How Search Engines Are Increasingly Driven By Machine Learning
Machine learning is an application of artificial intelligence (AI) expected to proliferate even further throughout the business world in 2017. When specific, rule-based algorithms are impossible or ineffective, machine learning is advantageous - smart machines can utilize data by teaching themselves to improve from it. Machine learning’s versatility is such that almost any process could be improved by it. When dealing with business information, maybe no process is more important than search tools, an area where machine learning has been truly revolutionary.
Ranking is Getting Smarter
At the end of 2016, Google search engines were optimized by an expanded integration of machine learning technology. This feature known as Rankbrain self-improves over time, using AI to monitor and adjust its core search algorithms to be consistent with an evolving understanding of semantics. Much earlier Google had introduced features such as “stemming” and the ability to recognize synonyms that widened direct responses to queries, but RankBrain effectively learns from experience and modifies itself. There are conditions that Google imposes on the data that RankBrain integrates (it is spoon-fed selective “offline” queries and information), but RankBrain’s addition improves pre-determined responses.
RankBrain is an important search ranking signa, the third most important signal after content and links in Google’s algorithmic system called Hummingbird. Instead of looking at links or directly at content, RankBrain is able to approximate what words or phrases in a query have similar meaning to others. It groups queries with linguistic similarities together, and maps new ones that seem to match to those groups. Google is pleased with the success of RankBrain, and has continued its investment in neural-network-modeled machine learning. One of the company’s latest advances is a new service it calls Google Neural Machine Translation, which reduces translation errors between certain languages by as much as 85 percent.
The introduction of machine learning has enabled search engines to better understand natural language input. A feature sometimes called “Query Intent Disambiguation” enables a search engine to recognize searches that contain homonyms (for example a search for “desert” could be a search for a food recipe, or a search for arid land), and respond according to applicable forms of concept or semantic ranking.
Machine Learning Goes Deep
Deep learning is a variety of machine learning where a set of algorithms is modeled to simulate the neocortex and used to process complex abstractions in data. Deep Learning has been applied to everything from the identification of faces in photographs posted on Facebook to improving the sophistication of machine translation. “The key thing about neural network models is that they are able to generalize better from the data,” says Microsoft researcher Arul Menezes. “With the previous model, no matter how much data we threw at them, they failed to make basic generalizations. At some point, more data was just not making them any better.”
Microsoft has applied machine learning to its own digital assistant program Cortana. And Facebook wants machine learning to improve its messaging program Chatbox. The head of Facebook’s AI research, Yann LeCun, offers this view on artificial intelligence: “AI is not magic, but we have already seen how it can make seemingly magical advances in scientific research and contribute to the everyday marvel of identifying objects in photos, recognizing speech, driving a car, or translating an online post into dozens of languages.”
How Much Can Machines Learn?
The future capabilities of machine learning remain uncertain. Defense Advanced Research Projects Agency (DARPA) is exploring the limitations of Machine Learning, through a program called FUNLoL (Fundamental Limits of Learning). DARPA is attempting to measure and track this technology, and has found that even small changes in goals requires the programming of new teaching processes. Machine learning programs can't generalize data from older versions, requiring programmers to input significant amounts of new information, restructured to accommodate changes in the rules.
Regardless of how it develops in the future, machine learning is impacting our world now. Every time you type a phrase into a search engine, the information involved becomes part of that learning process. Yet when it comes to drawing conclusions about the results of a search query, artificial intelligence still calls for human intelligence to draw actionable conclusions. As DARPA has found, humans are still needed for machine learning, but as business information continues its explosive growth, humans increasingly need machine learning to find what they seek.
Tame the Ever-Increasing Flow of Information
InfoDesk has created the world’s smartest platform for managing and sharing information. With our comprehensive solutions, you can bring all your information together, filter and select relevant content, and deliver the right intelligence to the right people. InfoDesk has been providing actionable intelligence to multinational corporations, government agencies and other organizations since 1999. InfoDesk is based in New York with offices in London, Washington, DC and India. Learn more about InfoDesk.