The news this week that Google is merging its two AI initiatives, Deep Mind, and Google Brain caused a stir, but commentators have struggled to interpret this development consistently. According to the Financial Times (““Deep Mind reinvents itself for AI counterattack”), this move is in response to Google facing an “existential dilemma”:
Its search unit, which accounted for most of Alphabet’s 2002 annual revenues of $283bn, appears to be under direct threat. Models such as Chat GPT can respond to queries with comprehensive written answers, rather than a set of links. These tools could allow millions of users to bypass online advertising.
The article continues to suggest how difficult it is for a research team to be integrated with a commercial organisation.
Even worse is the unsparing description of Google by veteran commentator Stephen Arnold:
Over the years, my team and I have observed Google’s struggles to innovate … Steve Ballmer allegedly described Google as a “one-trick pony” … Change is difficult for Alphabet …
However, Arnold eventually abandons the attempt to make any sense of this merger. “My team and I attempt to continue tracking innovations in smart software. We cannot do it. Too many people are doing too many new things.”
That’s quite a confession, but also a rather elliptical and probably unintended compliment about Google.
Perhaps we should think for a moment about what Google has achieved with its one main product, the search engine. Over the years, a simple search engine has become progressively more sophisticated. Google in fact frequently responds to queries with comprehensive answers – I heard an estimate that 14% of Google queries today are in the form of natural-language questions. In fact, it goes one better, it jumps from a query to delivering the answer at a relevant site. For example, I queried “Saxon churches in Norfolk”, and I was taken to a map of Norfolk showing churches with substantial Saxon remains. When I ask “How tall is the Eiffel Tower?”, Google responds “300 m, 330m to tip”. That looks like an answer to me.
We take for granted that Google is moving from listings to answering questions, because we are so familiar with the Google interface. We hardly notice when AI features are introduced, and we silently make use of them. There is no fanfare, but behind the scenes, the system has to separate search-type queries to natural-language questions, and then identify what questions can reasonably be answered. If you ask “Did Trump win the last election?” you are taken to a reputable site such as the Wikipedia article on the 2020 presidential election, not (at least in the first few hits) to a site about conspiracy theories and vote-fixing.
By comparison, I spent years trying to persuade reluctant academic publishers to move away from manual tasks to adopt new AI-based tools, only to have them rejected because of uncertainty at not having sufficient human input. Google’s achievement is there: it has moved the world towards an acceptable use of AI, so that AI is in the mainstream without our knowing it. For the most part, Google just works, and you realise how well it works when you talk to users about the search system they want, and they respond that they want it to work like Google.
Next time you read a newspaper article about the threat that AI poses to our civilization, remember that Google silently changed the world of search without any cries of complaint. Perhaps Google has done more to take search beyond Solr and Elasticsearch than anyone else.