Partying like it's 1999

(This will not be one of our usual posts; it is a thought piece on the current state of AI and its potential implications in the long run. This will be the first of a series of texts which we will be publishing on a regular basis. In writing them, we intend to communicate to you, our FollowFox community, our learnings on business history and strategy as they pertain to AI.)

…I was dreamin' when I wrote this

Forgive me if it goes astray

But when I woke up this mornin'

Could've sworn it was judgment day

…The sky was all purple

There were people runnin' everywhere

Tryin' to run from the destruction

You know I didn't even care

…Say, say, 2000-00, party over

Oops, out of time

So tonight I'm gonna party like it's 1999

“1999”, by Prince (song from 1982)

Imagine you are a teenager in 1999. You love music, as do all your cool friends. You own a Disc Man, and a collection of CDs as large as your allowance let you buy.

You also own a computer, which you share with the whole family. A true powerhouse: 500-MHz Pentium III CPU, 256 MB of RAM, and Voodoo graphics card with 8 MB of VRAM for any gaming needs (1999 saw the launch of Age of Empires II, HoMM3, and Unreal Tournament). It also had a Sound Blaster sound card, a 56kbps dial-up modem, and finally – and most especially – a brand-new CD-R, that allowed you to ‘burn’ both audio or data CDs for the first time at home.

Your father also subscribed to AOL; the dial-up connection achieved download speeds of over 15 megabytes per hour if you were online after midnight. As such, you forgo sleep navigating the web, looking for interesting things that required some bandwidth and, well, the cover of night for you to consume. One day, you stumble upon something that changes everything: a website called Napster, that lets you download any song you want for free, in a new format called “mp3” – a format created by a German research institute based on measuring how humans perceive sounds to understand what could be cut without loss in perceived audio quality. At only 3-4 megabytes per mp3 (versus over 50 megabytes of the original audio file), you could download a CD worth of songs in just a few hours, all for you to consume.

You are amazed. You can listen to any genre, any artist, any album you desire. You can create your own playlists, and even burn them on CDs and share them with your friends. You can discover new music and artists every day, all without even touching your meager teenage wallet. The opportunities were endless – you feel like you unlocked a treasure trove, a secret code to enjoying life as a teenager.

But how did this happen? How did music go from being something you had to pay for and own physically to something you could access and enjoy digitally? And what does this have to do with AI?

Music and AI: a tale of two Tech Stacks (and nth-order effects)

The year of 1999 saw the beginning of the story of how music got free. In his excellent book, Stephen Witt tells the fascinating tale of how a series of technological innovations transformed the music industry forever. He traces the origins of mp3 format, the rise of peer-to-peer file sharing, and the complicated impacts of these developments on musicians, record labels, and consumers.

But this is not just a story about music. It is also a narrative about how technologies evolve, combine, and eventually disrupt our lives in ways we can’t predict or control – a story chock-full of lessons that are useful to understanding our present day. Something similar is beginning to happen right now with Artificial Intelligence, and we may benefit a lot from seeing the parallels.

Just like music, AI does not depend on one single technology; it needs a special combination of different technologies, from distinct fields, that work together to create systems that can perform tasks normally requiring human intelligence. These technologies include, but are not limited to: sufficiently powerful floating-point computation; fast, expansive and inexpensive memory; robust optical telecom backbones; distributed file systems that allow the handling of very large datasets; commoditization of internet infrastructure with the likes of AWS and Azure; containerization of processes; as well as, finally, the multitude of algorithmic layers involved in machine learning, natural language processing and computer vision – which is what we’ve been calling ‘AI’ nowadays. Compare that list with the tech stack that began the upending of music in 1999: sufficiently powerful CPUs, high-fidelity sound cards, high-speed modems, CD-R drives, cheap dial-up internet, the MP3 format, and last but not least, Napster.

As in music, the AI tech stack required progress in parallel fields, which slowly matured and coalesced within the last few years. And as was with music, once all these fields achieved a minimum level of performance required, stuff begins changing fast. And changing fast they are.

But I digress. Let’s get back to our music story.

As great as Napster was in 1999, it had a major flaw: Napster’s architecture was centralized, making the network fragile to any troubles of the company that operated it. Indeed, when the heavy hand of music industry justice fell over Napster, the whole system was taken out, leaving all users without a way to continue sharing their files.

A guy named Bram Cohen was set to change that. While Napster’s legal issues were unraveling, Bram quit his job to commit to a new idea: a method to distribute files using a swarm of low-bandwidth peers as the source of a single file, acting in a fully distributed manner. By July 2001, the same month when Napster was shut down and its users were scrambling to find alternatives to it, Bram launched its product: BitTorrent.

BitTorrent became an instant, massive sensation. It was an incredibly smart algorithm that made downloads blazing fast, and the peer-to-peer architecture was structurally more reliable than any other alternative – the more users, the more robust the system became. In a short time, over a third of internet users worldwide became users of BitTorrent. It became the Hydra of digital piracy, threatening to spill over the music industry’s undoing to any other software business.

But wait, there’s more.

Fast forward to 2008. The days of dial-up connection are no longer – cable modems have delivered 100x performance increases in internet connection and grant you download speeds to the order of half a megabyte per second. Those were the early days of the Great Financial Crisis; the world held its collective breath as it witnessed the meltdown of the financial markets under Alan Greenspan’s Fed purview, and lots of people were angry at the Man for abandoning Main Street in favor of Wall Street.

While this was happening, a guy in the cryptography mailing list metzdowd.com believed that, by marrying the peer-to-peer concept of the BitTorrent protocol with a novel method for cryptographic proof he invented, he could build a fully peer-to-peer financial system that could free the world from government control of finances. In an overt homage to BitTorrent, he decides on naming its system… Bitcoin. Thus, beginning the story of how money (presumably) got free.

So let’s recap.

A German research institute studies how humans perceive sounds, and figure out that a lot can be cut off of audio without us noticing it. A compression algorithm based on this study is created, which allows people to use dial-up internet connections to share pirated songs that they burn on their CDs to listen and share. Seeing this happening, some guys (Napster) build a service to help them share the songs, which becomes a huge sensation but is quickly taken out by a lawsuit by record labels. Then comes a dude that creates a file sharing system that is fully peer-to-peer, and thus, immune to being taken out. A few years later, once broadband internet became widespread, another dude marries this file sharing protocol with his ‘Sakamoto consensus’ cryptographic method and invents cryptocurrencies.

The story of how music got free is the origin story of digital money. Who would have guessed that, while downloading a pirated song in 1999?

We are SO early yet – or: The best, and worst, is yet to come

Absolutely no one would have guessed that, of course. And that’s the point.

Technological progress is not a straight line. It is a tangled web of interactions, feedback loops, and unintended consequences. It is a story of surprises, twists, and turns, with heroes, villains, even victims and perpetrators. It is a story that defies prediction and control.

Think about it. If you had asked someone in 1999 what the future of music would look like, you might have heard something like this: “CDs will become obsolete; digital downloads will take over; streaming services will emerge; artists will have more freedom and control over their music; record labels will lose their power and influence”. Whoever offered these predictions would have been an amazing pundit, as they would have gotten all these things right.

But they would have missed something crucial: the connection between music and money. They would have missed how the invention of mp3 format would pave the way for a peer-to-peer file sharing revolution that would challenge the very notion of intellectual property and ownership. And they would have missed how this challenge would be utterly and completely resolved by one Satoshi Nakamoto, who would propose a method to ensure assignable and enforceable property rights on literally anything, even on jpegs and memes.

Let’s think in numbers to put this more in perspective.

As far as value in music industry was concerned, all’s well that ends well. The music industry peaked in revenues by 1999, reaching $25.2 billion worldwide. It then underwent a painful contraction during the subsequent fifteen years, reaching its lowest point of $14.2 billion in revenues by 2014. However, through a number of business model revolutions, from the dollar-a-song days of the iPod to the all-you-can-listen buffet of Spotify subscriptions, it found its footing again and finally achieved a new all-time-high of $28.8 billion in revenues by 2021. The amazing pundit referred above could conceive one scenario where something like this happens – the pundit could say something like “the next 10-20 years will be tough, but the music industry will survive and thrive after that”.

But they would never grasp a reality way more relevant than that. The journey from music file sharing to cryptocurrency took about a decade and created a whole new asset class that, at time of print, exceeds $1 trillion in market cap and trades over $150 billion every 24 hours (if you believe CoinMarketCap). From a financial standpoint, the music industry is nothing but a shadow of the cryptocurrency sector, at least an order of magnitude less valuable, despite sharing a good percentage of its ‘tech stack DNA’.

Today, AI presents itself just as the music industry did in 1999. AI had its ‘MP3 moment’ with the introduction of the Transformer architecture; it had its ‘fast dial-up’ with reasonably cheap GPU machines in the cloud; we are now well within AI’s ‘Napster moment’, with the launch of ChatGPT – the fastest-growing app ever launched in terms of monthly active users. We, AI enthusiasts, are indeed partying like it’s 1999.

AI hasn’t had its ‘BitTorrent moment’ yet, and we haven’t really got a clue as to how that would even look like. My bet is that this will involve the mass introduction of small LLM models, capable of being locally run in consumer-grade GPUs, even while achieving better-than-ChatGPT performance. But who knows? Only time will tell.

The only thing that’s certain is that, however creative we may be, we will never be able to anticipate nth-order effects that will come as a consequence of the events we are partaking in today. The next few years will be weird, and we’ll see the emergence of at least one completely new, revolutionary sector that will generate multi-trillion-dollar wealth – a sector we could never predict today, however clear our crystal ball could be.