For AI / Web3: Think Spotify, Not Google / Facebook (/ Yahoo)

There are any number of obvious ways companies will make money via AI (and blockchain and metaverse and….) But the business model(s) that fund the most basic online activity - finding information - are likely to look more like Spotify than Google or Facebook. Lemme ‘splain; but first, a(nother) walk down memory lane….

Web1. This may sound crazy a quarter century after Google was founded, but among the early criticism of the company (and there was a lot of criticism) was that rather than keeping users *on* site like Yahoo and the other portals (yes, I said portals), Google seemed to want to get people *off* of its site as quickly as humanly (and algorithmically) possible. Remember, in the early(er) days of content, the goal was to keep people staring at, listening to, watching, or otherwise engaging with *your* content on *your* media property so that you could show them ads the brands paid you to show them. Google didn’t do that; “worse,” Google didn’t even have banner ads to monetize the large but ephemeral traffic they were getting. We all know what happened though. In short: Google adopted (some say borrowed and some say stole) sponsored search results (AdWords) and got paid a ton for traffic origination and got over a trillion dollars in market cap. 

Web2. In 2012 I wrote about how the early criticism of Facebook (and there was a *lot* of criticism) was analogous to my Google example in that (1) Facebook was originating an increasingly large percentage of Web traffic and (2) Facebook was not using a standard monetization strategy like (its brief flirtation with) banner ads or sponsored search results (which I wrote about here <- but please don’t read because I got it at least half wrong). Facebook compounded the “problem” by cleverly convincing the world to move to mobile (jk) where they had even less of a semblance of a revenue model until they introduced sponsored posts and became a nearly trillion dollar company. 

Web3. And here we are again. (Yahoo ->) Google -> Facebook -> ChatGPT. 

The prediction (for lack of a better crystal ball) is that in 5-10 years our primary source of finding information will shift from Google search and Facebook feeds to bots like ChatGPT and Bard and whatever’s next. To compare, we won’t have good measures like percentage of traffic originating from Google or Facebook or ChatGPT or… and we won’t have measures like time-on-site or clicks or taps that correlate well. And this also throws a curve ball to the model we once didn’t understand (“no banners?!”) and then hated (“sponsorships ruin the integrity of search / feeds!) and then fell in love with (ooh… money money money)… because this isn’t originating traffic. And it’s not necessarily keeping traffic on-site to get ad views. So what is it? Short answer: dunno. BUT… (and, to quote the late Paul “Pee Wee Herman” Reubens, “everybody… has a big but(t)…”) This isn’t going away. Generative AI is here to stay. For better or worse. 

What’s happening now?

ChatGPT and others are ingesting (some say scraping and some say stealing) as much available content as humanly (and algorithmically)  possible and unleashing hoards of tireless GPUs to make sense of its booty before puking it out in (very) nicely packaged CliffsNotes that, when they finally put voice to ChatGPT’s text, will sound either like Kubrick and Clarke’s gratuitously polite HAL 9000 or like Cameron and Schwarzenegger’s genocidal SkyNet. (Salient side note: both tried to murder their creators.)

Meanwhile, content creators (like Sarah Silverman, love her) are suing OpenAI and its contemporaries for the aforementioned ingestion, transformation, and regurgitation of their content because they’re not getting paid on it. (Fair complaint.) And OpenAI released GPTbot to crawl the web and ingest more content to train its models and / but simultaneously shared how to block said bot from sucking the lifeblood from said content creators.

And, well, users are doing what they do: opting for the most effective ways of getting the content they want; i.e., increasingly opting for ChatGPT. So what’s going to happen? Should we look to Google or Facebook for what model will solve this? After all, both companies got content creators very excited for mechanisms to surface content and bring traffic so content creators could make money. Nah. Look to Spotify.

A brief history of time music streaming.

June 1999: The Fannings and Sean Parker start Napster and we all start downloading all of the music we can get our hands on because we know there’s no way this party will last. Key point: this was not just a cheaper (free) way to get music, but a faster / better / more enjoyable way to get music.

December 1999: The Recording Industry Association of America (RIAA) sues Napster because it was facilitating music theft. Nobody stops using Napster. The RIAA would later sue college students too, which was stupid.

2000: Dotcom bubble party! 🎶

March 2001: RIAA wins legal battle and Napster becomes “illegal.” (The party - and the dotcom bubble - are over.)

July 2001: Napster shuts down. (For the time being.) And then nobody listens to music for five years. (Ok that part isn’t true.)

April 2006: Daniel Ek and Martin Lorentzon start Spotify… in Sweden. They convince artists and labels to put music on the service in exchange for sharing ad and subscription revenue. Now everybody is happy and the world is in harmony again.

I may have taken a few liberties with that walk down memory lane (is Apple in the music business?) But the point is: Napster created a new way for people to listen to music that wasn’t only free but better. Turns out people were willing to pay for that or put up with ads. And turns out that it was a really nice way to get money from music consumers to labels and artists.

So wtf does that have to do with ChatGPT? Easy…. Analogous to Napster (and Spotify), ChatGPT created a new way to consume content that users love. But now content creators need to find a way to get excited to have their content used to train these AI models. ChatGPT (et al) will still crawl and scrape and ingest all the content it can, but could pay for access to some content <- that GPTbot could be a nice way to figure out who needs to get paid. Consumers will “pay” for varying levels of access to AI generated content. That $20/mo for ChatGPT seems like a nice start, but I suspect there is a lot more coming to that rather than just better GPT models and more features. Everybody is happy and the world is in harmony again.

(NB: See here for some thoughts on how we’ll track all that content churning and transforming through LLMs and such.)

(Also: pardon my hyperbole above; it’s entirely possible and likely, well, likely that the GenAI chatbots also find revenue models in sponsored content / results, traffic origination <- e,g, “To learn more, see source here….” etc. Just thought this was a good way to look at it too. Carry on….)