ChatGPT sells better than humans

FACEBOOK doesn't sell “your" data, contrary to what many often claim. As someone who has worked in media, I've always found this notion puzzling. Facebook, like an old-school newspaper, sells attention while fiercely guarding its data. Moreover, it's not just “your" data; your interactions on the platform are with other people, giving them meaning and value because they're part of a social network. While your personal data may hold significance to you, its value to businesses is tiny in isolation. Only when data is collected at scale and scope, linking together, does it start to have real value. This value is often fully realised when the data is collectively processed using machine learning algorithms to extract meaning.

So, for quite some time I have found myself at odds with the prevailing narrative on data privacy, particularly in the context of online advertising. The concerns raised by critics, in my view, have been disproportionate to the risks involved. Influential works, such as those by Shoshana Zuboff on “surveillance capitalism”, have struck me as not only grandiose in their claims but also riding the wave of a moral panic that’s swept up public discourse. I have long suspected that the reason behind the widespread misunderstanding of these issues, exemplified by the EU Commission's General Data Protection Regulation (GDPR), lies in the pervasive influence of neoliberal thinking.

Proposals such as granting individuals property rights over their data are fundamentally misguided. Such an approach would probably result in a scenario where only those who are financially desperate would be willing to sell their data, as the monetary value of an individual's data is low. Moreover, the idea of data ownership becomes murky when considering the nature of social interactions. Who, for instance, can claim ownership of a conversation between you and four friends? My suspicion has been that these proposed solutions are nothing more than a clumsy attempt to apply individualistic market-based thinking to a problem that is, at its core, a collective one.

The efforts of privacy activists such as Max Schrems, the introduction of Apple's app tracking transparency measures and the GDPR have had a tangible impact on the landscape of targeted advertising. Through my involvement in a fashion startup, I witnessed the effects first-hand as our advertising costs increased. This experience is not unique, as several studies have corroborated the significant impact of these changes that have made targeted advertising harder.

The “long tail" of the Internet, heralded by Wired magazine's editor nearly two decades ago, promised a media landscape catering to niche interests and tastes. While many early internet promises fell flat, the “long tail" of small businesses serving niche audiences with specific product and service interests remains a vibrant reality. These businesses depend heavily on targeted advertising to effectively reach their intended customers, a vast improvement over the old-fashioned, derisively nicknamed “spray and pray" advertising approach that suited mass undifferentiated products.

I believed that advertising on platforms like Instagram had become so effective that it had transformed from a nuisance into a useful tool, showing users content that genuinely piqued their interest. However, because of these measures in the name of privacy, the landscape has shifted. Independent advertisers must now purchase more ad impressions to maintain their sales figures. If Facebook chose not to increase the number of ads displayed to avoid annoying users, it would lead to two other negative outcomes: an increase in the cost per sale and a decrease in the total number of sales generated. This situation is detrimental to users, particularly harmful to small businesses and startups, and ultimately damaging to the economy as a whole.

Drawing from the extensive literature on the spread of “viral” information, I found myself less concerned about the notion that people were being manipulated into purchasing things they did not desire, as Zuboff alleged. Studies suggest that information does not behave like a virus, spreading indiscriminately; rather, its spread is more akin to a spark. Just as a spark's ability to ignite a fire is contingent upon whether the forest is damp or dry, the effectiveness of a message in influencing behaviour is heavily dependent on the receptiveness of the audience. In other words, the likelihood of individuals acting upon a message is more a function of their predisposition to the content than the message's inherent persuasive power.

However, that was before the arrival of large language models (LLMs) such as ChatGPT. A recent paper seems to confirm Zuboff's fears: LLMs are already as persuasive as humans, even without additional training. But where they had minimal demographic information about their human interlocutor, such as gender, ethnicity and political affiliation, LLMs were significantly more persuasive than humans. Interestingly, when two humans interacted, they became less persuasive when given more information about the person they were trying to persuade.

Other recent studies have demonstrated that LLMs can accurately categorise people into one of the widely accepted Big Five personality traits by analysing just a handful of their Facebook status updates, all without any additional training. Similar tests have yielded comparable results using posts from Reddit and X (formerly Twitter). Moreover, other studies have indicated that by examining a user's friends and likes, machine learning algorithms can predict sensitive attributes, such as sexual orientation.

In light of these findings, it's not far-fetched to imagine that targeted advertising may soon be capable of not only identifying the right prospective buyers but manipulating people into purchasing products they don't actually want. Even more concerningly, this ability could be harnessed by many actors. Instead of ads, they can scrape conversational data off large platforms and use that to influence opinions in political and other campaigns using bots masquerading as other users. Perhaps I owe Zuboff an apology.

♦ VWB ♦

BE PART OF THE CONVERSATION: Go to the bottom of this page to share your opinion. We look forward to hearing from you.