Fortune Telling Tech's Next Three Years

No one can predict the future — least of all me. Yet I want to write this post for future me, to look back on. It feels like we have passed the event horizon in technological progress, and there’s no way back. I do sincerely hope I’m wrong on most of these. At any rate, it’ll be interesting to look back on this post later just to see how wrong I was.

The reason this post covers the next three years in technology rather than - say - five, is because I’m two years late to the party. I woke up late while the AI party was already in full swing. To make up for the lost time, all that’s left is the next three years.

Preface

Let me preface this post by saying that I do not think current generation Artificial Intelligence to be actually intelligent in the sense of it being sentient. It is a novel technique for processing large swathes of information accompanied by an extremely natural interface: chat and speech. The technology is the most potent form of accessing information yet, but what truly makes it revolutionary is the ability to just “talk” to it as you would to another person.

Before, a person with a computer was - say - 10 times more efficient than a person without. People who could instruct computers through code would be even 10 times more efficient than that, making them 100 times more efficient than a person that didn’t have a computer. What AI has done is democratise computing. Anyone can instruct a computer to do whatever they please now, and they can do it in their own language.

With that out of the way, here’s how I think this will impact our lives and society in the near future, in no particular order.

1. The Honeymoon Will End

We’re currently in the honeymoon of AI. Models are getting progressively more powerful, there multiple different models on offer, as well as some niche services that combine these services into niche offerings. All of these are easily accessible through the browser, and relatively inexpensive.

Most services offer a free tier that is sufficient for most users. For those that rely more heavily on AI, or extensively use AI for their work, premium plans cost around $20 per month.

The problem here is that these offerings aren’t sustainable long term. The environmental impact of AI is huge, with the Guardian stating that AI could account for half of datacenter power usage by the end of this year. Even saying “hello” and “thank you” to AI costs OpenAI millions in electricity.

What’s more is that even the largest company - OpenAI - is bleeding money. Last year, it operated at a net loss of 5 billion USD. That is with a (likely inflated) 400 million weekly users. It’s very unlikely they will turn a profit any time soon, as Edward Zitron has outlined in much more detail than I ever could.

What we’re seeing is Big Tech’s classic playbook: gain market share at tremendous cost, outlive the competition, then squeeze the users for as much money as possible. There’s no reason to believe it’s any different with AI.

Our devices will have AI capabilities at their core, and won’t be able to function without. Companies will come to rely on AI to do the work rather than hiring humans for their expertise. Software will come to rely on AI models to do the difficult parts. When the time’s right, so too will be the monthly subscription fee.

2. The Free Internet Will Die

We’ve been witnessing the decay of the free, open internet for over a decade now. Black hole platforms like Discord, Facebook, and more recently Large Language Models, have long replaced the freely searchable forums of yesteryear. Websites cost money to run, and without visitors and declining ad revenue, many have gone dark. This platformisation of the internet has been going on for years. AI will prove to be final nail in the free internet’s coffin.

Google Search used to be the gateway to the free internet. Any website could rank — beit at the mercy of Google’s ranking algorithm — and (almost) anything that was searched for, could be found. This is no longer the case with AI consuming the underlying websites, and subsequently regurgitating their content through AI generated responses.

One could argue that Google’s hand was partly forced, as it had been manipulated into peddling SEO-optimised garbage in the years prior, with no clear response from Google on how it would deal with it.

The information these websites contained will live on in the belly of the beautiful beast that is the LLM. Except, rather than being able to search it freely, we are now at the mercy of the tech company that operates it for profit. In the process, it strips the very websites it relies on of their lifeblood. Devoid of visitors and the ad revenue they bring, many will invariably go dark.

3. The Knowledge Economy Will Shift

Our knowledge economies will undergo a radical shift as we transition from a primarily human-centered knowledge economy to a computer-centered one.

Truth is, Large Language Models are far more efficient in analysing vast swathes of information, like legal precedent and relevance, than even the best lawyer in the world. It can do so in a fraction of the time, at a fraction of the cost. Economically, it makes zero sense to have expensive lawyers draft up legal agreements when a computer can do it at almost zero cost.

This notion isn’t limited to just the legal field. The effort to output ratio of an LLM is vastly superior to that of human knowledge workers, be it in the field of law, economics, or engineering. In almost all cases, it will take the human more time to produce the equivalent of what the Large Language Model can produce, than it will for the human to review and correct whatever it is the LLM produces.

Exactly this imbalance will seep into our modern economies in the coming few years. Instead of working on the material itself, knowledge workers will shift to providing high quality input for LLMs to learn from, and subsequently, will be tasked with validating the LLM’s output.

Companies will capitalise on this by shifting from expensive human knowledge workers to inexpensive computer models. The humans that do remain will focus their efforts on validating the output of the computer model, and ensuring it fits the business’ goals.

Conclusion

It’s very likely I’m hallucinating more than the LLMs I’m writing about, and I’ll be wrong on most of this, if not completely. There’s obviously a non-zero chance that there is no AI boom, and what we’re seeing now is just a bubble.

Whichever ends up being true, I think we’re at a period in time where the capabilities of what computers can do in terms of psuedo-reasoning and computer vision will potentially be even more transformational for society than the previous two decades of computing combined.

We should carefully consider what information we contribute to the machine, because like the internet itself, AI is unlikely to forget. As the machine itself becomes more and more opaque, it’s ever less likely that regulatory institutions will be able to keep up, let alone protect the interest of regular people ahead of the unrelenting advance.

Then again, the more optimistically inclined among us might also argue that what we are witnessing is the liberation of mankind from mundane labour through technology.

At any rate, these are very interesting times.