Value in AI Pt 2

Yesterday I wrote about the evolution in how information is created and spreads in human society as a function of technological progress. You can read the full post here. To summarize the assertion at the core of my argument:

The net effect of past technological advance can be read as a discrete set of sudden and powerful step function decreases in the cost of replicating and distributing content. The cumulative set of technological progress - which we’ll shorthand to “AI” for brevity - will have the same net effect on the origination of content.

So the central question is: who are the winners, and what are they doing, in a world where content origination, replication and distribution are all effectively zero?

As always, I have no answers, only half-baked ideas. But it’s a place to start.

Curation

In a world where content can be created, remixed and transmitted with zero friction; where that content can be delivered to an audience of millions or and just as easily an audience of one; and if that content can be delivered directly without filtering, editing or programming by a middling interlocutor, then what might provide differentiated value to the consumer of all this bountiful content?

One place might be taste and taste making.

Though the pool of available content may be infinitely deep and the possible permutations of any instance may be limitless, attention remains limited. Furthermore consumers look to others to contextualize and interpret that content. From this memetic pattern of consumption emerges the role of the taste maker - trusted brands towards which consumers develop kindred sentiment through repeated exposure, each “interaction” building trust progressively. Increasingly these brands are faces and voices more than they are logos and brand marks. In a world where machines confer objects and amorphous entities their own intelligence, this affinity towards faces rather than logos is likely to attenuate.

By a taste maker, do I mean an algorithmic recommendation engine? Yes but no. Algorithms optimize engagement and time on platform but I see algorithmic engines more as an evolution of search (solving the problem whereby the disruptive action pattern of prompting is excessive friction for the user). Algorithms are very good at understanding patterns and serving up more of the same - where “the same” may not have been obvious to a mere mortal - but algorithms are also prone to optimize for a local maximum and stagnate there. Even Tiktok specifically employs human taste makers to discern social trends early and “heat” trends in the algorithm, and in this way put their thumb on the scale of the machine to give it that extra sprinkle of Tiktok magic.

I do believe there is something fundamentally human about taste, something which an algorithm and even a transformer based model cannot capture regardless of its size. Perhaps as a result of advancing the frontier of machine-intelligence we will develop, as a consequence, a more nuanced understanding of “intelligence” as a general term. Where I’d argue today intelligence is used colloquially to mean “advanced stuff only humans can do,” we will find that machines develop along the axis of some types of intelligence (reasoning, information synthesis, others) and there are other types “advanced stuff” that humans will continue to demonstrate comparative advantage - taste being one of them.

When I say taste makers I’m talking about influencers - everything from individual political podcasters to Twitch streamers to Tiktokers. I’m talking about celebrities and visionary founders and anyone who creates a personal brand and attracts an audience and defines a niche.

It seems reasonable to me that with AI entrenching its roots deeper into humanity’s social fabric - and thereby changing society, work and play - we will see the emergence of new types of curator and taste maker roles among humans.

Data

If the future of intelligence is to be commoditized, metered and deployed like transportation is charged by the mile on Uber, and intelligence can be deployed on demand against a problem or situation, then what is the difference between a high quality vs a bad result?

The raw inputs to that intelligence. Specifically I mean the data inputs that intelligence uses to derive its solution.

The obvious first order winners are the existing warehouses of large scale private datasets: the Googles and Apples and Facebooks, the Salesforces, the Bloombergs.

To see beyond current incumbent winners, I wonder about the following: consumers themselves, or solutions that empower the end-consumers of intelligence to own the data themselves and thus apply the metered intelligence themselves; sensors - the eyes, ears and touch of the machines - and the other datasets which bridge the analog world to that of digital bits the machines think in; novel methods that can provide haptic signal back to the machines per inference without the delay of cycle of RL dataset compilation and model training run.

(As an aside: sometimes I wonder whether technologists and investors are thinking too big as they imagine the future they will profit from. Or rather, that by thinking very big they are in fact thinking too small. I’m not yet fully sold that the future shape of our cyborg economy will take the shape of all-purpose-all-powerful-monolith-frontier-intelligence-models and human-shape-all-purpose-robots with robot eyes and ears and hands that neatly slot in and elbow out humans. The future of globalized trade was not the monolithic East India Trading Company but the disaggregated, messy and resilient supply chain of thousands of ports, rails, container ships, cranes, operators, equipment, and all the supporting manufacturers and suppliers they themselves depend on. Small, specialized, diverse and open may yet win.)

Attention

I almost called this section “Distribution,” but I caught myself as it would be confusing to say “distribution is an axis of differentiation” in the breath after saying “distribution costs are zero so what’s next.”

While the width of the pipe to distribute content gets ever wider, and so the cost for content to be transported from A to B continues to decrease, the size of the node at the end of the pipe remains fixed. I am of course referring to attention.

Limitless, low-cost, elasticly scalable intelligence will make it possible to go from concept to product faster and easier and with a smaller team. This will particularly be true if machine intelligence extends its reach into the real world via advanced robotics. No matter how easy and cheap it becomes to create, however, the problem of getting to the consumer for them to buy remains. In fact if it becomes harder to differentiate via production competence then the marketing and distribution to the buyer becomes the greater opportunity for differentiation.

I can make all the products, sites, portals and apps that I want, I can’t sell if I don’t have a way to get the right people’s attention on them! More products on the marketplace only makes that task harder.

Live experiences

There was a time musicians made music and burned that music onto branded CDs and sold them in stores and a really successful musician would made a hit tune and bundle it with some other tunes onto a disc and sell more of those discs than other people and that’s how they became rich. Then came the internet and pirating and streaming. Things have changed. Now, although streaming revenue is appreciated, the purpose for an artist to be on a streaming platform is to have the necessary exposure to listeners and build and maintain a fanbase (ahem - sounds like attention) and the musician makes their margin on ticket sales coming from live shows.

Produced content that incurs zero-replication-cost competes for scale and reach in a crowded marketplace, and so requires dialing competitive outperformance along one or both of two axes - targeting a more tightly defined niche or increasing the value of investment the production. These are the books and movies and shows and even research papers that require progressively more and more upfront investment just to be able to be competitive on the global content marketplace at all. If the producer has onhand a talented and discerning curator of taste then they have the opportunity to outperform competitors on consumer uptake and enjoy outsize profit due to that zero replication cost, but understanding taste is difficult and an upfront risk.

On the other hand, constraining and imposing cost of replication confers opportunity for differentiation and margin!

Live creates ephemerality and means that content might be able to move across location (live sports, news, streaming) but cannot travel across time.

In person confers exclusivity and means that content cannot travel across time nor distance.

Both are valued by humans for creating connection, authenticity, community and unique experiences. Unique is just a colloquial way of saying economically differentiated.

Books -> readings by the author, AMAs

Movies -> screenings with the director

Podcasts -> live recordings and Q&A

Research papers -> symposiums

Music -> festivals and concerts

A bonus dangling thought

One obvious but unrelated question I’ve shifted on over time is “what happens to content if origination is free? what is the meaning of replication if cost(origination) == cost(replication) == zero? will we watch the same movies and read the same books and play the same video games, or will we each exist in our own little personalized content worlds?”

Said in a concrete example: why would I read a book about a British wizarding orphan boy if I could instead read Harry Potter but have myself as the main character? Should there be as many variations of Harry Potter as there are readers?

I began my response by thinking that personalized content was the obvious answer and that before too long we would all be reading our own books and playing our own video games and reading our own news. Now I’m not so sure. The more I reflect and observe the more I think we value community and connection and shared experiences. Part of the value of watching a movie is to have one more shared experience with our kindred human and feeling connectedrather than isolated.

Customization exists in tension with community. The answer might exist somewhere between the two extremes and vary according to use case.

Maybe the future looks a bit like what’s happened with real world events and news, where the reporters and editorializers on real world events have gradually become more entrenched and opinionated to represent the viewpoint of their increasingly defined audience - reading left leaning vs right leaning news spheres can seem alternatively opinionated all the way to alternative-reality - and these content producers deliver their own variation on the original version of the real world events.