How Generative AI Hallucinations Can Benefit Brands
Let's think of AI-generated hallucinations as a feature rather than a bug. Jon explains why they are creative sparks, not something to fear in his recent Ad Age article.
(An insightful 3-minute read with a fresh lemonade and shortbread biscuit)
About a month ago, there was a flurry of pieces popping up telling us all what we needed to know about AI.
One thing we do know about them now is that they are totally out of date.
This month I was invited to judge a Turing Test in New York where humans went up against AI to make ads and we had to tell them apart… it was damn hard… already. I also spoke to some boffins from Dublin who are building AI for the big four. I had a natter with the computer science faculty at perhaps the UK’s most prestigious university. Spoke to procurement at a massive global client. In between I seem to be watching a lot of YouTube videos out of the west coast at double speed at 2am in the morning. The landscape is moving fast. But honestly if you meet anyone who tells you they got it all worked out. Run a mile.
So. Right now. Today. This is what I believe we all need to know.
Next week it will be different.
But for now, here is the most up-to-date thinking I could find.
Hallucinations are great - they are the feature, not the bug
Does AI dream of Electric Sheep? Contrary to popular belief, AI-generated hallucinations are not going to be anyones downfall and are not something to be scared of, as many people seem to think. In fact, they could become something of a dream opportunity for marketers. They actually offer unique and unexpected combinations that can drive creativity in new and unthought-of ways. Isn't that just what a creative ‘spark’ is? They could be perfect for identifying, exploring and manipulating brands themselves as a concept. By experimenting with AI-generated content, you can tap into fresh ideas and engage customers in innovative ways. Get yourself off the blank page. They could actually turn out to be the most interesting feature of this new technology.
Chatbots with personality
The bot has grown up, found a calling and developed a personality. By integrating bots into the whole process, qualitative insights can be gathered and used to develop specific audience personas, which in turn inform/train the AI-powered brand persona. Instant audience quick dip. However, you need to ensure that the personas created encompass diverse perspectives, don't forget the internet is not a perfect data source by any means. AI can also be used to enhance client conversations, improving the overall understanding of client's needs and preferences. And when a detailed audience persona has a chat with a developed brand persona… hmmm. There's a feedback loop.
Up close and personal
When interacting with brands through large language models (LLMs), marketers can employ conversion tactics, meters, and sliders to fine-tune and personalize the AI-generated content. This means you can align the AI output with a brand's desired tone, voice, and messaging. Customizing the content according to specific metrics and preferences creates a more tailored and impactful brand experience.
Conversion Tactics: Optimize conversion rates using AI-generated content through tactics like A/B testing, experimenting with different CTAs, and incorporating social proof elements.
Meters: Gauge personalization in AI-generated content by adjusting meters for the level of personalisation, tone, style and language complexity to align with their target audience.
Sliders: Use sliders to fine-tune AI-generated content by prioritizing product/service features, adjusting emotional appeal, and controlling content length and format.
Train the bias out
This has rightfully grabbed the headlines recently but it’s such a hugely important topic it’s worth revising. It is vitally important to recognise that biases can arise from both the creators of AI models and the training data used. To head this off you need to create diverse teams that bring different perspectives to the table and to foster inclusivity…but hey, we’ve been saying this for years right, so who knows when this will happen? RLHF (Reinforcement Learning from Human Feedback) is so important here. And when you’re producing a sea of content at machine speed do you let a machine check it at machine speed. Is that trust or idiocy? Regular human intervention and evaluation of the output are also essential to identify and correct biases, you’ve got to train it like a puppy. Or it will disgrace itself. For now….
Understanding AI in advertising is like trying to ride a comet; it's incredibly thrilling, incredibly risky, and it's almost impossible to predict where it's going next (and could cause an extinction level event).
We've established the intrigue of AI-generated hallucinations, which, like dreams, can spawn a world of fantastic, creative ideas. Today's bots have grown, becoming more sophisticated, more personalized, more useful. The power of personalisation offered by large language models stands to transform how brands connect with their consumers. This dynamism, this ability to create tailored experiences, speaks to an evolution of branding that's not just unique, but uniquely resonant.
The promise of AI is enormous, but a lack of diversity, and bias in training data could all lead to significant problems. Therefore, we must apply the same vigilance and ethical scrutiny to AI as we would to any influential force in society.
As we stand on the cusp of a new era of AI integration in advertising, we need to remember that the goal isn't just to keep pace with the rapid advancements. Instead, we must actively shape and direct the trajectory, remembering always that AI is a tool, not a master.
But today's news is just tomorrow's chip paper as they used to say where I come from. Be prepared to evolve, to adapt, to redefine what you thought you knew. After all, the brave new world of AI waits for no one.
Photo by Gertrūda Valasevičiūtė on Unsplash
