Google search and GenAI – a lost opportunity?

In retrospect, Google missed a promising opportunity to build a GenAI tool before anyone else, when it let go off, over ten years ago, the ‘search within results’ feature in its search engine, which let you search within results with newer search clues. You could keep repeating the process with every set of search results, in the manner of a structured iterative process, which, by the way, is the foundation of chatGPT or any similar GenAI tool that may yet emerge. Since it was searching (again) within the results, which is a small part of the internet, the search engine could find better and better reference points linking newer search clues, improving with every such step, the only limit being your own imagination and understanding of the subject you were searching for. Imagine peeling off layers and layers of irrelevance to finally encounter the quintessence of what you were looking for, an activity no different from the way a sculptor works. I did this for a living for over ten years, which meant that I did this almost every day.

In my view, the precursor to this feature was a ‘structured query’ option in some of the sites, especially journals. I remember an excellent such feature in the Euromoney website during the mid to late 1990s, when search was in its infancy. Relevance of results is the most fundamental attribute for any search engine, and it improves when the engine navigates a closed family (figuratively speaking) of textual data, as was the case with Euromoney, which dealt with banking and finance. The structured query feature let the search engine see better connections to the search clues, making finer and finer connections and correlations, because the language of the site is a well-defined population and limited, in the strictest statistical sense of the word.

In my view, Google must have seen the potential to generalise this feature with the ‘search within results’ because it is simply a question of refined structure. Each iteration in effect leads to a new set of sites thus reproducing the ‘structured query’ feature of a single site.

Google even invited a debate on it in which I participated, beseeching the search giant to keep the feature and not let it go. But they did, replacing the feature with Boolean logic, which based on intelligent use of + and -, as well as ‘and’ ‘or’ ’no’ for whatever set of words or word. In my view, Google search results were never the same, reinforcing a simple point: the familiar is the foundation for relevance. 

I have wondered several times what logic drove Google to withdraw arguably its most promising feature. The only answer I can think of is advertising, because by then it had become ‘the world’s premier online advertising agency disguised as a search engine’ as The Economist so aptly described. Perhaps, the advertising algorithm was comfortable with a certain degree of irrelevance, generating more ads than the ’search within results’ would have. 

Who knows what else could have emerged out of the Google search engine had the feature continued? There is a logic to my saying this. In an earlier article ‘Searching for an answer’ ( https://www.challengingintelligence.com/tech-info-data-analysis/searching-for-an-answer/), among other things, I wrote of Microsoft’s proclaimed intention to use chatGPT with Bing, to produce more relevant results. And, as I have observed in another article, Professor Sutton, the pioneer of Reinforcement learning, predicted in 1995 that the future of natural language processing lay with search (http://incompleteideas.net/).

 And there were already domain-specific search engines emerging; healthcare was among the first to develop a domain-specific search engine. I am sure Google was aware of these developments. Perhaps, management attention was drawn to the other businesses Google got into through acquisitions, funded by the enormous profits from search. For all its sprawling businesses, bulk of Google’s revenues arise out of its search-linked advertising, across multiple channels. Meanwhile, a cottage industry has already mushroomed around chatGPT, which will force Google to forge a new and decisive path, as it did with the PageRank algorithm.

A Samsung CEO observed, when it was enjoying extraordinary success, that the most slippery period is when you are extremely successful, because serious mistakes can get made and not be seen until it is too late. The mistake is dwarfed by the enormity of success only to be suffered later as a gigantic blunder. This has happened quite often in the tech industry.

This is where vision comes into play, understood as the ability to ‘see’ how businesses – your own and others’ – will shape up the future environment and the ability to define a business in deep simplicity as Steve Jobs did when he argued that computers can be intuitive leading to the Mac & iOS, that mobile phones can become mini-computers leading to the adoption of a different chip architecture and that music was central in the lives of ordinary people, leading to the by now famous line of ‘1000 songs in your pocket’.

‘Learn to let go’ was an advice that Peter Drucker used to give. ‘Learn to hold on to’ would have been good advice. For Google.      

Takeaways

Search within results could have led to GenAI tools

There is no guarantee that you have completely understood the potential of your own product

Caution amidst success is the mark of a great business leader

Vision is the ability to see what others miss

Intelligently used and with purpose, hindsight is useful