How to Make AI That Improves the Wealth Gap -

Income inequality is increasing across the entire world. Those that have the capital seem to keep getting more of it, while those that don’t are getting less every generation. Jesus once told his followers,

To him who has much, much will be given. To him who has little, even what he has will be taken away. -Matthew 25:29

He meant that being diligent in service would be rewarded, but because of his phrasing, the statement gave the name to The Matthew Effect, aka “The rich get richer and the poor get poorer.” This is often explained as those who start with an advantage continue to use it to gain even more advantages. One of those advantages has been ownership of technology.

Every major technology since the Industrial Revolution has increased The Matthew Effect, leading to greater accumulation of capital by the rich and increased exploitation of labor. Is there any hope for AI benefiting everyone?

The Evidence So Far

Replacing

First: the bad news. AI and computers are probably going to replace some categories of jobs. In particular, I fear the transportation sector is going to be hit hard by self-driving cars. Taxi company owners and trucking fleet owners will get the benefits while laying off their drivers.

I believe the reason why AI will replace so many drivers is because AI can do nearly all of the job tasks. Especially with Uber-style taxis, driving from computer-specified place to computer-specified place is the only work to do. Once that task is automated, there’s nothing left for humans. Other jobs that consist of only one well-defined task are in trouble also. Autonomous vehicles rely on neural networks and the recent increase in available compute, but they are not language models. Is generative AI going to cause the same problems?

Assisting

The situation for language models impacting jobs has so far been very different. I think it’s illustrative to look at the most obvious job to replace with AI, translation. Machine translation has been around since the Alta Vista Babel Fish in 1997. AI models have improved significantly since and have been competitive with human translators for 20 years. How has that affected the demand of translators?

According to the Bureau of Labor Statistics, in May 2007 there were 33,680 translators and interpreters working in the United States. In 2023? 51,560. Average wages increased from an average of $37,490 to $63,080. This is a 14% increase after inflation. How can this be?

Domain expertise

As great as language models are, they are limited to well-defined and contained tasks. In this case, who is going to tell the AI what needs to get translated? The name of the restaurant on a menu shouldn’t be translated, for instance. A dish’s name in German may need to be shortened to fit in the same space; AI can shorten titles too, but it’s still the human translator that is telling the AI to do it. This translator is going to be much better at using translation AI than the restaurant owner. In fact, a good translator is an expert user of translation software.

In recent studies and my own experience, domain experts get more benefit from AI in their domain than others. This shouldn’t be a surprise; no domain is made up of singular and isolated tasks. Context, scoping, and review are always necessary, and always domain-specific.

No one can say what the overall labor market will be in five years, but for the last 20 years, translators have adapted extremely well to improving AI. How can we make sure the future has a lot more examples like the translation market and not the transportation market?

Making AI Products Benefit Everyone

The benefits of AI could go either way: to taxi company owners or to individual workers. I believe we have a duty as technologists to ensure technology is as beneficial to society as possible, and that means we should make sure AI benefits individual workers.

If you’re a product builder, design your product for use by individual professionals, not managers or owners. Jobs should not be replaced by AI, but instead accelerated with AI. This makes a more capable and useful AI product anyway, as the professional becomes the best user of the AI.

If you’re an employee, you should work to become an expert at using AI in your domain. If your boss can do your job with AI better than you can, you are in trouble. Similarly, try to improve the job tasks around the main task: gathering, understanding, defining, and communicating. If your job only has one well-specified and automatable task, find a new job!

If you’re in sales and consulting, sell to end users, contractors, and entrepreneurs, or failing that, sell per-seat licenses to businesses. Selling a usage model directly to the Accounts Payable department is too easy to turn into layoffs.

If the technology industry pulls together, this AI wave doesn’t need to have the same effects on income inequality as the last several technology waves. Perhaps we can revert the Matthew Effect to what Jesus’ phrase originally meant: working hard will get you ahead.

The translator example was inspired by the excellent Planet Money article, AI Was All The Rage At AEA 2025. Read it for more about economics and less about product design.