In Frank Herbert's Dune novels, there's a historical footnote that plays in one of the underlying themes of the series. There was a time in the history of humanity, still in our future, when humanity dev thinking machines, computers with human-like minds that quickly replaced human labor in almost every sense. In Herbert's vision, this leads to a brief utopia for humanity. But it doesn't last, because very quickly "people became slaves to other men with machines". This leads ultimately to an uprising against the "thinking machines", known as the "Butlerian Jihad", and the subsequent immutable law/commandment "Thou shalt not make a machine in the likeness of a human mind".

Take careful note: It is the people who control the machines, not the machines themselves, who are the problem in Frank Herbert's timeline.
(Also, take careful note, we're talking about Frank Herbert's original DUNE series, not his son's unfaithful bastardized ripoffs.)

For years I have been thinking about this concept. While books and novels and movies and YouTube video essays all talk about the dystopian future of a military AI apocalypse, and maybe a few mention the basic expectation of increasing unemployment as automation replaces jobs faster than it creates new ones, this idea of slavery to those who control the machines always felt more realistic to me, more inevitable. But I could never clearly draw the line from here to there, or adequately describe what it might look like.

GPT is starting to make it more clear.

I hesitate to make any predictions, or especially to put a timeline on them, when I'm not the most expert in the field. There are certainly people more qualified in neural networks and economics than I am. But I do work in a field very closely related to both topics, so what follows is not a wholly ignorant thought experiment.

But this is a thought experiment, not a fortune telling.

With the release of Dall-E, and subsequently ChatGPT (running GPT3.5), last year, the conversation surrounding AI is exploding. A significant portion of this dialogue focuses on how people are using AI to write stories, to write blogs, to write news articles, to illustrate books, to generate code, to write video games. And a lot of this dialogue has focused on how creators in these fields feel threatened. What chance does a human have of getting a writing job if they have to compete with a veritable cascade of AI written material that is just as good as (if not better than) what they could write?

If you ask a lot of pundits (like John Oliver), they'll tell you this isn't as big of an apocalypse as people make it out to be. Writers and artists can use AI as a tool in their workflows, and many actual writers are using GPT as a tool of their own trade. Traditional artists might be displaced by artists who use AI, but they will still be humans creating art. And as AI becomes more capable, it might start to inundate more traditional office jobs - lawyers who use AI will displace the lawyers who don't.

But on the other hand, in the IT field, we're integrating neural network language models into our ticket staging and categorization processes. In effect, "ChatGPT" is now doing the work of a dispatcher. It's not that our dispatcher is using AI to make their job more efficient, the AI is simply doing the dispatcher's job. Dispatcher is a job that no longer exists. That's today.

In the IT field, we're integrating other automation tools to automate common tasks. The low-hanging fruit, as it were, in this case is the basic stuff, like user creations. The kinds of things a tier 1 help desk tech might do. Because this automation can be run - automatically - based on ticket categories, the tier 1 tech is not using AI to make their job more efficient, the AI is simply doing the tier 1 tech's job. That's a job that no longer exists. That's today.

These are jobs that are disappearing but not being replaced. This is what's happening with standard language model neural networks, and GPT3.5 is only enhancing these tools and capabilities.

All of that is coming in the next 6 months.

Yesterday, OpenAI displayed a demo of GPT4. GPT4 can build your website. GPT4 can write your Discord bots. GPT4 can rewrite this essay into a rhyming poem. GPT4 can do your taxes.

I'm a high tier IT systems administrator & engineer at a managed services provider. My job involves a tremendous amount of knowledge, analysis, and critical thinking in order to design, deploy, administer, and support my customers' infrastructure. GPT4 can do my job. Or, more precisely, it will just as soon as someone uses GPT4 to write the code to build a software app that does my job. That person may even be me, hoping to make my own job easier. But once my boss has an AI bot that can scan logs, detect problems, search the internet, identify the correct solution, and apply it to a computer system, I'll be obsolete.

Very soon we'll see more of these desk jobs replaced by AI. There may be a short-term surge in job openings for AI-app developers, but then those developers who make the AI apps that replace my job will write AI apps that replace their own jobs.

This will mean IT, finance, lawyers, doctors, operators and receptionists, software developers, and eventually AI-app software developers will all be displaced by AI.

That's coming in the next year.

What will this look like? Within 6 months or so unemployment will start to rise by a few percent. These will be the people displaced because they didn't adopt AI. The number of people displaced will be fairly high, but many of them will be able to retrain or find new employment, or will learn how to integrate AI and find new jobs. Within the next 6 months after that, as more and more "white collar" jobs begin to get replaced, unemployment will rise even higher and even more quickly. We'll have double digit unemployment by 2024.

But critical to this thought exercise, there is one class of positions that AI won't displace: the executives. The people who make the firing decisions. The people whose paychecks get fatter when they cut "costs". They'll still have jobs.

And what jobs remain will be the jobs that machines can't (yet) do. Up until now automation has slowly been displacing blue collar jobs and leaving white collar workers alone. That's about to flip upside-down and in a big way - because a GPT4 lawyer could advise you in your divorce, but MIT's ATLAS can't (yet) wield a shovel (at least not cost-effectively). This means that white collar workers will only be able to find rainbow-collar jobs - mining, landscaping, HVAC, construction, roofing, remodeling, cleaning, maintenance, food service, sex work. There will be some grey-collar jobs for maintenance techs to support computer/network/AI hardware and infrastructure. But all of those cushy desk jobs people have been doing from home since 2020? They'll be gone.

That's coming in the opening months of 2024.

This is where things start to get real fuzzy, because it's hard to imagine what comes next. AI philosophers call this an "event horizon" or "AI Singularity" (something of a branch off of the more broad "technological singularity" topic). At its simplest, a "singularity" in this sense is a point in time where an event or collection of events occurs that causes so much profound change that it's impossible for us to predict or imagine what comes after. It's a horizon in time past which we cannot see. The change is so dramatic that the possibilities - both good and bad - are literally endless and all (more or less) equally probable. In this case, the singularity is AI.

• Will we stand witness to the end of humanity as we know it as we watch the downfall (most likely at our own hands) of our post-modern civilization?
• Will the displaced workers rise up in violent rebellion?
• Will displaced workers have access to AI themselves and be able to create their own new income opportunities?
• What happens to rent and mortgages while all of this is happening?
• Will governments step in to provide social safety nets, such as UBI or GMI, to displaced workers?
• Will AI allow us to solve societal problems faster, leading to a new technological revolution that ushers in an era of peace and equity?
• Will we find a balance with AI and step into a future where books, stories, art, video games, and all the entertainment we desire can be conjured on-demand just by asking a computer to make it for us?

One possibility, the one that I have been pondering on since I first read Frank Herbert's DUNE novels in 1998, is that all of us workers who get displaced by AI will become slaves to those who control the AI that displaced us. We'll be their landscapers and gardeners, their kitchen maids and trash collectors, their HVAC techs and pool cleaners. We'll be their construction workers building their 5th mansion. And since there will be so many of us fighting for the same jobs, we'll have no bargaining power for wages. We'll work for pennies a day to live by the skin of our teeth in a collapsing suburbia.

Many writers have commented on the inevitable widening wealth gap and increases in inequality, social unrest, and socio-political discord. Those things are happening anyway, with or without AI. The difference that AI will make, though, that's what will be catastrophic. AI accelerates the timeline. This isn't 10 years away, it's happening now. We don't have time, anymore, to debate the best solutions or write our governments to tell them to prepare. We have a year.

A year from now lies a singularity, an event horizon past which we cannot see.

On the other side of that horizon could be a Star Trek utopia, where we simply ask the computer to do what we want it to do, and it does it. You could turn on your computer and say "Hey Bing, make me a video game about Sherlock Holmes" and 10 minutes later you put on your VR headset and suddenly you're surrounded by the dank streets of 19th century London in a fully customized and individualized video game built to your exact specifications. 

Or, on the other side of that horizon, could be slavery for impoverished billions left stranded in the wake of an AI apocalypse that nobody saw coming.

In this little thought experiment, I've proposed that these changes could happen within a year. That timeline is realistic. Perhaps it could take longer, as people and industries are slow to adopt AI and slow to adapt to the changes it brings. It could be shorter, as a highly competitive marketplace pushes faster adoption of a technology that nobody is prepared for.

I hope for the Star Trek future. But that is the hardest future, and it will take all of humanity working together to build it. More on that next time.