz about just how quickly the mainstream internet/AI is barreling towards General Intelligence. My excitement is tempered by the fact that I can’t get the expensive version of ChatGPT4 to give me Mexican business headlines that are less than 6 months old. AI is institutionalizing the asymmetry of information in all the wrong ways. Systems are getting faster and deeper – but narrower. Mexico may be one of the places getting narrowed out.
Infotech entrepreneurs in Mexico and LatAm (and I presume other developing economies) are at a crossroads. On the one hand, the opportunity to add value to global systems is huge. On the other hand, global systems have a long history of ignoring or devouring innovation from developing economies. The race is on.
Institutionalized Data Bias – Your New Operating Environment
If you work with developing economies, you are familiar with the asymmetrical distribution of business intelligence. When it comes to news and information, they’ve already seen yours, but you’re not able to see theirs.
Up until now, that has meant operating in the semi-dark until you could develop contacts on the ground and become a knowledgeable insider. But in an AI world, you won’t know what you don’t know. Your AI systems will fill in the gaps with authoritative-sounding nonsense and you’ll be none the wiser until your project blows up or the inspector shows up.
In the US and Europe, the good news is that the systems are getting smarter, and they are inventing garbage less often. In digital backwaters like Mexico, there is less cause for optimism. Instead of improving the data accuracy of responses about Mexico, it seems that AI is just getting more comfortable operating without it.
When working with ChatGPT in Mexico, I’m getting more and more “sorry, we don’t have access”, and that’s with the right plug-ins. Other times, I’m getting data that is far, far too old to be useful. Occasionally, the output is just plain wrong.
The challenge for analysts in Mexico is that traditional methods and tools for gathering business intelligence are disappearing. Once AI creates the new normal for data collection, the old ways (live people reading reports, visiting places, asking questions, analyzing findings) will dry up and blow away.
Mexico as Case Study
1. Mexico as Source of Blind spots.
This one is happening to me now. There are some basic economic questions that are inexplicably difficult to get answers about. Basic pricing data. Straight answers on salary. I’m certain that the information must be out there, but the AI systems can’t seem to find it.
One of my big concerns about working in Mexico (and before this in Viet Nam) is the asymmetry of data. There is so much data available on the US that the problem is that you get overwhelmed. In China you may not trust the data but there are hundreds of voices discussing what the right numbers might. All of that shows up in the AI platforms – both as part of the initial data set and in the ongoing training.
But that’s precisely what’s missing in Mexico. There isn’t a strong tradition of publicly reporting data, and inquisitive reporters are not part of the environment. Doing basic web searches in Mexico is hit or miss, and AI is likely to bake in that deficiency.
I found this out the hard way after spending hours trying to research what I thought was a simple project. I was considering a weekly roundup of Mexican business headlines, focusing on local sources. Just the kind of task a premium AI should be great at.
The closest I got was a reprint of Reuters headlines, with no sensitivity to publication dates. When I asked for more sources or Spanish language input, I got more apologies.
The international media only sees 3 angles to news about Mexico – illegal immigration, drug crime, and resort hotels. It looks like the same biases are finding their way into AI platforms.
2. Mexico as Home for Outdated Data
I recently tried to research real estate prices in Mexico. I needed industrial rental rates in major urban areas around Guadalajara. It’s normal to have trouble finding the prices of specific buildings or sites, but getting general price levels for famous neighborhoods is usually no problem. It’s a widely available statistic.
That’s why I was a little surprised that I was hitting brick walls when searching for a base figure for industrial property. I had to do a little digging — and the number I ultimately came up with seemed a little funky. I know land prices are lower in Mexico – but this was REALLY low.
I must have been using Bing’s ChatGPT set-up because I was able to trace the source of the statistic that caught my eye. It turns out that the bot was quoting a one-year-old report from a commercial real estate firm. The report in question cited a 3-year-old report by another commercial real estate firm — which was no longer available online. I have a finance background and this would not pass muster as usable data. But if I hadn’t done the digging, I would have confidently included the number in a report.
This bad data was revealed in a routine review of the data — but I get the feeling that fact-checking is going to be a casualty of the AI revolution. AIs routinely return outdated results with that supreme machine confidence you never think to question.
3. Mexico As Source of Just Plain Wrong Data.
I like to use a certain statistical dataset. You don’t have to know which one. Don’t worry about it. But it’s what the data people refer to as, “robust”. I’m not saying it’s old, but it’s been confirmed in more than a couple of field trials.
I asked ChatGPT4 to use the data to analyze a problem having to do with Mexico, and the response was just factually inaccurate. The numbers ChatGPT was returning were significantly different from the numbers on the researchers’ own site. The analysis was also dead wrong. And the AI persisted in its erroneous conclusions after I pointed out the errors and requested it to recalculate.
Even when faced with clear contradictions in its own logic, ChatGPT wouldn’t give a satisfactory answer.
Mexico’s AI Crossroads.
Like most business issues in Mexico, AI stands at a crossroads.
The Bad News: In Mexico, you are using outdated, inaccurate, or downright crazy data. Your competitors aren’t, and that is putting you at a competitive disadvantage when negotiating and marketing. You can’t trust potential partners because they have much more information about you and your economic drivers than you have about them. You can’t trust your own AI because its Mexico analysis is unreliable.
The (potential) Good News: You have the opportunity to build out a sustainable competitive advantage. If AI systems are used to add value, then this could be a golden age of innovation for Mexican infotech. But if the sole application of AI in business is to reduce expenses and automate away jobs, then Mexico and other developing economies will be lost and if not forgotten, then at least mis-remembered.
Final Word
AI is a full-on gold rush. The fate of developing economies are big question marks. Places like Mexico may turn into those quirky stories about unlikely heroes who prospered in unlikely ways. Or they’ll be industrial roadkill that couldn’t keep up with the flow on global information pathways.
The Digital Divide is going to result in two tiers of business intelligence.
On the one hand, everything that happens in the land of plentiful data and bandwidth will be analyzed, augmented, and added value.
Meanwhile, those of us toiling away in the wasteland will not only be ignored by the new AI. We will also lose access to our old platforms and tools as AI concentrates the availability of platforms to advertisers or members of the same corporate family. Big MNCs will have the wherewithal to combine and collate – turning paywalls and restrictions into a competitive advantage (since they will be the only ones able to integrate competing systems).
Those of us on the wrong side of the divide will have to read about it on the newsfeeds when the information is released to the public.