Will AI really lower labour costs?

AI-based automation is often seen as a lower-cost alternative to human labour, but Carlos Espinal, investor at Seedcamp, used a historical comparison to explain there are other forces at play than simply lower costs:

“In the case of the 1960s Brasero program [a guest worker program for Mexican farm workers to come to the US], it was just that the local worker was feeling displaced by the immigrant worker. So, they cancel the Brasero program. And the expectation was that by canceling it, all Americans could get their jobs back.
But the problem is that the companies had already optimized their models around the costs and margins associated with a lower cheaper labor costs, [making it difficult to return to a model based on higher labour costs]. Rather, what happens is the industries adapted themselves and shift the way that they use their land for produce that was more optimized towards the kinds of cost they could afford.”

Carlos compares this situation to customer service outsourcing to cheaper work forces, and then to AI-based automation:

We’ve been exporting customer service to countries where maybe English isn’t a first language, and that became difficult. This was frustrating both for customer service representatives and the customers themselves. The cost basis to bring it back to the UK or the US is too high.
We won’t see those jobs return full force, but rather a mass migration to some kind of automated customer service whereby it’s good enough, and works with specialists that deal with [particular] situations.

Rather than accept a generalization that this as a kind of self-evident economic force, I wanted to dig deeper to see why exactly did farmers end up adapting to technology, rather than absorbing labour costs. It wouldn’t be the first time the price of certain crops rose, or the first time a UK bank invested in a local customer service capacity at a higher cost.

Cheaper labour sources didn’t affect the rising cost of labour

Abolishing the Brasero program, thus removing cheaper labour, was supposed to increase farm worker wages. Looking back, a study analyzed what effect ending Brasero actually had.

When comparing states with high use of Brasero labour (>20%) to states with low or no use, we see they all reacted the same. This shows that ending Brasero in 1964 didn’t have an effect on wages. Other things were causing farm labour costs to rise, and reducing the supply of cheaper labour had no effect.

States with high Brasero dependence peformed the same as those with low Brasero dependence, and wage changes we the same for states with no Brasero program at all. From: Immigration Restrictions as Active Labor Market Policy: Evidence from the Mexican Brasero Exclusion. Clemens et al.

Equally, it’s an oversimplification to say that when labour costs drop, industries simply get used to the low costs and find other ways.

After Brasero, production in peanuts, flaxseeds, oats and other crops continued be produced in the same or greater quantities without technological advancements allowing greater yield, all while labour costs were rising.

So what was the difference? What was going on?

How farmers plan crops

External dependencies like weather and crop prices affect farmers profits, so their seasonal planning tends to be conservative.

The 1950’s saw price volatility for potatoes; declining prices for corn and rye; but higher yields per-acre for those because of tractor technology which balanced that out. If you grew rye, you’d earn less per bushel, but produce more bushels from your land, so would earn the same amount.

But at the same time, crops like barley, beans and oats didn’t offer better per-acre yields or higher prices, yet land allocation for them stayed fairly constant.

It’s what you can make out of your land that matters, and labour and higher-efficiency technologies are just two factors in the equation. Other risks often carry heavier weight.

Farmers weigh up risks of certain crops against profits. Mechanisation helped de-risk rising labour costs, but if that wasn’t an option, farmers would simply hire less labor and produce less.

They would play it safe and do what they could with what they had.

Replacement of horses by tractors on US farms. From: Economic History Association

Low additional capital costs

By 1960, a US-wide transformation from work animals to tractors was just finishing. General-purpose tractor technology was mature, and tractor add-ons such as the wheat combine-harvester had become popular. Attaching a tomato harvester to your tractor was a relatively low additional capital cost.

Losing negotiating power

Around the 50’s, advanced in refrigeration technology were affecting supply chains and changed demand and selling prices for farmers. Frozen produce was starting to be popular among consumers, and distributors who invested in cold storage closer to farms had stronger negotiation leverage because they weren’t in a rush to buy, but farmers had to sell now or let their crops rot.

This was part of a shift that had already begun a few decades earlier — from small, family farms that produced a range of food towards fewer, larger farms specialising in specific types of produce. This caused farmers to become more strategic in their long-term crop planning. They were pressured to invest in ways to increase yields, but those investments were often specific to certain crops.

Advances in biology

The shift to machine-harvestable crops started with gasoline-powered tractors (at first, called “traction engines”) in the 20’s, and by the 40’s wheat and corn production was growing because of the combine-harvester.

But in the 50’s other technologies than those purely focused on farm mechanisation emerged. Farmers started shifting to tomatoes because of a new breed of tomato which yielded all at once and at a specific height. This meant tomatoes could be harvested by a specifically-designed tractor-pulled mechanical harvester — no need for human eyes and hands. And this time, the benefits weren’t just for the farmer. These tougher-skinned, flatter tomatoes were easier to distribute and process. These new tomatoes benefited distributors, who were becoming more powerful.

Bruce Hartsough, chair of the UC Davis Department of Agricultural and Biological Engineering, explains that both the tomato strain and harvester were developed closely together:

“The tomato harvester is a perfect example of the shared history of agricultural engineering and biology. It required the parallel collaborative development of tomato varieties that were resistant to mechanical damage, were easily detached from the vine and ripened uniformly.”

Did the farmers adapt to the technology, or the technology to the farmers?

In the case of farming in the 60’s, the issue of labour cost turned out to be misleading. Understanding why farmers shifted, in some cases, to crops that required less human labour, there were a number of factors:

  • How farmers make crop-planting decisions
  • Existing investment in tractors allowing for low additional capital costs for shifting to new machine-harvestable crops
  • Advanced in biology allowing leeway for farmers to shift crops and avoid higher labour costs
  • Strong pressure on farmers to adapt, caused by changes in consumer demand and distributor negotiating power

These were critical factors in the Brasero story — do you think they have equivalents in a modern customer service scenario?

What’s the relationship between lowering human costs and technological efficiency in other industries?

Usually, technology development is described as adapting solutions for users or an industry, but Carlos recognized the inverse:

“We might actually have to shape how we look for business, and how we look for interactions, to be able to adapt to the needs of AI.”

Rather than over-generalise and claim that AI will simply take over by creating cost-efficiencies and force industries to adapt, what other factors do you see when you look at specific industries?

How do investment decisions get made — what risks, capital and marginal costs are at play? What external pressures are on the decision makers? Who has the power? Are existing technology platforms in place that can be easily extended to use AI? Are other technologies assisting a transition to AI?

Do you see the shift to tomatoes in the 1960’s as similar to the 2010’s shift to AI in customer service? What factors matter there?