GenAI as automobile for the mind, and exercise as the antidote: A metaphor for predicting GenAI’s impact

Some of you may remember the Apple ads that emphasized the computer as a “bicycle for the mind.”

(From https://folklore.org/Bicycle.html)

GenAI is not like a bicycle for the mind. Instead, it’s more like an automobile. I’m finding that comparison to be useful in thinking about how GenAI may impact our world.

A bicycle extends our abilities. It allows us to do more with our legs and bodies than we can without the bicycle. The automobile also extends our abilities, but it doesn’t use those abilities. As Paul Kirschner recently wrote, GenAI is not cognitive offloading. It’s outsourcing. We don’t think about how to do the tasks that we ask GenAI to do. As the recent Anthropic study showed, you don’t learn about the libraries that your code uses when GenAI is generating the code for you (press release version, full ArXiv paper).

Automobiles have had an enormous impact on modern society. We can go places and do things that we couldn’t previously. Most of us can’t bike across the US, but many of us drive across it. So, we drive a lot and bicycle less. But there’s a cost — our bodies and minds atrophy if we do not use them.

Ford sold the Model-T as a general tool. (“You can have any color you want as long as it’s black.”). Users made changes to it to adapt it for various conditions and tasks. Today, those changes have led to a wide range of vehicles for a wide range of purposes: sedans, minivans, pick-up trucks, SUVs, and big rigs for hauling materials over long distances. Ford could not have anticipated all those different uses and specializations. They evolved over time.

To gain the benefit of automobiles, we made enormous changes to our infrastructure. We have freeways and driveways, garages and parking structures, gas stations and tire shops. We have changed our environment in order to use the automobile more.

But over time, we have seen those costs. Automobiles (and associated infrastructure, like blacktop parking lots) have had a large, negative impact on our ecology. Neighborhoods were destroyed when freeways were built through them.

We are starting to roll back some of our society’s earlier decisions that favored the automobile. We develop hybrid and electric cars that have less negative impacts on our ecology. In my town, bike lanes are being added, explicitly to choke automobile traffic in order to encourage more biking and less driving. We want people to use their bodies more. Exercise is an antidote to many of the automobile’s ills.

Here’s what I predict based on this comparison:

• We are going to see more specialized forms of GenAI, that we are going to have difficulty imagining today. Already, I am seeing the best learning outcomes from tools where GenAI is built into the tool (like Xu Wang’s work and Barbara Ericson’s). Chat is the Model-T of GenAI. It’ll get used for lots of purposes, but we’ll eventually figure out the specialized forms that will be much more useful.
• We are going to change our infrastructure to enable GenAI. More power plants, more power distribution, more data centers. Eventually, we’ll figure out that we went too far, and we’ll scale those back. But right now, it’s hard to estimate what’s “too far.”
• We will likely overuse GenAI and some of our abilities will atrophy, without a lot of self-regulation and careful consideration of how we use GenAI. Generative AI is a marshmallow test. We will have to figure out that we need to exercise our minds, even if GenAI could do it easier, faster, and in some cases, better.