Words: Calum MacLeod – Senior Consultant
Category(s): Branding, Content
Date: 1st October 2023
The AI debate tends to focus on exhausting and potentially terrifying macro musings: AI stealing our jobs, AI being used by enemies of the state to attack our way of life, or AI becoming self-aware and going all Skynet on us. For brands, however, there are many AI in marketing questions that are no less fundamental but with a lot more bearing on the practical day-to-day.
- Can we (how can we) use generative platforms such as ChatGPT, Claude or Bard as part of our workflow in an ethical and legally compliant way?
- Does using AI in marketing, and folding these platforms into the creative processes, risk leaving us reputationally and legally exposed?
Google hints at AI validation
Google’s stance on this debate has recently become clearer. Its September 2023 search algorithm guidelines no longer contain a stated preference for human-created content. This can be taken as implicit – yet significant – support and validation for the use of AI in marketing and specifically AI-generated content.
Google now suggests that content creators should focus on producing “original, helpful content created for people”. It’s perhaps worth reminding ourselves at this point that all the content produced by conversational AI tools such as its own Bard platform or ChatGPT is, by definition, unoriginal – an amalgamation of content scraped from the web.
Exactly how Google’s algorithm squares this circle will probably remain a mystery.
A lot of lawsuits
There’s currently no consensus between legislators and operators on how to frame AI’s rules of engagement, particularly with regards to how the platforms ingest (or pilfer, steal, plunder – as some would have it) copyrighted material. Consequently, and a little inevitably, the situation has got litigious.
A number of high-profile lawsuits are under way, with some very unhappy content creators suing platform owners such as OpenAI for alleged copyright infringements. Specifically, they have objected to the use of their copyrighted content to train models like ChatGPT in order to deliver an enhanced, more human AI output.
It probably doesn’t help that the platform operators haven’t exactly been forthcoming about the source titles their digital training use. OpenAI for example described its training materials as “two internet-based books corpora” (corpora is a term used in language research meaning “databases that contain texts”). It has labelled these corpora, with beautiful anonymity, Books1 and Books2. It’s believed that they may contain titles accessed via book-aggregating online databases called ‘shadow libraries’ (which sound like reading rooms on the Death Star) such as Bibliotik, Library Genesis, Sci-Hub, or Z-Library. Some of these databases have themselves been subject to legal copyright challenges.
For the extraordinary potential of AI in marketing to be realised, and for marketers to use the generative platforms with confidence, issues of trust, ethics, copyright and accuracy need to be overcome. The quality and integrity of desk research can’t be tainted by fears that the source material contravenes copyright law or has been pulled from behind a paywall or from gated content. While this may not be happening, legitimate information provenance needs to be beyond question.
Sensible or cynical?
When OpenAI launched its new GPTBot this summer, it attempted to distance itself from impropriety by simultaneously giving content creators the means to block the web crawler. By inserting a little bit of code in a website’s robots.txt file a creator can ensure content can’t be scraped to help train GPT-4 or ChatGPT. Hundreds of organisations, including the likes of Amazon, Bloomberg, Reuters and IKEA have now put this block in place.
It’s difficult to know what to think about this gesture. On the one hand, OpenAI didn’t have to provide the means to block its bots and it’s undeniably a reasonable stopgap while copyright lawmakers figure out how to untangle the whole mess.
On the other hand – given the many months of content-based training the likes of GPT-4 and ChatGPT underwent before the bot blocker was released – it feels a bit like providing a beautifully gilded padlock for the stable door long after the horses have bolted.
No path, no precedent for legislating AI
The trouble is there’s no AI or LLM bill of rights (and wrongs) – yet. There’s no framework for how the principles of creators’ copyright can be squared with this new and evolving set of technologies. Even the US Copyright Office has opened a public consultation on how and where copyright might be applied when it comes to AI. In November 2023, UK Prime Minister Rishi Sunak is set to host what has been hailed (by Mr Sunak and co) as the world’s first artificial intelligence summit at no lesser venue than Bletchley Park – Alan Turing’s old stomping ground and one of the spiritual homes of computer science.
No doubt the great and the good in attendance will focus on those larger, loftier existential AI pre-occupations than AI in marketing and content copyright, but any practical progression of discussions would be a welcome surprise. What’s clear is there’s no clear or easy road ahead and everyone is playing catch-up.
Let’s hope that sensible heads prevail and a compromise akin to that reached in the digital music sharing business might be reached. Until then, use of generative AI platforms remains a big grey area.
The more things change, the more they stay the same…
So how should brands (and marketing agencies) proceed for now? Staying up to date on legislation, regulations and best practice would be a good first step. Brands with high ethical standards might feel the best use of AI in marketing is as a tool for research and inspiration – a supplement rather than a replacement. Certainly, a consensus seems to be building that at least being transparent on how AI is being utilised in the creation of content is the right thing to do.
Some brands might want to take a stronger stance and (be seen to) do the right thing with regards to ownership and copyright laws. Others might be wary of a coming legal reckoning and the reputational and business risks associated with being on the wrong side of history.
Either way, there have always been solid business reasons for brands to ensure their content output is original and factual. The existence of a new tool to aid content creation doesn’t change this.