A BLOG FROM CHARLIE CARTER

For two primary reasons, the manufacturing industry in the United States shrunk significantly in the last 50 years. First, as globalization took root, it became cheaper to manufacture in low-wage developing countries. Second, technology slowly and steadily automated more and more manufacturing processes. The implications for the communities in “middle America” that were traditional manufacturing hubs were significant – job and opportunity loss, economic decline, and rising drug abuse are just some of the symptoms. Beyond these tangible losses, blue collar workers have lived for generations with the their jobs – and their standard of living – under threat. (Bruce Springsteen (the Boss) in many ways built a career tapping into this anxiety).

The economic gains in our economy over the last 50 years were largely capture by “knowledge workers”. The data supports what we’ve widely assumed — a college degree was, with of course exceptions, an important component of job and income security.

Now artificial intelligence (AI) is progressing so rapidly, the Boss of this generation may sing songs about the plight of the knowledge worker. AI has largely been an unfulfilled promise for over 50 years. There have been various moments of excitement, followed by “AI winters”. We are entering another hype cycle, but this time may actually be different. One example is GPT-3, a neural network machine learning model trained using massive amounts of internet data to generate any type of text. Using GPT-3, a small amount of input text generates large volumes of output text that appear to be written by a human. GPT-3 is a language prediction model. It takes a small amount of input text and, having been trained on the vast body of internet text to spot patterns, predicts what the most useful result will be.

Already, incredibly valuable software applications are being created using GPT-3. For example, Jasper generates very good adverting copy and visual images based on a small amount of input text. Other well-known use cases include writing software code and gaming applications. And, other less know use cases are no doubt being developed. As a (recovering) transactional lawyer, I’m sure these models can be used to write at minimum very good initial drafts of contracts and litigation pleadings.

GPT-3 is a proprietary applications, and is expensive to use. But, open source competitors are rapidly being developed, including Stability AI, which just announced a big financing. As more competitors develop models, the cost to the user of the models will go down, opening up these models to more creators, and we should see a Cambrian explosion of new and useful applications

For now, models like GPT-3 largely enhance human performance, making copywriters, coders (and even lawyers!) more efficient and effective. And, perhaps that remains the paradigm for the foreseeable future. But, full automation may come more quickly than expected. Just as machines have replaced many factory and similar jobs, software may replace many white collar jobs. Indeed, when the Boss of this generation writes songs about the lack of jobs for knowledge workers, he or may only need to write the first few lines of the song.