AI may not replace your favorite content, but it might make it more useful.
Welcome to People vs Algorithms #60.
I look for patterns in media, business and culture. My POV is informed by 30 years of leadership in media and advertising businesses.
Sometimes it’s nice to read in the browser.
AI moves to the interface
Commenting on a feverish couple of months in AI, former Microsoft exec and computer scientist Steven Sinofsky makes an important observation: “In the next 6–12 months every product (site/app) that has a free form text field will have an “AI-enhanced” text field. All text entered (spoken) will be embellished, corrected, refined, or “run through” an LLM (large language model). Every text box becomes a prompt box.” The comment got me thinking about AI’s impact on the media consumption experience. How will AI augment how we read and watch?
It's pretty obvious that AI is going to change the creation side of the content game. The tech is rapidly being built into every writing and office productivity application, content management system, video and music editing suite, etc. There is much hand-wringing about if and how this should happen, including Wired’s declaration this week that their use of AI will be extremely limited, which strikes me as premature and kinda stupid. Times like these seem to benefit the savvy hackers. CNET’s embattled and now departing editor-in-chief, Connie Guglielmo, seemingly agrees, so smitten with AI’s role in media making that she decided to make it her full time gig.
But, I wanted to think through how AI changes the consumption experience. How might it alter the act of reading on the web or app? Of watching? What will it do to ecommerce interfaces? In the intersection of the content and shopping?
Sinofsky makes the point that AI is spreading like wildfire because base technologies of mobile / cloud / social are established — AI integration does not require a drawn out platform reset. Last week’s launch of Open.AI’s ChatGPT API signals the start of the race in earnest. Any third party can now quickly build AI-enhanced applications and fine tune custom models with proprietary data sets. Open.AI has just made leveraging the tech far more cost effective (more below). In short, watch for a cascade of experimentation… everywhere.
Snap's new "My AI" ChatGPT integration is an early example of this distributed innovation, and a logical extension of Snap’s chat-centered experience. “My AI” will offer Snapchatters a reliable robot friend to “answer a burning trivia question, offer advice on the perfect gift for your BFF’s birthday, help plan a hiking trip for a long weekend, suggest what to make for dinner.” Thank you new robot friend.
For the purposes of thinking through media extension and, in particular, where media hits “service” and shopping, Instacart’s proposed ChatGPT implementation is an interesting example to contemplate. "Ask Instacart" will connect a chat query to recipe content and the 1.5 million product SKUs listed by the shopping service. The utility of moving from chat to recipes to product listings then fulfillment offers a good sense of the potential of the technology as it becomes more central to content and commerce interfaces. Here’s a quick demo:
A new dimension to media
Imagine you are reading a story on a web page or app, like the old fashioned kind of story written by a human. How might one augment the experience with AI functionality:
Summarize / rewrite / translate this article;
Compare this content to things recently written on the topic;
Fact check the author’s arguments and show supporting sources;
Deconstruct the bias of the author;
Present opposing sides of the argument;
Show me the best / most relevant related content on the topic;
Give me background on a person mentioned in text;
Summarize how this relates to other things I’ve read;
How would “put notable thinker here” react to this article;
Show more images/videos relating to the topic… or generate new ones.
A piece of content becomes a starting point for a deeper contextual investigation, with little additional user effort. As such, the AI functions as a democratized, free research assistant.
More disruptive to people in the advertising and affiliate ecosystem will be the ability to take the context of media and instantly connect commercial outcomes to it. Imagine a new AI layer sitting on top of media that:
Finds products, prices and availability of any items reviewed in the content;
Gathers what others have to say about any product, service or company;
In a video context, effortlessly reveals what products appear, points to where to buy them;
Creates personalized renderings of me wearing / using the product;
Instantly takes my personal information a fulfills and order, an inquiry, etc.;
Gets a group people to book / buy with me;
The impact will be pretty staggering. Broadly, AI offers the potential to connect the world of content to the world of products and services as a SKU level with zero friction. Consequently, it presents an opportunity to completely rethink digital advertising in the image of what we now call affiliate marketing.
New take, old tricks
Much of this has been contemplated before. A generation of technology providers sought to put a layer on top of content in exchange for advertising revenue share, data access or fees. The list is endless. Disqus and Open Web provided "commenting" functionality to publishers. Medium enabled inline commenting. ShareThis traded social buttons for user data. Vibrant tried to turn "links" into a pop-up video ad product. GumGum put ads on photos. Countless companies sought to connect video with contextual product recommendations.
Mostly, this functionality served to slow and clutter the consumption experience and add another toll booth between advertisers and media owners. One might argue it hastened the shift away from media owners distribution points (websites) to faster, more convenient places to consume media (social media). The question of who controls and exploits middle layers (see this week’s bundling discussion on the podcast) is ever present in the media business — will it be publishers, tool providers, networks, the browser or the OS. AI will represent a new scramble to control a contextual layer, this time one with the promise of much more consumer and commercial utility.
Arguably, the biggest impact of AI on media date has not been around the content itself, but how it’s fed to us. Social media, broadly considered, represented not just a layer on content, but an entirely new interface to it. The personalization of content delivery via algorithms inside of Facebook’s feed was media’s first AI use case. As we have discussed before, new AI chat is the potential end game of aggregation, deconstructing content at a semantic level, an existential problem for content creators. Obviously a one-sided exchange where AI completely disembowels human-made media is not sustainable. All of which makes the application of AI technology against human content a hugely important area for consideration.
If the previous era of content enhancement proved anything, it was just cause you could do something, doesn’t mean you should. In truth, efforts to forge new types of storytelling have not changed media much. Remember Snow Fall? In December 2012 that the media world freaked out when the New York Times debuted the first of a new type of web-based multimedia storytelling. It was an auspicious moment that showed how feature packaging could be more than simple text and pictures. Admittedly, this was not about tools around content but presentation of content itself.
Snow Fall promised to turn the web into a bold new storytelling platform for media companies. One might argue it evolved how ambitious content was presented in a new digital media age. But in truth, not much changed. It was too costly and laborious and mostly, we went back to packing stories the same old way. Today, email has become a delivery mechanism of choice for text content. For good reason. It’s simple, fast, uncluttered and comes to your mailbox. It does one thing very well.
The cost of intelligence
Back to the Open.AI API offering. Today, the service is priced according to the number of "tokens" that move through their API, both in and out. A token is a series of characters that represents a single unit of meaning. It's a bit like a word, but in many cases a world is broken down into parts. In the end, a paragraph is about 100 tokens. The API is priced at $0.002 per 1000 tokens (depends on the service used, but directionally accurate). I don't have hard data on the average number of tokens per session and it obviously varies dramatically by use case, but one might assume it's in the 1000 token range.
If you were to price the service as a "cost per thousand" or CPM, the currency used for media pricing, a thousand users using the ChatGPT API would amount to about a $2 CPM, adding a pretty significant toll to the cost of delivering a media experience to a user. To put this in context, while media prices vary widely, advertisers typically pay a $5 to $50 CPM (depending on environment, audience, ad experience). Adding AI chat to a media experience would, in short, weigh significantly on variable costs. Assuming an enhanced experience would boost session length and depth, add value to the advertising offering or potentially transaction effectiveness, lots of media purveyors will be eager to experiment.
The new link
I suspect this time it will be different because the AI extensions are overflowing with potential. Watch for media's first great AI Snow Fall moment, as ChatGPT is leveraged by storytellers in interesting ways. Odds are it will happen in the next few months. Look for a new cohort of AI-powered tech providers leveraging the adtech playbook to find new ways to exploit the middle.
Considered more broadly, AI augments the "link" as the next important building block of the internet and media. The link is the connection between pages, between places and ideas. Increasingly, the act of digging deeper or manipulating a piece of content will be supported by an AI query. Microsoft's awkward combination of chat and Bing search is the first of many attempts to refactor how we leverage AI to change the web experience. The tech is about to alter the content experience in ways that offer new dimensions to media contextualization, co-creation (human/AI), commerce integration, content/service hybrids and, I’m sure, yet unimagined experiences and business models.
Advertising models will change the most as AI provides better ways to match buyers and sellers around content. Which, given the web's suboptimal relationship with its primary breadwinner and most maligned constituent, may be the best news of all.
Have a great weekend.../ Troy
On the podcast: Alex wants to burn down the ad industry
We get sidetracked on "what is premium media?”, go deep on “bundling” and how AI will change it at a fundamental level. Plus: Alex hates advertising, and I love my dog, Harriet.
The best pod in media, says Harriet. Listen here.
A different grind of a different kind
Fifteen years ago the Go! Team made this and I was smitten.
New goodness below.
The Go! Team are an English six-piece band from Brighton, England. The band initially began as a solo project conceived by Ian Parton; however, after the unexpected success of The Go! Team's debut album, Thunder, Lightning, Strike, Parton recruited band members to play for live performances and subsequent albums. Musically, the band combines indie rock and garage rock with a mixture of funk and Bollywood soundtracks, double Dutch chants, old school hip hop and distorted guitars. Their songs are a mix of live instrumentation and samples from various sources. The band's vocals also vary between performances: while live vocals are handled mostly by lead vocalist Ninja, vocals on record also feature sampled and guest voices.