Advertising and AI Don’t Mix

Who does your AI work for?

At the beginning of another technological era we have an opportunity to reset the incentive architecture and design a system that promotes human flourishing rather than the harvesting of human attention.

Last month Anthropic posted a clutch of cute videos that effortlessly reinforced their stance that an ad-driven business model to monetise AI represents a pretty dystopian future. They were straight out of a black mirror episode.

[embed]

Apart from being a masterclass in how to land a complex point efficiently, these videos deal with a topic of deep importance. The history of the internet is in many ways the history of advertising as a dominant commercial model…. It didn’t have to be that way — other business models were, and are, theoretically possible — but there are some strong reasons why it was. There are also some pretty profound problems that this has led to and we stand at a point in history where we have an opportunity to choose a different model for how the next phase of our technological progress gets monetised with wildly superior outcomes for pretty much everyone.

Why did we settle on the ad-based model in the first place?

There are two core explanations for advertising’s dominance. The first is that advertising is a natural fit for the internet in that it keeps things fundamentally open:

1. Users can traverse the breadth of the web unhindered by paywalls
2. Content creators can reach the widest audience, because their content is ‘free’
3. Advertisers can reach customers they’d never have been able to reach otherwise

As Ben Thompson puts it:

The original web was the human web, and advertising was and is one of the best possible ways to monetize the only scarce resource in digital: human attention. The incentives all align.

The second explanation is that at the time, we were technologically unready to implement more direct monetisation models…. The basic story hinges on one of the core patterns of the early internet which was (like many innovations) one of unbundling and rebundling. For example, physical newspapers and magazines were replaced by individual articles from diverse sources, and these were duly rebundled into personalised feeds. The direct purchase of a physical newspaper or magazine made sense with the payment tech we had at the time (a single small but decently sized transaction). However, we couldn’t handle the lots of tiny transactions you’d need to directly purchase your morning reading article by article due to the structure of payment fees and limitations on transaction volume that then existed. At the time of their creation, (as told by one of their creators, Marc Andreessen) micropayments could not be built into browsers due to the inertia of Visa and Mastercard who dominated payment infrastructure at the time.

The internet’s original sin

And so we found equilibrium in The Internet’s Original Sin… the advertising business model became dominant, unlocking vast economic value and powering the growth of the web until today. It’s been incredibly successful commercially and has delivered some noble benefits in terms of the democratisation of information — advertising is the only model that ensures universal access… whereas subscription models, for example, would surely entrench inequality. Nevertheless, it has also set the conditions for the surveillance economy that is its logical extension.

As far back as 1849 Thomas Carlyle wrote that advertising, or ‘puffing’ as he called it, deserved to be taxed out of existence. No doubt he’d be horrified by the fully automated and industrialised advertising systems we have built today…. but what specifically are the drawbacks and dangers of such a model?

Thomas Carlyle

What’s wrong with advertising?

For a start — people don’t like it…. Since the dawn of the internet advertisers have been in a steady arms race with users. As Maciej Cegłowski describes it…

One of the first banner ads had a click-through rate of 78%. That’s mind-boggling. Do you know what two words were on that banner ad?

“Shop Naked!” CLICK!

But banner ads turned out to be like poison ivy. People clicked them once, and learned never to touch them again. The industry collapsed in a flurry of pop-overs, pop-unders, and shame.

Courtesy, Maciej Cegłowski

Eventually, Google discovered you could serve little text ads in context and people would click on them…. At least for a while, but then we all got smart to promoted links, sponsored results and the like. In general the pattern consisted of advertisers continually innovating to put ‘impressions’ into people’s minds, and people continually learning to tune out or indeed block out the nuisance….

Eventually the big social media networks showed up and evolved the model once more, wrapping ads not in useful content, but in infinite feeds of irresistibly addictive digital junk. Worldwide internet users average 2h and 20 mins per day on social networking. The incentives that have led to this have been covered effectively by many writers and film makers, with perhaps the most well known being ‘The Social Dilemma’, directed by Jeff Orlowski and featuring many of the clearest voices on the topic.

[embed]

Facebook, the first large social platform did indeed begin as a way of connecting human beings together, establishing a digital overlay to genuine human communities. I remember it launching at my university in its early days and it was magical. The industry still refers to itself as social media, but today’s more sophisticated platforms have largely shed any pretence of being a mirror of genuine human society. TikTok is a pure addiction machine where the construct of friends and connections itself has been dispensed of in favour of unmodulated virality.

“Show me the incentive and I will show you the outcome” — Charlie Munger

Without rehashing the complete narrative of what I’d suggest should be compulsory viewing, the fundamental difficulty boils down to incentives taken to their logical extremes. Social platforms monetise themselves via advertising. Revenue is a function of the number of ads per day per user x number of users. Which in turn is a function of how many hours a day they can keep users scrolling. The primary objective then is how to keep users glued to the feed — and the industry continues to innovate in this direction. The result is something that behaves less like a digital utility, and more like an addictive digital narcotic. From the design of the alerts, icons and interfaces to the personalised concoction of content, our best minds and our most powerful technology has been directed towards creating and sustaining addiction.

What’s the harm?

Spending hours a day scrolling through a stream of disconnected harmless digital confections would be one thing — a waste of time, primarily. But it turns out that much of the cocktail of content we’re served up can be genuinely harmful, albeit sometimes through 2nd or 3rd order effects. What’s notably missing from the incentive set above is anything to do with what we might broadly describe as ‘truthfulness’. As a brief refresher of a well-documented phenomenon, what most users find most engaging is often in direct opposition to truth… fake news spreads 6x faster than real news, conspiracy theories you’re susceptible to will be introduced and then reinforced, and outrage (achieved by whatever means) seems to be the most effective emotion to illicit if you want to capture and hold someone’s attention.

The midwit meme applied to itself

The departure from anything that systematically anchors a person’s information diet to something resembling the truth is clearly unhelpful in many ways. The harms appear in numerous domains but perhaps most clearly in healthcare (anti-vax for example) and politics (all manner of extremism and populism)…. And politics and health are in many ways upstream of a lot of the rest of life.

What makes this even more nefarious is the observed phenomenon of social algorithms not just serving up persistent stream of provocation, but of systematically walking us gradually and imperceptibly from moderate positions towards the extremes — incremental persuasion campaigns leading us step by step down whatever rabbit hole we’re most susceptible to, and leaving us in wracked disharmony within our echo chambers. It starts with a little mistrust of mainstream media, pharma industry scepticism and legitimate discrimination grievance, and ends with QAnon, anti-vax and jihadism….

Social media radicalisation pipelines from Anthropic’s Claude

From a mental health perspective, we’re left anxious, envious, scattered and insecure with the most vulnerable tragically going on to cause physical harm to either themselves or others. Does anyone leave a sustained engagement with modern social media with their dopamine circuits unscathed?

And what about the opportunity cost? American teenage girls average around 2 hours per day on TikTok alone, out of over 5 hours across all platforms. Between the ages of 13–20 that is an entire waking year of your youth spent… scrolling TikTok….

Has a generation of our most talented technologists spent its effort on perfecting the science and execution of harvesting adolescent consciousness?

An advertising-funded AI ecosystem would not merely repeat the mistakes of social media — it would universalise them.

This time we can’t afford to become the product

The challenge is that the coming AI wave amplifies all of this. AI is a bigger deal than social. It is so general and so powerful as a technology that it’s likely to become our primary window to reality — its utility is such that I’d expect it to supplant many of the existing feeds, apps, sites, etc that we would normally use to view and understand the world around us beyond our immediate physical line of site. It will mediate some large fraction of our information diet. And because of this breadth it will be very difficult to turn away from. For this reason I’d argue we should want this window to be as transparent as possible — incentives that drive the distortion of this lens should be undesirable for anyone who thinks the truth matters and humanity’s species-spanning attempt to apprehend reality is of value.

AI Mode in Google

By virtue of being a general purpose technology who’s breadth of application is arguably infinite, it will also be very difficult to avoid. It will be woven into the fabric of everything around us, often invisibly. Your favourite social app occupies very narrow physical real estate in your life — it’s a single app on your device. AI will be infused into almost everything, and if it ends up fundamentally as an engine of commercial manipulation it recasts the entire contact surface of your life as a great glowing billboard. In time even AI’s physical form factor is likely to reflect this intimacy. OpenAI is reputedly working on a trio of devices, some of which will be worn or carried close to the body and we’re approaching a time when this technology is literally wired into our brains. Futurists like Ray Kurzweil speculate about a future where we begin to merge with this technology — it will become part of who we are. A version of AI that isn’t wholly ours and working purely for our benefit under such a future scenario, however distant, feels nightmarish.

And beyond simply being difficult to avoid — we genuinely need AI. Its potential in fields such as healthcare, education, and general wisdom is just too important to mismanage. These are deep and intimate aspects of life… wouldn’t we naturally be appalled if advertising started showing up in hospitals, classrooms and temples? Is there a chance that applying AI as a giant advertising platform dwarfs the size of the other markets above in which it could do genuine civilisational good? Will we again see a generation of technologists getting rich focusing on the former and neglecting the latter?

The fact that the AI market is likely to exhibit high concentration makes the above particularly pressing. Today there are a small handful of corporations in the race to extend the frontier of AI capability. Like many other digital technologies that are unconstrained by physical location and have near-zero marginal cost dynamics, are we heading into another winner takes all scenario? Or at least a very small number takes all? Given the likely pervasiveness and importance of this technology, this to me feels highly problematic. Anything even remotely resembling a monopoly on intelligence sounds like a bad outcome. There are some highly respected voices, notably Yann LeCunn, erstwhile Chief AI Scientist at Meta, who envisions a preferable industry structure that looks much more like a broad, communal open source endeavour with many distributed parties contributing resources, data, and other elements to an open, shared, accessible form of AI. Whilst a topic for another time if we are indeed heading for extreme concentration in the structure of the AI market, this should only heighten our concern, as the potential for abuse of a dominant position is all the greater.

So what options do we have?

First up we have the law…. Scott Galloway is an NYU professor and vocal critic of our collective failure to effectively regulate social media and avoid its most obvious harms — particularly with regard to its impact on children. To paraphrase him — if we don’t have laws to regulate this effectively, we have the wrong laws. Today the first bans on social media for children are coming into force as nations recognise the harm an unchecked ad-based commercial model can cause when taken to its fully optimised extent. Should we take a similar legislative approach to a technology which promises even more power to command attention and manipulate under the wrong set of incentives?

Second we have leadership…. And today we have some striking diversity amongst the loudest voices in the AI industry. Of the currently dominant AI corporations, Anthropic has probably the most ethically sound and mature stance on this question — stating that they believe advertising has no place in the commercial model of this technology. It appears their primary approach to monetisation is direct license fees from users. It’s worth noting that their moral fortitude has recently been tested (successfully so far) in a separate domain — the use of AI in war — through a public showdown with the Pentagon.

Dario Amodei, CEO of Anthropic, and Pete Hegseth, US Secretary of War. Courtesy of Business Insider

In contrast, and leaving it at the mercy of Anthropic’s creative PR, OpenAI is already experimenting with embedded advertising, although they have defined some clear guiding principles which suggest how and where this will apply. As stated, these feel sensible, thoughtful and situated within a broader mission of doing what’s right for people at large and thereby creating a service that’s compelling enough that people grow willing to pay directly. Nonetheless I think we need to keep incentives in mind here. What would happen if it becomes clear that consumer AI is the most powerful selling engine ever created and using it to monetise a user’s attention is worth far more than any reasonable subscription fee that they themselves might directly contribute? Will we continue to stick to our principles in the face of such a monstrous unclaimed economic prize?

Google occupies an interesting space historically in this story. Although they are the OG of digital advertising and still dominate much of the web outside of the large social platforms, their corporate motto of 15 years, ‘don’t be evil’, does seem to have guided their actions and diffused out into public perception of them as an organisation. Perhaps the sheer indisputable utility of their offering (email, maps, search, etc), particular when contrasted with the lack of genuine utility offered by the social platforms, coupled with the thoughtful attempt to make their advertising itself as genuinely useful as possible (with the exception of youtube imho) suggests a credible middle ground approach similar to the one articulated in OpenAI’s principles?

[embed]

As ever, xAI’s Elon Musk offers a pithy articulation of his position on the topic. Grok is designed to be ‘maximally truth seeking’. Without doubt Musk is a hyperbolic, unpredictable and highly controversial figure, hyper-logical in some fields and seemingly awash with conspiracy thinking and confusion in others. Without dipping into that swirling debate, some of his early actions at twitter — experimenting with different forms of direct subscription models and efforts to remove bots from the platform suggest he is at least partly aligned to an approach which avoids some of the abuses incentivised by a pure attention harvesting model. And stripping that core phrase from its wild context, maybe ‘maximally truth seeking’ does in fact hit the ethical nail on the head.

Hearteningly, one of the first major AI companies to launch advertising, Perplexity, this week decided to remove it from the platform. One executive stated: “the challenge with ads is that a user would just start doubting everything . . . which is why we don’t see it as a fruitful thing to focus on right now”.

Architecture is more durable than personal principles

Thirdly… seeking a solution which is more durable than the principles and intentions of today’s corporate leaders, some groups and thinkers are proposing practical architectures that provide an alternative model compared with the data hungry, ad-driven and highly centralised status quo. In particular, Tim Berners Lee, inventor of the internet itself, has launched a standard (called Solid) and a company (called Inrupt) designed to redistribute the centre of gravity of the ecosystem at large back towards decentralisation as a core principle.

He envisions a world where instead of a handful of unassailable platforms capturing and monetising the personal data of billions of users, sovereignty is returned to the individual. Fundamentally an individual would take control of their own data, storing it in a construct called a pod and choosing which organisation to share it with and under what circumstances. In theory the right organisations might gain access to vastly more relevant data (think healthcare etc) but with a user’s explicit permission.

Inrupt Agentic Wallet

This to me represents a far more reliable and ethically sound equilibrium which repositions the user as the customer rather than the product and empowers genuinely useful services to create far more compelling experiences. Perhaps the question is whether we, the people, are sufficiently mature at sufficient scale to help shift the equilibrium against such entrenched interests as the big centralising platforms of today’s internet?

Is the future agentic?

As a final thought, it is worth speculating on what a few more years of innovation and adaptation might bring and how this question might be applied to a world which looks very different from the human-centric one we inhabit today. There are early signs of a world in which humans are not users of AI as such, but rather delegate their tasks and needs to a growing number of increasingly autonomous AI agents. Gartner talks about ‘machine customers’ and many of the big AI services are laying the foundations for ‘agentic commerce’.

From ArkInvest Big Ideas 2026

In a world where we have not just consumers and producers, but also one in which both sets of participants are joined by a proliferation of autonomous agents acting on their behalf really ups the complexity of the ecosystem. Broadly I think there are two flavours of outcome here… one in which we build for ourselves a new kind of interface to the world which behaves like a talented and trustworthy chief of staff. Peace and stillness are returned to us as our agents head out into the world, eating complexity and bringing us only the most self-actualising of information and decisions to be made. Let’s call this one ‘self-sovereign AI’. This is not to say that out of our human view there isn’t a frantic, silent struggle ongoing between our personal agents and those seeking to make a profit from us — an inconceivably complex war of negotiation, claim making, broadcasting, and trickery that makes today’s battle for human purchasing power look tame. One might reasonably hope that through this agentic wrangling, the system moves towards efficiency and utility rather towards manipulation and demand-farming.

The other flavour is obviously darker — and involves a scenario where our ‘personal AI’ is insufficiently powerful or inappropriately designed. In this world we could imagine all manner of malign outcomes— perhaps our personal agents are able to be ‘manipulated’ by commercial AI, perhaps our own agents begin ‘managing upwards’ keeping us well-fed, but still hungry for more, appeasing the worst of our human natures and making us individually weaker, less discerning, less capable. Ultimately it could be a combination of both, and responsibility will yet again return to the individual to make best use of another technology that could either fatten and anaesthetise us or elevate us to greater flourishing.

Time to act

Today, at the beginning of another great technology cycle we have an opportunity at multiple levels — societal, organisational and individual — to act and establish an incentive architecture for AI that will shape the way this technology evolves. History demonstrates that incentives will be followed to their logical extremes and where these don’t align with collective human wellbeing, we must intervene. Our choices today help either set the conditions for radical manipulation at scale or radical increases in human flourishing.

Societally we need a public discourse that results in and legitimises the right legislation. Organisationally, we need long termism and sound ethics as well as governance structures that can bolster these against powerful commercial pressures. And individually we need to be able to recognise when we’ve become the product — our attention being farmed and harvested — and learn how to resist this through self-sovereignty over our personal data and selecting the right material from which to construct our new window on the world.

This story is brought to you by Llewelyn Constant, Innovation Evangelist at TCS Pace and member of the AI@Pace Collective.

Contact the TCS Pace London Team or Llewelyn on LinkedIn