Life|Aggregated

Peter Norvig

Director of research at Google

Internet search as we know it is just one decade old; by 2020 it will have evolved far beyond its current bounds. Content will be a mix of text, speech, still and video images, histories of interactions with colleagues, friends, information sources and their automated proxies, and tracks of sensor readings from Global Positioning System devices, medical devices and other embedded sensors in our environment.

The majority of search queries will be spoken, not typed, and an experimental minority will be through direct monitoring of brain signals. Users will decide how much of their lives they want to share with search engines, and in what ways.

The results we get back will be a synthesis, not just a list. For example, today if I ask ‘compare approaches to nuclear fusion’, the major search engines agree that a general encyclopaedia article on fusion power comes first, followed by other similar articles. A decade from now, the result will summarize the major approaches, contrast their differences, automatically translate any foreign documents into my language, and then rank the results by efficacy or place them in a table or chart as appropriate. If I then ask for ‘background mathematics for fusion theory’, I will get an outline for an impromptu course concentrating on the necessary complex analysis, customized to specific applications in fusion and to my level of mathematical understanding. If I stumble, the course will be readjusted to fit my needs, or perhaps the search engine will connect me to a tutor or another student in a similar plight. Interaction with search engines will be an ongoing conversation; one that is integrated with the other ongoing tasks of our lives.

One big challenge for search engines is to implement a measure of quality that is not based solely on popularity. Search engines must determine both relevance (is the item pertinent to the user’s query?) and quality (is the item inherently accurate, useful and understandable, independent of the query?). Current relevance measures do reasonably well. Measures of quality require better models of the concepts and relations expressed in documents and how they relate to the reality of the world, as well as models of the trustworthiness of authors. Thus, a site that claims that the Moon landings were a hoax and seems to have a coherent argument structure will be judged to be lower quality than a legitimate astronomy site, because the premises of the hoax argument are at odds with reality. Understanding and improving these models is a key challenge for the coming decade.

via nature.com

Fascinating look into the future of search. Makes us wonder where we will all be… I originally came to this through http://ahier.posterous.com