The limits of generative AI

Generative AI at its worst

What becomes obvious on anything other than a cursory look at the event’s website is that it was either “crafted” (the wrong word as it suggests skill) by a person with a strange grasp of language and interesting views on aesthetics, or, in greater likelihood by generative AI.

Generative AI for marketing - a warning
Created by AI [Microsoft Designer, prompt: "terminator wandering around a chocolate experience with lollipops and bunnies"]

Both the use of language and the “style” of the images used strongly suggest the use of AI. Take the opening paragraph:

“Dive into the whimsical of Willy's Chocolate Experience!, a place where chocolate dreams become reality. Book your adventure now and embark on a journey filled with wondrous creations and enchanting surprises at every turn!”

Compare that to ChatGPT’s effort with this prompt “thirty or so words for Willy's chocolate experience”:

“Willy's Chocolate Experience invites you to indulge in a world of cocoa wonders. From interactive workshops to delectable tastings, immerse yourself in a sweet journey that promises delight and unforgettable chocolate moments for all.”

ChatGPT 3.5

The strong use of superfluous adjectives is a strong giveaway, it’s the type of thing a human copywriter will do when they haver copy to fill but not enough information to fill it with, it’s also indicative of the content AIs are “trained” on.  Also note worthy are the associations, “assumptions”, AIs make with even the smallest of prompts.

We’re all experts – the danger of AI

This highlights one of the key dangers of AI – it can make users of it feel like they know what they’re doing – when fundamentally, they don’t.

The above quoted paragraphs could loosely (and very generously) be described as “marketing spiel”, ready to be uploaded onto a website (and in the former example, it was). But it’s not. They are a collection of words which “sound nice together”, give the right “vibe”, but fundamentally fail because they haven’t been sanity checked (in this case checking whether they are actually describing the event accurately).

Generously, we can ascribe this to an oversight, ungenerously…?

AI is a tool not a replacement

The key learning from this, and the undoubted thousands of other examples out there, is that AI is a fantastic tool for marketeers (and many other professions), but it is not a replacement.

Whilst not impossible, it would be very rare for AIs in their current form, to give perfect responses from simple uneducated prompts. There needs to be a basic level of understanding of the subject matter by the user in order to generate usable, robust output, and to assess and edit that output to make it meaningful for the purpose it’s designed for.

Quite simply – “garbage in, garbage out”.

From our experience of AI, it’s a very iterative process to make it work effectively. We ask a question of it; we get an answer – 9 times out of a 10 it’s not perfect. We refine that question, keep asking follow ups, keep refining the query until we have something that we can use.

Understanding the limits of AI

It’s important to recognise that AI, in human terms, isn’t intelligent, it doesn’t comprehend what it’s being asked – that would be Artificial General Intelligence (think HAL 9000 from 2001: A Space Odyssey, or Skynet).

Instead, current AIs are very good at analysing data, discerning patterns, trends and relationships within that data, then creating statistical models based on that data which it can use to output responses within a specified format for specific queries.

This can give the impression of being “intelligent”, but in reality, is a series of statistical probabilities that we perceive as being intelligent.

It’s important to note that errors, mis-truths, biases in the underlying data the AI has been “trained on” will be reflected in its responses to queries.

Generative AI the future of marketing?

The simple answer – yes and no. 

AI is broadly going to give you, if you prompt it intelligently enough, the basics – but it won’t give you the finished article. Remember, the “basics” here is not a unique piece of prose (or other form of content) focussed on your business but is instead a collection of statistical probabilities.

That might work for some, for many it won’t. In any case, any form of generative AI output does need to be reviewed, checked and checked again.

If you would like to be kept informed of when we add another blog post, signup below.


Please see our privacy policy for how we manage and store your personal data.