Reading Time: 3 minutes
Photo by Steven Wright on Unsplash

If you ask Google “What is the ugliest language in the world?” it will give you an answer (Kannada, according to Google, at least until 2021, when there was a minor scandal in the Indian media when Google’s response was revealed). The problem is that Google doesn’t think; just like generative AI, it simply repeats what it finds, and has no concept of self-filtering (except when there is a public outcry, and presumably Google staffers impose a more acceptable response).

Despite internet-based search only having been with us for relatively few years, we have, I suspect, become accustomed to its limitations. It sometimes (as above) gives us rather unpalatable answers that are unacceptable. But this is just one of many limitations of search that we have learned to live with. A 2022 “perspective” paper, Situating Search, takes Google to task for not really solving the problem of search. It’s great to raise the question, but I don’t think this paper really comes up with an alternative.

The paper, which predates the big launch of generative AI models, makes a fundamental point: “A search system needs to do more than just generate an answer, but should interact and make sense of information.” That is very true, but quite how any search system can do such a thing is quite a challenge. The authors describe rather dismissively what they term “Google proposals”, a basket term to cover all the innovations from Google in responses to a user queries, such as answering whole questions. Google has certainly improved, but it can miss the big picture of why people seek information.

Of course, Google has only one interface, and we know there are several types of search. There are many papers describing different types of search, but no single universally accepted classification, so the authors list several models of classifying types of search, all of them valid to my thinking. We can distinguish three ways of searching:

  • Lookup
  • Learn
  • Investigate

Or, to put it another way, search can be:

  • Searching as exploration
  • Searching to accomplish tasks
  • Searching as learning

But why stop there? Another author, Belkin, comes up an elaborate model of  four dimensions of searching, each varying between two parameters:

  • Searching or scanning
  • Selection or learning
  • Specification or recognition
  • Information or meta-information

But identifying the different kinds of search is a long way from interacting with a user who may not even ask a single question. “Based on your searching, I believe you are interested in the history of tennis. Would you like to know more on this topic?” is not something that Google has ever proposed to me.

The rise of generative AI has not really solved the problem of search. It might even have made it worse, since generative AI still requires a question, or a prompt, a question, before it can deliver an answer.  

One day we might get the kind of search we need. Until then, we haven’t progressed very far beyond the information desk of the library, where we ask an expert searcher how to find the answer to a question. They use their knowledge of information resources to point us to the most relevant answer. Come to think of it, that system worked pretty well. I distinctly remember a librarian saying to me, in answer to my question, “Did you mean … ?”