ChatGPT is a disruptive technology. But it is not just about engineers being out of a job as Henrik Kniberg explains in this well written post. Sure, engineers who write boilerplate code in mainstream languages will need to look for a new job. Just as FORTRAN and COBOL programmers had to do in the 1990ies. Or as Cloud changed the way software is created and hosted in the 2010s. Smart engineers pick up new skills, languages, tools and technologies all the time. This is what humans excel at. This is not the main issue.
The real disruption comes from how AI tools change the flow of money. Who are the new gatekeepers? Which existing business models loose out? These are the questions we should ask. The answers to these questions will help us understand the structural changes taking place.
At the end of this post I offer my predictions for winners and losers. Skip ahead if you want. Read on as I share my reflections and observations that drove me to these conclusions.
Is good-enough good enough?
ChatGPT provides good-enough answers, not correct answers. This will work in many cases, just like looking up something in Wikipedia or on Stack Overflow provides good enough answers for most of your needs. Let’s be honest: For most of the things we humans do, good-enough is sufficient. For some tasks there are better tools. I’d still trust a compiler to generate correct code over ChatGPT and I will use my calculator over ChatGPT for basic arithmetics. You will need humans to validate the output. You need humans for those cases where good-enough is not good-enough.
Your input is your IP
Prompts is the new source code. What your enter is your IP. Who creates it? Who maintains it? Who has access to it? People who can create and validate prompts will be in demand. You need to think about which AI providers you trust to consume your precious IP. If your prompts contain sensitive information, think twice before sending them to a standard AI tool over the internet. Enterprise requirements comes at a premium price.
Humans provide the data
ChatGPT is trained on open data created by millions of volunteers. Wikipedia, Stack Overflow. Content created for free by people for people, now consumed by AI crawlers that turns it into a business. Will these communities continue to let AI crawlers tap into this wealth of information? Elon Musk was quick to act on this threat on behalf of Twitter and to increase the price for API access to tweets. As Twitter itself makes money from content provided for free by millions of volunteers, the irony is striking.
Just as social media is only as good as the content the users upload for free, so are AI bots dependent on the data sets it learns from. Communities that provide quality content may decide to charge for access. Or as fewer people write boilerplate code and look up answers on Stack Overflow, there will be fewer answers for AI crawlers to learn.
Leading the race comes at a premium
It takes time to train an AI. ChatGPT is not up to date with the latest events in the world. If you need an AI that considers the latest news, you will pay a premium price. Bloomberg provides share ticks for free after 15 minutes. To get data faster costs money. Companies try to generate alpha by being fast to change tweets into trades. However, an AI that only considers the latest news is at risk of being manipulated or tricked by bad input. This is why flash crashes happen. A trading algorithm that knows enough about the past can protect you from these events (and even profit from them).
OpenAI shareholders. Obviously. ChatGPT is not free to use and while the cost is based on usage, the pricing model is decided by those who control OpenAI. It can and will change in the future. This is not a charity. Like Google, Amazon, Facebook, the investors are in for the long haul.
Premium product providers. Those who provide AIs that can beat ChatGPT in niche markets. Areas with special needs for accuracy, speed, secrecy and data protection. Stock markets. Military and intelligence. Who will they be?
Data center owners and hardware providers. Those providing the compute power required to train the AIs.
Engineers. Instead of spending hours writing boilerplate code, they get to spend their time understanding the business problem and solving the core problem.
Technology companies that are too slow to adapt. There will be a drop in demand for developer licenses and training in traditional tools. Time to review pricing models. Interesting to see how quick Bing took a 10% share from Google Search.
Society may lose if we let ChatGPT become a new monopoly in how we apply IT to solve business problems. I think this is unlikely. Let’s instead celebrate that we have one more tool in our toolbox that allow us to solve certain problems faster.
As a post script, I leave you with a final thought:
The Electric Monk
All this talk about AI reminded me of the Electric Monk from Dirk Gently’s Holistic Detective Agency by Douglas Adams. Written in the 1980s, Adams reasons that just as people bought video tape recorders so they didn’t have to watch all the tv shows that were broadcasted, in the future people will buy electric monks to believe things for them so they didn’t have to believe all these things themselves.
With today’s endless flows of information and disinformation, the time is right for electric monks. Maybe ChatGPT is an electric monk? A Google search shows you top ranked links that you then need to review and evaluate for accuracy and trustworthiness and usefulness. ChatGPT is the perfect tool for the lazy brain: It reads these links plus millions more and summarises everything in a short memo.
Soon we don’t even need to read the memo, just knowing that we can get it any time we want is comforting enough that we can keep on sharing cute pictures of our pets and kids on the internet. For electric monks to click, like, and subscribe to.