Google researchers are rethinking search, and it should not come as a surprise to anyone, especially as data privacy concerns take more attention from companies and consumers.
The company, founded in the late 1990s, might be one of the most successful in the past century. Our society and the industry have transformed, forcing the company to adapt, change and discard many of its processes.
Experts at companies such as Google, Microsoft, and IBM have talked for years about how artificial intelligence and machine learning will change ad targeting and support the retrieval of information in search queries.
But there has also been talk about inherent AI bias, because the data is only as good as the models that support the system.
Timnit Gebru, who co-lead Google's Ethical A.I. team, tweeted in December that she was fired because of content in an email sent a day earlier to a group that included company employees. In the email, she wrote about bias in artificial intelligence, among other things.
Classic information retrieval systems do not answer questions directly, write the authors of a paper published earlier this month. They instead provide references to answers, hopefully authoritative. They typically contain pieces of the answer, not the complete answer.
Google researchers suggest in a paper that they can make the internet more searchable and the results more accurate through big language models made from machine-learning algorithms that could replace today's system of index, retrieve, and rank.
In the paper -- titled Rethinking Search: Making Experts out of Dilettantes -- the authors sketch out something they call an “aspirational task,” with an example of what this approach might look like.
The answer to the query would provide the benefits and the risks -- something that Google and Microsoft have worked on providing for years.
For example, a user might query, “What are the health benefits of red wine?” The engine would return a detailed answer from an authoritative source that highlight the potential benefits and risks of drinking red wine.
The technology envisioned in this paper would produce a response that provides references to source material making it much easier to highlight the authoritativeness of the content. The example shows how deeper connections between sequences of words and documents can become more useful.
Researchers suggest the next generation of search engines would become a large-scale AI retrieval system.
The model researchers envision can understand terms and documents, and their relationships to each other. They can be trained to generate content with proper citations, but it will have challenges.
The paper discusses several difficult research and engineering challenges such as modeling, training, and response generation that must be solved before the system can be realized. They also include information retrieval, natural language processing, and machine learning research disciplines, and will require interdisciplinary research to become successful.
Marketers should not expect, but prepare, for major changes in search during the next few years. Not only in Google and Microsoft, but Yelp, Facebook, IBM, and other platforms that make retrieval of information and advertising possible.