What the heck is going on with Google AI Search?
The story about Google's new "AI search" going haywire needs no introduction at this point. From the outside, it appears that the AI is pretty bad, and Google, despite being behind the core tech that made generative AI possible in the first place, is unable to "get the thing right".
I am not here to defend Google. However, I think there is a lot of misconception going around that needs honest and realistic discussion.
As we all know, Large Language Models are trained with a vast swath of data from the Internet, and Google is uncontested when it comes to crawling the web.
So, this begs the question (or the dilemma) of what data should be used to train or ground the AI, and what to exclude. This is a challenging question because on the one hand, being selective introduces "own" biases. On the other hand, using all the data means that the AI is trained with whatever is pushed to the web, introducing "others" biases.
You might then ask: isn't the same "dilemma" also applicable to indexing web content (i.e. deciding what pages surface in search results)?
I think you are right. However, this doesn't address the "responsibility" aspect.
Whenever you use Google search, you get all the "relevant" results, and you are free to pick what suites you and ignore what doesn't. For example, if a result states that Mr. Obama is "Muslim", you might simply ignore it. What's important here is that you are unlikely to blame Google for bringing it up.
However, the expectations are very different when using the AI search, or indeed any GPT-like conversation. You get one result and so you have an expectation that it's exactly what you are after. It's simply a black and white situation.
Technically speaking, the data that the AI relies on to generate answers comes from a vector database that's optimised to pick whatever context "resembles" the search query. If the input is garbage, the output is likely to be garbage too.
In short, this tech is new and is far from being fool proof, so for now, just mind the gap!
![Accuval Jaafar Almusaad](/static/website/images/jaafar_almusaad.jpg)
Jaafar Almusaad is a co-founder and CTO of REXSMART. Jaafar has over two decades’ experience in computer engineering, software development and data science. He holds a Master’s degree in Information Technology Management from the University of Sunderland.
![Accuval LinkedIn](/static/website/images/linkedin.png)