AI Productivity Evidence

AI Productivity Evidence

Fed Chair Jerome Powell recently remarked, “AI should lead to increases in productivity.” US GDP growth has been surprisingly resilient lately at a 3.4% real rate. Generally, GDP growth can be divided into changes in population, productivity and debt. The positive take on AI-driven productivity is that there is incredible demand (spending) for AI productivity tools. Successful implementation of these tools can lead to increases in revenue and drops in costs. Also, employee training can be dramatically enhanced by AI-enabled tools. The pessimistic take on AI-driven productivity is that the costs will outweigh the benefits. Of the names to follow in this arena, Eric Brynjolfsson is the optimist, and Robert Gordon is the pessimist.

Automation examples in big companies

Unilever has 50,000 ice cream freezers equipped with machine-learning computer vision systems. These systems scan the current inventory and automatically order. The system can automatically order new ice cream when and where it is needed. Many computer vision models are open source, so building a system like this can be done with freely available tools.

Walmart implemented an employee LLM called “My Assistant” that answers simple questions on employee benefits and helps employees locate items in the store. The goal of this system is to free employees from monotonous and repetitive tasks. Every employee has the same basic HR questions, so imagine how helpful an HR LLM would be for everyone involved. The Sam’s Club division has also initiated a pilot to use computer vision to scan carts as they leave the store so they don’t have to have an employee check the receipts. As early as two years ago, they began rolling out automated robotic floor cleaners that also scan the shelves with machine vision to count the inventory on a nightly basis*. These little automations add up.

I like to balance the apparent negatives with creative, optimistic outlooks. Suppose automation will replace low-level jobs like checking receipts at the door of a Sam’s Club or driving cleaning machines up and down the aisles of a grocery store. In that case, AI can also help retrain people faster than ever for new positions where they will have more impact and probably earn more.

AI accelerates cognitive work although, at this stage, tools like ChatGPT or wrappers that sit on top of OpenAI are only about as good as a talented intern. I like the meme about the programmer who 10xed their debugging workload because they had AI write the code in just five minutes. AI can assist with cognitive workload through data organization, entry, and retrieval, or it can assist with creative work like idea generation and discovery. Another way to say that is better efficiency and more innovation. Many of the AI advantages available today are not as game-changing as we want them to be, but they do add real value in subtle ways, and they are improving.

A handful of other startups that caught my attention in AI are:

• Greyparrot – machine vision sorting of landfill waste to assist in recycling efforts. I remember volunteering to sort recycling as a teenager. This is a better way to approach the problem.
• Canvas – automated robotic dry-wall finishing. Sanding vast areas (like the walls of an airport). It was recently used in a renovation project at SFO airport.
• Monumental AI – automated stone carving for public design, construction and architecture. Here is their Instagram page.
• Shield.AI – AI fighter pilot technology (over $1B raised at a $2.8B valuation)

*If you’re interested in following the automated robotic floor cleaning story, the AI is powered by a private AI company called Brain and the machines are manufactured and sold by a public company called Tennant. Know there are other competitors in this space. Not advice.