Google doing what I said they would
Big-tech AI updates + Google moving to AI first search
Over the last week or two, there have been a few big-tech AI announcements. OpenAI announced a more sophisticated, all-round model GPT-4o, which looked impressive in the demo, but a bit creepy when the AI seemingly tried to flirt with the presenter after he said that he loved ChatGPT (We won’t be able to move for AI companions in the not to distant future).
Meanwhile, Microsoft had updates about it’s AI Windows machine offering further us invasive functionality called (Total) Recall, giving your machine photographic memory by constantly capturing screenshots of what you are doing (come on, how could this possibly go wrong?! I mean sure, if someone gets access to your machine they have visual audit trail of everything you have been doing, but AI, am I right?).
But really, I’m here to talk about the Google Gemini Search update, well, mostly to say called it.
Google’s AI Search
Predictably, Google has announced a move from traditional search to an AI search experience (I described such a potential move at the start of the year).
Despite suggestions that Google’s golden-goose ads product would limit their willingness to change the search model, they appear to be heading all-in on the RAG-the-internet approach similar to what Perplexity offers.
As I got into in my Perplexity AI overview (and general thoughts on AI vs Search), Google have got a very clear advantage here over newcomers, simply because a huge determining factor (possibly the biggest factor) is the quality of the retrieval step and Google have been perfecting and fine-tuning their search engine for over two decades, so competing in that space is always going to be hard (although I have no doubt that Google will also have it within them to completely mess this up, such is the nature of large organisations and innovative products, group-think and the like).
In a write up on the topic in The Verge, it’s not a surprise that Google go out of their way to mention the search side of things:
That combination of the Knowledge Graph and AI — Google’s old search tool and its new one — is key for Reid and her team. Some things in search are a solved problem, like sports scores: “If you just actually want the score, the product works pretty well,” Reid says. Gemini’s job, in that case, is to make sure you get the score no matter how strangely you ask for it.
Google is redesigning its search engine — and it’s AI all the way down
Sports data/results queries are the same example I walked through when looking at Perplexity previously - it’s a good example of exact, specific data types being looked for that also needs to consider chronology. Something that Google have long solved (I have googled many sports results over the years and not once seen anything incorrect or needed to doubt it) - Perplexity and other challengers have to solve this now, and given their AI centric starting point they might approach it with AI providing the answer from search results, but I think it’s still not going to be as reliable.
Interestingly, it also ties into another idea of “verified answers” that I have pitched before. That is, in an ideal RAG AI system if the retrieval step is able to fetch data that is close enough to the answer in itself, they system could skip the AI generation part altogether and surface the result, effectively as what I’d call a verified answer. This way it completely removes any doubt in the AI - this is what Google are essentially suggesting with scores, if someone asks for a score, the retrieval (search) step can solve that with a high level of confidence, so there is no need to throw AI in to the mix at that point.
And of course, it’s only been days since the announcement, but publishers are already up in arms about it.