top of page
  • Mal McCallion

Google's Gruelling AI Glitches

Updated: Jun 11

IMAGE: Google Gemini


Google’s AI rollout has seen many self-inflicted disasters – but it’s the latest that’s the one which should worry business owners the most.


If you’ve never thought of applying glue to your pizza, or chewing the occasional rock, you’re not alone – these are terrible ideas that should not be tried anywhere. Yet Google’s new ‘AI Overview’ says you should do it. Some people did.


How did this once-surefooted search behemoth find itself so lost and humiliated? It all started, as it often does, with someone else breaking all the rules.


Back in the halcyon days of October 2022, things were looking pretty peachy at Google. Whilst there had been some grumbling about the steady decay of the border between ‘organic’ search results and those that you were served up as ‘sponsored’ links – which had become more plentiful and also more camouflaged, so that you accidentally hit them more often and flipped more coins into Google’s pocket – it was the only place you ever really went to find stuff.


This settlement, the idea that this was how things had been for so long that it would always be so, had its tsunami, earthquake, iceberg and volcano all at once on 30 November that year. Little-known outside the geekiest of AI crazies, a small outfit called OpenAI released a chatbot called ChatGPT – and everything changed.


The way that OpenAI’s leaders now frame this moment is all a bit too ‘humblebrag’ for me. According to their whimsical history, they had this thing – ChatGPT-3 – that they thought was useful but didn’t really know what to do with it. The leadership team, led by CEO Sam Altman, decided to go out for pizza and – just before they turned the lights out in their small San Francisco office – thought they might set it live and see what happened.


Suddenly, they had the fastest-growing consumer product ever on their hands. 100 million people – more than the entire UK – registered to start using it in month one.


Quickly, its applications to the business and consumer worlds became a focus – and there were plenty. But at its heart it conquered one particular use-case incredibly well – it helped you discover straightforward answers to your random questions. It searched its memory of everything that had ever been written online – up to April 2022, when its training data started to be prepared – and replied, quickly, with what you wanted to know.


Want to learn more about AI's impact on the world in general and property in particular?

Join us on our next Webinar!


As fast as we are becoming used to this functionality – and, being humans, already starting to complain about its speed and/or recency – it’s worth remembering that initial interaction you had or will have with it. I am old enough to remember my first ever Google search, in 2000, and how amazing that was.


The first time I popped in a ChatGPT query (“Who is Mick Stockwell?” since you ask – and I’ll return to why, and how powerful that question remains, in another blog soon …!), the impact for me was more extraordinary that even that initial Google moment was.


Which brings us back around to that matter in hand – Google and its ‘AI Overview’ feature.


Whilst we in the UK were crashing around trying to escape the suffocating effects of Liz Truss’s disastrous kami-Kwarsi budget in Q4 2022, Google was issuing a ‘Code Red’ to all its staff – this meant an existential crisis that could put their entire business model in jeopardy. Because let’s face it, as lazy humans, why would we scroll through (and accidentally click) some highly questionable sponsored links that are kind-of-adjacent-but-don’t-really-answer-the-question on gnarly Google, when we can ask handsome, one-shot ChatGPT and get a curated single answer straight away?


The last eighteen months in Googleplexes across the world have essentially been endless internal meetings entitled – “How do we stop people, who like getting one-shot answers from ChatGPT, leaving our extraordinarily profitable Search Results pages?’


The answer, in the end, was ‘AI Overviews’.


They trialled them recently as ‘Search Generative Experience’ (SGE), which was really boring, so they changed the name. Meanwhile, the rest of AI was charging ahead – OpenAI recently launched the awesome, multimodal (ie it can see images and understand voice commands as well as text ones) ChatGPT-4o, Claude 3 from Anthropic is here, which can take in a Macbeth-length chunk of text and answer all of your GCSE coursework instantly, Meta’s Llama 2 as well, which is opensource and therefore free forever. Google launched Bard to try and keep up; it was pretty rubbish. They then launched Gemini and pretended it wasn’t Bard 2; this time, it was OK – but still not the best.


And that’s before we even get to its disastrous image generator …


IMAGE: Google Gemini

So finally, last week Google launched AI Overviews out of testing. And it’s another embarrassing misstep to rank alongside Google Gemini's black popes, vikings, Founding Fathers and Elon Musks.

But it’s a misstep that they’ll improve upon – because they can’t not – and this where it’s going to get challenging for every business owner out there.


Because AI Overviews want to find that one-shot answer. And if the question is ‘Which estate agent is the best in [your area]?’, what happens if the one-shot answer isn’t you?


Google’s flailing attempts to protect its revenues against insurrectionists such as ChatGPT are going to hit us all. Getting on the front foot, ensuring that you’re maximising your chances of appearing at or near the top of the answer to those questions is critical to future success.


And, if the best time to start this 01 December 2022 – the second-best time is right now.

22 views0 comments


bottom of page