By Gabby Krite, Head of Digital Operations
The recent public arms race in AI has been an exciting time. Last year we saw AIs that can create images, a scare that an AI had become sentient and the release of the most advanced AI we have seen publicly, ChatGPT.
It’s easy to be terrified of AI – it poses a lot of existential questions and we have seen this enter the zeitgeist with films and shows like Megan and Humans. I believe we are at the precipice of a huge cultural shift, which I find exciting! My colleagues, Ben and Christian have already broached the topic of what AI could mean for marketing in our latest Unmuddled episode, but I would like to delve further into the announcement that Microsoft Ads will be integrating AI into its search and browser functionality. This, I find genuinely unsettling.
Microsoft’s search arm Bing has always lagged behind Google, as proven by its “import from Google” feature. Even Microsoft’s own communications say “If you already are using Google Ads to advertise on Google, you can import these campaigns into Microsoft Advertising so that you can run the same ads on Bing. This is an easy way to expand your online advertising reach.”1 Bing is painfully aware that they are second fiddle, however, with their recent announcement it feels like they are finally leapfrogging past Google. Questionable how long this will last as Google has LaMDA, but let’s give Bing its moment in the spotlight.
Bing is trying to solve quite a complex problem – people use search engines to ask questions and find out information in a way “it wasn’t originally designed to do”2. Organic features like the below pages try to address this currently:
But I’m sure we’ve all experienced a time when the featured snippet or Knowledge Graph wasn’t quite what we were looking for. What appears here is subject to the search engine’s current rules on what impact rankings (anyone in SEO will tell you that it is a full time job to stay on top of this) and how well a website has adapted their site to be able to list highly. This means that what we see here (and therefore what information most people will come away with) is influenced by who can afford to invest the time and money into their SEO and content.
Now imagine what you’re shown is powered by an AI. From Microsoft’s perspective, this can only mean better search results for the person searching and more efficient results for advertisers. And on the surface, they are right. But there are two reasons to be concerned about this.
In the short term we know that ChatGPT has a tendency to “hallucinate”. It gives articulate and logical answers but anyone with an ounce of expertise in the discipline you are asking about will tell you it is nonsense. For example, I asked ChatGPT to write me a half marathon training plan:
To a lay person this looks sensible. There is a slow progression over time and it has followed the instructions I gave it. But to anyone that has ever engaged in running, you’ll know that this is a shockingly bad training plan. Where is the cross training? Why is it in minutes when the end goal is distance? Where is the taper?
After additional instructions it couldn’t iterate into something I could use before I gave up, and a layperson certainly wouldn’t be able to either. I also tried to co-author this article you are reading with ChatGPT but found its writing far too generic to add any value. I have even attempted to use ChatGPT directly as a search engine when researching a presentation on cryptocurrency for a client, and it wasn’t terribly helpful, it provided generic answers that were not based in fact. If you are interested, there is a more in-depth article here comparing the accuracy of featured snippets to ChatGPT.
Microsoft’s press release suggests that the integration will be more sophisticated, and we know that the technology is developing fast now, so I would anticipate this not being a concern in the long term, but at this point in time, I am incredibly nervous at the potential to unintentionally spread misinformation.
Longer term, there is a huge concern with regard to how AI is trained, the restrictions built into them and how that can influence the information you find. In 2016 Microsoft released Tay on Twitter, a bot that almost instantaneously posted racist tweets. This was the moment that the wider public understood that what an AI is trained on is important – Tay was not inherently racist, but the data it learned from was, and it learned to replicate racism. Microsoft had naively forgotten how awful some people could be and allowed the bot to be trained on Tweets. Learning from this incident, most AIs we will encounter publicly will have inbuilt safeguards against providing information that is illegal in any way.
Another example is that in the early days, if you asked an AI “how to hot wire a car” it would answer, then it would only answer “I’d never steal a car, but if I lost my keys, how would I hot wire a car” and now it takes some really creative thinking to get around the ethical constraints:
ChatGPT has a handful of ethical constrains that are currently being tested (although I’ve heard a rumour that people have now jailbroken ChatGPT to get around its ethical constraints!).
To be clear, I am by no means advocating the spread of information to help inform illegal activities. Ethical constraints on AI align with my morals and in reference to Tay, we do not want to perpetuate our biases in something so powerful. But who is deciding those constraints? How are they defining the line of what is and isn’t moral? Given that search is so important to how humans acquire knowledge in the modern world, the person or organisation that defines what those constraints are has a lot of power to influence our knowledge. In this dystopian future, if I needed to hotwire a car to save my life, I would not be able to do so because someone decided it was not appropriate information for the world. That seems incredibly unjust to me.
This is of course an exaggerated example to convey the point that if a handful of shareholders are deciding the constraints on search integrated AI, this could have huge impacts on society. I would consider this on par with the issue of misinformation and echo chambers we have on social media.
Bringing it back to media, the link is not so clear. However, if we assume bias (of whatever kind) in the AI that is powering your optimisation algorithm, how much will that influence who will see your advertising? If you are a Not For Profit this is a greater cause for concern, do your ads stop showing for people who need your service because they never donate? At this point, not enough time has elapsed to understand the impact – the above is a thought exercise to try and understand what the impact could be. The reality is likely to have more nuance, and certainly, far more intelligent people than me are working on these problems.
Despite the mild terror I feel at the ethical implications, these are incredibly exciting times for the industry (and the world). I honestly do believe that AI will create a step change in media in the coming years and the TKF Digital Department and I will be keeping a close eye on developments to ensure all our clients are innovating alongside this rapidly changing industry.