Google is going to get even better at search which will improve its relevance with users and increase its dominance, but it won’t help in the struggle with Amazon for the smart home.
Google has announced the launch of a major revision to its search engine which sees it improve its relevance in the long tail of queries as well as understand its users better.
After a lot of development, Google is finally letting BERT (Bidirectional Encoder Representations from Transformers) loose in its commercial search algorithm.
It is also migrating search onto its Tensor Processing Units (TPUs) in the data centre.
Put simply, BERT changes the way Google understands the search queries that are entered into the engine.
Previously, words were understood on a stand-alone basis, but BERT tries to look at the context that is being used in order to understand the query more accurately.
Something similar has been in use for some time in Google Translate, but this migration to its core product signals that BERT is now accurate and mature:
I suspect that users will notice two main changes:
First, comprehension: With better comprehension will come a more natural interaction with the search engine.
Google estimates that 1 on 10 times it gets the sense of a query wrong (more in my experience) and it is these misunderstandings it aims to fix with BERT.
Google gives a couple of examples of how this improves comprehension which are similar in type and complexity to the areas where I have noticed the search engine going wrong.
The computation required to make this work is so great that Google has had to shift the load to its TPUs in order to run the algorithm as quickly and efficiently as possible.
If this becomes standard across all of the Google’s queries, this may begin to put further pressure on gross margins unless Google can also increase the accuracy of the targeting and hence the value of the advertisements it sells.
Second, long tail: The long tail is where Google really stands out from its competitors like Bing, Duck Duck Go and so on.
This refers to Google’s ability to accurately pick up some very obscure, but relevant results thanks to its huge knowledge graph.
The combination of this index and better comprehension should also produce significantly better results in the long tail where a large part of Google’s differentiation lies.
The migration of using BERT in search should improve Google’s dominance in search but I think it is not going to help much when it comes to digital assistants and voice search.
This is because Amazon Alexa is already known to be an inferior product but it is still gaining dominance in the digital assistant space.
This is a combination of Google’s inability to gain share with its own devices (Google is not very good with any product that is physical) as well as a deep mistrust from many of its partners.
This is particularly the case in the automotive industry where automakers are much more comfortable using Amazon Alexa than they are Google Assistant.
OEMs fear that Google will repeat in vehicles what it has done in smartphones and reduce the hardware makers to commodities and earning most of the profit for itself.
The net result is greater barriers to entry for the core business of search, but no meaningful impact for its ancillary products where I see the digital assistant continuing to struggle despite its superiority.