👀 On Reading in The Age of AI

A special, 🤠 FREE hi, tech. just for you

“I write because I don’t know what I think until I read what I say.” – Flannery O’Connor

Reader, I changed the title of this post soon after I started writing it.

Original title: On Writing in The Age of AI.

Writers are “vain, selfish, and lazy”, as Orwell so memorably put it. The rise of generative AI, which can seemingly create new content from digital air, has brought those very vices into sharper focus.

In the past 24 hours, I have seen the following articles:

• “I’m a copywriter. I’m pretty sure artificial intelligence is going to take my job” - The Guardian

• “Stanford students are using generative AI in online exams” - The FT

And I received an invitation to the “AI for Writers Summit”, too.

Next, I sat down to add a little more ink to the reservoir, because I am a writer and I am vain and selfish. Occasionally lazy, too.

You expect better from hi, tech. - and rightly so.

Instead, I decided to reframe the question.

What would it be like to read all of this autocreated content?

The most effective examples of GPT-3 content I have seen would fall into this category:

Content that no-one wants to write, and no-one wants to read.

In other words, the majority of content that is published online today. In fact, much of the content we create is already for machines. Why shouldn’t it be written by them, too?

Therefore, the impact of generative AI could be an acceleration of the sensory deprivation we already feel today, if you’ll pardon the paradox. Two-thirds of Europeans are immersed in ambient noise that is the equivalent of perpetual rainfall.

We can’t hear ourselves think.

Machine-written content and the reading experience

I gave a presentation years ago about the evils of so-called “SEO content”.

SEO content includes those predictable blog posts (5 things to do in Acapulco, 5 ways to do your tax return - in Acapulco, etc.), keyword-stuffed product descriptions, recipe blogs, the opening paragraphs of news articles, and just about anything that tries to grab Google’s attention.

SEO content is stunningly consistent in its delivery, whether it appears on behalf of Burberry or Barclays. Here’s an example I took from the Barclays website:

😬😬😬 Good luck, ChatGPT!

As sentence construction goes, SEO content is the equivalent of that time I tried to build an IKEA wardrobe.

Sure, the pieces were all there. Sure, they were joined together, albeit not with the provided equipment and not in the right order. And sure, I paid a guy off TaskRabbit £50 to take it apart and start again.

Except, we accept these syntactical constructions online. Or rather, we have trained our minds to ignore them. As long as it get the job done, right?

When Homer Simpson built a spice rack

The content is typically authored by people, who write it as though they were malfunctioning robots. I know - I have written so very much of it.

For many writers, any notion of idiosyncratic sensibility is toast once it meets the tyrannical Yoast WordPress plug-in.

*Keyword density too low*

*Simplify your syntax*

And we amend until the robots give us the green light to publish.

The reading experience is not the foremost concern here. Publishers care about communicating with Google, in the hope that it will direct readers towards their content. Readers = Clicks.

I am laying this out as a framework to illuminate the inevitability of generative AI feeding the content machine in the near future.

It is cheaper, it is faster, and it is better suited to meeting the strict criteria of WordPress plug-ins.

We could do better, if we cared

Nonetheless, we continue to commit the same errors if we take this framework itself as permanent.

Why should we view this highly impressive technology as little more than a means of automating content that has no true audience in mind? Shouldn’t we aim higher?

These questions perhaps belong to a future stage of generative AI, yet they are worth posing now. The future of generative AI will be dictated by people, after all. And it will arrive an awful lot sooner than we think.

What is the mental experience for a reader that is bombarded with unending SEO content, no longer constrained by the puny limits of human fingers?

Barclays could pump out 1,000 articles a day about the benefits of a credit card (in Acapulco), but to what possible end? We will learn to ignore it like we ignore terms and conditions. It only adds to the constant noise that surrounds us, desensitising us further. Reading, as a medium, is supposed to connect us to something - even someone.

As DALL-E and Stable Diffusion evolve, the viewer will be subjected to even more artistic creations from the hands of robots too. This is already happening on social media in China and it will happen here promptly. If the content receives “engagement”, it will have all the incentive it needs to multiply.

This scenario assumes that today’s technology improves slightly and is then industrialised. Every company will have a generative AI toolkit at their disposal and if I know companies, they will not use it responsibly. Only a new business model for online content would put the end user at the top of the priority list.

However, our next question is:

If we make the imaginative leap to assume that generative AI could create content that were equal to - or superior to - the quality of a skilled writer, how would this affect the reader?

Let’s say the above was written by ChatGPT.

I know, I know - the Muses have so amply stocked my quill that no meagre robot could dare replicate my all-too-human prose, but do play along.

Would you care? Would it matter to you that the output came from the process of an algorithm, and not of my frazzled mental circuit?

One thing - perhaps the only thing - I have learned from writing this newsletter is that readers appreciate a personal take on the big stories. That’s not what I expected at the outset; I figured that my own insignificance would be better married to an objective tone of voice.

My irreverence matters more than my irrelevance, I have learned.

An AI could learn to mimic any style, but could it develop its own? That is the key question here. Otherwise, it is a party trick, a mechanical quirk to marvel over rather than to engage with as an equal.

The most difficult task for generative AI will be to replicate content that is by people and for people. We have simplified the task for AI by stooping to the machines’ level for too long.

In a desensitised world, we should reframe the question. As the opening citation from Flannery O’Connor highlights, we are all readers - even writers are readers. Why not elevate that role and strive to improve the experience with the new tools at our disposal?

Instead of asking whether AI can churn out enough mediocre content to replace copywriters, we must ask how it can help us to create content that connects with readers on a deeper, human level. Content that neither people nor AI could create alone.

That, to my mind, would be a mark of true technological progress.