Say what we will about Bill Gates, the man created one of the most enduring and formidable organisms in the history of computing. By market cap, Microsoft is worth two trillion dollars almost 50 years after its founding in 1975.

Gates, who has seen his share of paradigm shifts, writes:

The development of AI is as fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone.

Decades on, we’re still watching the microprocessor super cycle play out. Even as we’ve hit a wall in the raw speed of desktop computers, every year we push more boundaries in power efficiency, as in silicon optimized for mobile, and in parallelization, as in GPUs. The stuff we can do for realtime graphics and portable computing would have been flatly impossible a decade ago.

Every technological advancement in the subsequent paradigm shifts Gates describes depends on this one field.

The personal computer, meanwhile, has stabilized as a paradigm—they’ve all looked more or less the same the last ten years—but remains the bedrock for how people in every field of endeavor solve problems, communicate and create.

In short, Gates is arguing that AI is a permanent transformation of humanity’s relationship not just to technology, but to each other and our drives to mold the world around us.

Look, I’m not the guy who mistakes wealth for trustworthiness. I’m here for any critique of Bill Gates and his initiatives you may want to argue. But on the particular subject of paradigm shifts, he has the credibility of having navigated them well enough to amass significant power.

So as an exercise, let’s grant his premise for a moment. Let’s treat him as an expert witness on paradigm shifts. What would it mean if he was right that this is a fundamental new paradigm? What can we learn about the shape of AI’s path based on the analogies of previous epochs?

Internet evolution

The internet had its first cultural moment in the 90’s, but that took decades of preparation. It was a technology for extreme specialists in academia and the defense establishment.

And it just wasn’t that good for most people.

Most use of the internet required on-premises access to one of a small handful of high-bandwidth links, each of which requiring expensive infrastructure. Failing that, you could use a phone connection to reach it, but you were constrained to painfully narrow bandwidth.

For most of the 80’s, 9600 kilobits per second was the absolute fastest you could connect to any remote service. And once you did, it’s not like there was even that much there. Email, perhaps, a few file servers, usenet.

By 1991, speeds crept up to 14.4 kilobits per second. A moderate improvement, but a 14.4 modem took several minutes to download even the simplest of images, to say nothing of full multimedia. It just wasn’t feasible. You could do text comfortably, and everything else was a slog.

America Online, Compuserve and other online services were early midwives to the internet revolution, selling metered access over the telephone. Filling the content gap, they provided pre-web, service-specific news, culture, finance, and sports outlets, along with basic social features like chat and email.

Dealing with the narrow pipe of a 14.4 modem was a challenge, so they turned the meter off for certain tasks, like downloading graphics assets for a particular content area. This task could take as long as half an hour.

In short, the early experience of the “internet” was shit.

Despite this, these early internet services were magic. It was addictive. A complete revolution and compulsion. The possibility of new friends, new conversations, new ideas that would have been absolutely impossible to access in the pre-internet world. The potential far outstripped the limitations. People happily paid an inflation-adjusted $6 hourly for the fun and stimulation of this experience.

Interlude: the web, a shift within the shift

Upon this substrate of connections, communication and community, the web was born.

What was revolutionary about the web itself was its fundamental malleability. A simple language of tags and plaintext could be transformed, by the browser, into a fully realized hypermedia document. It could connect to other documents, and anyone with an internet connection and a browser could look at it.

Instead of brokering a deal with a service like America Online to show your content to a narrow slice of the internet, you could author an experience that anyone could access. The web changed the costs of reaching people through the internet.

So, yes, a business like Amazon.com, fully impossible without the internet and the web, could be built. But everyday people could learn the basics of HTML and represent themselves as well. It was an explosion of culture and expression like we’d never seen before.

And again: the technology, by today’s standards, was terrible. But we loved it.

Bandwidth and the accelerating pace of the internet revolution

Two factors shaped the bandwidth constraints of the early consumer internet. The telephone network itself has a theoretical maximum of audio information it can carry. It was designed for lo-fi voice conversations more than a century ago.

But even within that narrow headroom, modems left a lot on the table. As the market for them grew, technology miniaturized, error-correction mechanisms improved, and modems eked out more and more gains, quadrupling in speed between 1991 and 1998.

Meanwhile, high-volume infrastructure investments made it possible to offer unlimited access. Stay online all night if you wanted to.

As a consequence, the media capabilities of the internet began to expand. Lots of porno, of course, but also streaming audio and video of every persuasion. A medium of text chat and news posts was evolving into the full-featured organ of culture we know today.

Of course, we know what came next: AOL and its ilk all withered as their technology was itself disrupted by the growing influence of the web and the incredible bandwidth of new broadband technologies.

Today we do things that would be absolutely impossible at the beginning of the consumer internet: 4K video streaming, realtime multiplayer gaming, downloads of multi-gigabyte software packages. While the underlying, protocol-level pipes of the internet remain a work in progress, the consumer experience has matured.

But maturation is different from stagnation. Few of us can imagine a world without the internet. It remains indispensable.

What if we put that whole internet in your pocket?

I’ve said plenty about mobile already, but let’s explore the subjective experience of its evolution.

Back in 2007, I tagged along with a videographer friend to a shoot at a rap producer’s studio. Between shots, I compared notes with this producer on how we were each enjoying our iPhones. He thought it was cool, but it was also just a toy to him.

It hadn’t earned his respect alongside all his racks of formidable tech.

It was a reasonable position. The iPhone’s user experience was revolutionary in its clarity compared to the crummy phone software of the day. Its 2007 introduction is a time capsule of just how unsatisfied the average person was with the phone they grudgingly carried everywhere.

Yet, objectively, the 2007 iPhone was the worst version they ever sold. Like the early internet, it was shit: no App Store, no enterprise device management features so you could use it at work, tortoise-slow cellular data, English-only UI.

It didn’t even have GPS.

But look what happened next:

• 2008: App Store, 3G cellular data, GPS, support for Microsoft Exchange email, basic enterprise configuration
• 2010: High-density display, front-facing camera for video calling, no more carrier exclusivity
• 2011: 1080p video recording, no more paying per-message for texting, WiFi hotspot, broad international language support
• 2012: LTE cellular data

In just five years, the iPhone went from a neat curiosity with serious potential to an indispensable tool with formidable capabilities. Navigation, multimedia, gaming, high-bandwidth video calls on cellular—it could do it all. Entire categories of gadget, from the camcorder to the GPS device, were subsumed into a single purchase.

None of this was magic. It was good ROI. While Apple enjoyed a brief lead, other mobile device manufacturers wanted a slice of the market as well. Consumers badly wanted the internet in their pocket. Demand incentivized investment, iteration and refinement of these technologies.

Which brings us to Pattern Synthesis Engines

I think AI is a misnomer, and causes distraction by anthropomorphizing this new technology. I prefer Pattern Synthesis Engine.

Right now, the PSE is a toy. A strange curiosity.

It has struggled to render fingers and teeth, when creating images. In chat, it frequently bluffs and bullshits. The interfaces we have for accessing it are crude and brittle—ChatGPT in particular is an exercise in saintly patience as its popularity has grown, and it’s almost unusably slow during business hours.

Still, I have already found ChatGPT to be transformational in the workflow of writing code. I’m building a replacement for this website’s CMS right now, and adapting an excellent but substantial open source codebase as a starting point.

The thing is, I’m not particularly experienced with JavaScript. In particular, I’m often baffled by its syntax, especially because there are often multiple ways to express the same concept, they’re all valid, and different sources prescribe different versions.

Now, when I get tripped up by this, I can solve the problem by dumping multiple, entire functions in a ChatGPT session.

I swear to god, last night the machine instantly solved a problem that had me stumped for almost half an hour. I dumped multiple entire functions into the chat and asked it what I was doing wrong.

It correctly diagnosed the issue—I was wrapping an argument in some braces when I didn’t need to—and I was back on my way.

Remember: shitty technology that’s still good enough to capture our cultural imagination needn’t stay shitty forever.

So if we extrapolate from these previous shifts, we can imagine a trajectory for PSEs that addresses many of the current issues.

If that’s the case, the consequences and opportunities could indeed rank alongside the microprocessor, personal computer, the internet, and mobile.

Fewer errors, less waiting, less brittleness

The error correction is going to get better. Indeed, the Midjourney folks are thrilled to report they can make non-terrifying hands now.

As PSEs evolve, we can imagine their speeds improving while the ways we interact with them grow more robust. The first iPhone was sluggish and the early experience of the consumer internet could be wrecked by something as simple as picking up the phone. Now these issues are long in the past.

Persistent and pervasive

Once, we “went on the internet.” There was a ritual to this, especially in the earliest days. Negotiating clearance for the phone line with various parties in the household, then initiating the phone connection, with its various squeals and chirps.

In the broadband age, the network connection became persistent. If the cable modem was plugged in, the connection was there, steady. Today, thanks to mobile, we are for better or worse stuck online, tethered at all times to a network connection. Now we must make an effort to go offline, unplug.

Today PSEs require significant hardware and computation, and exist centralized in the hands of a few players. As investment makes running them more efficient, and hardware develops optimized for their execution, we can imagine a day in the future where specialized PSEs are embedded more closely in our day-to-day activities. Indeed, for many business applications, this capability will be table stakes for adoption. At a minimum, large companies will demand their own, isolated PSEs to ensure they aren’t training a competitor’s data set.

Internet of frauds

With rapidly improving speech synthesis, plus the ability to construct plausible English language content based on a prompt, we are already seeing fraudsters using fake voices to scam innocent people.

Perhaps most daunting about this is the prospect that, in time, the entire operation could become automated. Set the machine loose to draw associational links between people, filter out the ones where you can generate a voice, and just keep dialing, pocketing the money where you can.

PSEs bring the issue of automated spam into whole new domains where we have no existing defenses.

Digital guardians

It has always struck me that the mind is an information system, but that unlike all our other information systems, we place it into networks largely unprotected. The junky router you get with your broadband plan has more basic protections against hostile actors than the typical user of Facebook or Twitter.

PSEs could change this, detecting hateful and fraudulent content with greater speed and efficiency than any human alone.

Gates describes this notion as the “personal agent,” and its science fiction and cyberpunk roots are clear enough:

You’ll be able to use natural language to have this agent help you with scheduling, communications, and e-commerce, and it will work across all your devices. Because of the cost of training the models and running the computations, creating a personal agent is not feasible yet, but thanks to the recent advances in AI, it is now a realistic goal. Some issues will need to be worked out: For example, can an insurance company ask your agent things about you without your permission? If so, how many people will choose not to use it?

The rise of fraud will require tools to can counter with the same economics of automation. Without that, we’ll be drowning in bullshit by 2025. These guardians will be essential in an information war that has just gone nuclear.

Learning and knowledge will be transformed

I’m a first-generation knowledge worker. I grew up working class. If my mom missed work for the day, the money didn’t come in.

My economic circumstances were transformed by microprocessors, personal computers, and the internet. I’ve earned passive income by selling software, and I’ve enjoyed the economic leverage of someone who can create those outcomes for venture funded startups.

What made that transformation possible was how much I could learn, just for the fun of learning. I spent my childhood exploring ever corner of the internet I could find, fiddling with every piece of software that showed up in my path.

It was a completely informal, self-directed education in computing and technology that would never have been possible for someone of my socioeconomic status in earlier generations.

Gates takes a largely technocratic view of PSEs and their impact on education:

There are many ways that AIs can assist teachers and administrators, including assessing a student’s understanding of a subject and giving advice on career planning.

But to me, the larger opportunity here is in allowing people who don’t learn well in the traditional models of education to nonetheless pursue a self-directed course of study, grounded in whatever it is they need to know to solve problems and sate their curiosity.

Forget how this is going to impact schools. Just imagine how it will disrupt them. It goes beyond “cheating” at essays. Virtuous PSEs will create all new paths for finding your way in a complex world.

The challenge, of course, is that anything that can be done virtuously can also be turned toward darkness. The same tech that can educate can also indoctrinate.

Cultural and creative explosion

There is now a messy yet predictable history between the technologists building PSEs and the creatives whose work they slurped up, non-consensually, to train the datasets. I think the artists should be compensated, but we’ll see what happens.

Nevertheless, I can imagine long term implications for how we create culture. Imagine writing Star Trek fanfic and having it come to life as a full-motion video episode. You’ve got hundreds of hours of training data, multiple seasons of multiple series. How sets and starships look, how characters talk.

It’s just as complicated a notion as any other we’ve covered here. Anything fans can do, studios can match, and suddenly we’re exploiting the likenesses of actors, writers, scenic artists, composers and more in a completely new domain of intellectual property.

This one is far beyond the current tech, and yet seems inevitable from the perspective of what the tools now.

Still: what will it mean when you can write a TV episode make a computer spit it out for you? What about games or websites?

We’re going to find out sometime. It might not even be that far away.

A complicated future

Merely on the downsides, it’s easy to grant Gates a premise placing automated pattern synthesis on the same level as the internet or personal computing. These technologies created permanent social shifts, and PSEs could do much the same.

Nevertheless, there’s also serious potential to amplify human impact.

I’m allergic to hype. There’s going to be a lot of bullshitters flooding this space and you have every reason to treat this emerging field with a critical eye. Especially those hyping its danger, as a feint to attract defense money.

Nevertheless, there is power here. They’re going to build this future one or another. Stay informed, stay curious.

There’s something underway.