Organizations are increasingly becoming insight-driven. Together with the rapid development of AI technologies like machine learning and natural language processing (NLP), the demand for AI-enabled search and analytics solutions will become more prevalent in the enterprise – a trend I recently discussed in Forbes’ 120 AI Predictions for 2019*. As we begin 2019, let’s look at some of the most notable observations around these capabilities.
Achieve meaningful results sooner with Pragmatic AI
There’s a lot of talk around AI, but let’s first define Artificial Intelligence (AI) in our context before we delve into how it can be applied to enterprise use cases.
According to a survey published in the Journal of Artificial Intelligence Research, AI experts predict a 50% chance that AI will replace all human jobs and perform full labor automation in about 125 years. That is “pure AI,” which has a long way to achieving tasks mirroring human intelligence.
<<< Start >>>
Extracted from “When Will AI Exceed Human Performance? Evidence from AI Experts”**
<<< End >>>
But rather than “pure AI,” we should consider the practical AI technologies that today’s organizations can use to produce results in the short-run. In the enterprise context, we are talking about “pragmatic AI” – a collection of multiple technologies that mirror human brainpower and behavior. Compared to “pure AI,” “pragmatic AI” is narrower in scope and encompasses deep learning, NLP, search, machine learning, intelligent sensors, and robotics.
<<< Start >>>
Businesses with a pragmatic vision will focus on how to integrate these AI technologies to augment specific humans’ tasks and achieve real outcomes.
<<< End >>>
Transform the workplace with AI-powered search and analytics
As Google changes the consumer expectations for search and enterprises become more data-driven, traditional keyword search will no longer meet enterprise users’ needs. The combination of search and technologies like machine learning and NLP (collectively known as Cognitive Search) is enabling enterprises to deliver a better experience – and more insight – to users. I’ve noticed growing demands for AI-enabled search and analytics solutions in multiple client use cases.
<<< Start >>>
<<< End >>>
For instance, in the Oil and Gas industry, we are helping a global client develop an AI-powered knowledge platform for well drilling planning, design, and incident reporting. This platform would rely on search and NLP to extract and normalize ambiguous content that lacks structure, such as text documents, images, seismic surveys, well logs, exploratory drilling reports, and more. The results would be real-time insights at the user’s fingertips, eliminating the manual review process and allowing users to focus on strategic tasks.
Another example is in the Recruiting industry. Previously, recruiters were using traditional search to look through millions of resumes to match candidates to jobs posted by hiring companies. But, how could we improve recruiters’ experience and business results? Well, search interfaces continued to evolve, allowing recruiters to filter and refine search results. For instance, they can use facets like maps to account for the distance between the candidate and the job. In the most sophisticated search and match applications, custom machine learning algorithms can analyze the text in CVs and job postings to provide a suggested list of the best candidates – automating the task of sifting through millions of resumes. Recruiters could also provide feedback into what they thought was the most successful match so that the algorithms can “learn” the patterns and improve future results. The outcomes were universally positive. People got jobs faster. Hiring companies were happy. Recruiters made more money.
These use cases are the type of search and analytics evolution we’re seeing. They take manual data-heavy tasks and apply “pragmatic AI” technologies to enable task automation or adding intelligence to the process. As a result, workers are not necessarily replaced, but their decision-making and efficiency are enhanced. The success of these applications relies on understanding what is in the data as well as how that data can benefit the user. But given the nature of enterprise content, this is not a challenge easily solved.
Automate unstructured content analysis to drive knowledge discovery
<<< Start >>>
of enterprise content is unstructured. It consists of fast-growing human-generated content, including memos, emails, text documents, research and legal reports, voice recordings, videos, social media posts, etc.
<<< End >>>
Unlike structured data (tables, forms, log files), it is difficult to search for, let alone analyze, meaningful information from unstructured data.
Fortunately, there are existing technologies for acquiring, processing, and tagging massive unstructured content to make it available for search and analysis.
Adding NLP and machine learning capabilities, we can automate unstructured content analysis processes: extracting entities (people, locations, companies, etc.), identifying sentiment (can we tell if our customers like our products based on their online reviews?), and categorizing topics (documents with certain indicators should be routed to the corresponding departments for processing).
These “pragmatic AI” technologies have also powered chatbots or virtual assistants that analyze the user’s queries and intent to deliver responses or fulfill tasks. Well-implemented enterprise chatbots can do simple tasks, improve customer services, and boost user productivity, internal and external alike.
It doesn’t stop there. While enterprise chatbots are thriving, they mostly support narrow-domain, low-complexity tasks like order fulfillment and basic customer requests. We can do more with “pragmatic AI” to enrich the user experience.
Enrich the knowledge-seeking experience for users
Going beyond chatbots, I expect to see NLP and machine learning fuel more powerful applications: question-answer (QA) systems. While harder to develop, enterprise QA systems will ultimately be more useful as they support expansive domains and natural language understanding (NLU) capabilities. Let’s look at how this is already happening in our daily Internet search experience.
Google has pioneered question-answer capabilities in an effort to transform its “search engine” into a “knowledge engine” with Google Knowledge Graph. A Google search for “How tall is Mount Everest?” will return not only its elevation but also a snapshot of things you’d likely want to know about the mountain. This is enabled by Mount Everest’s knowledge graph node – a compilation of data about Mount Everest (description, first ascenders, mountain range, did you know, etc.) from various Internet sources (Wikipedia, Wikidata, news articles, websites, etc.).
As Google advances the knowledge-seeking experience, businesses will need to match their users’ rising expectations. So how would an enterprise QA system powered by knowledge graphs look like?
Take an internal search application for example. A question like “Who are our Analytics experts in North America?” would return not only a list of the experts’ names, locations, and skills, but also a complete snapshot of the projects, collaborators, and clients associated with each expert. This is achieved by bringing data fragments from multiple databases together (e.g. Employee Resumes, Project Documents, Client Reports, etc.) and then create meaningful connections between them. For instance, NLP can extract:
- Names, locations, skills, and education from the Employee Resumes database
- Project names, project requirements, and the people working on a project from the Project Documents database
- Client names, associated projects, and associated employees from the Client Reports database
From those fragmented data points, NLU algorithms can create an interconnected information network indicating how each employee, project, and client is linked to each other. This network is the enterprise knowledge graph, from which the QA system can instantaneously pull a snapshot of connected information and deliver relevant insight when the user asks a question. The knowledge graph will be ever-growing as new data points and insightful relationships are added indefinitely.
And with the rapid development of NLP/NLU and graph technologies, we will soon be able to answer highly complex questions faster and more accurately – think detecting financial fraud, identifying market trends, finding medical treatments, etc.
Think practicality and outcomes over buzzwords
Looking ahead into 2019 and beyond, I expect “pragmatic AI” technologies – NLP, machine learning, and cognitive search – to further evolve and be leveraged more broadly.
While technical leadership may have a full understanding of the technology, there's still a knowledge gap between understanding the capabilities versus translating them into business benefits. Thus, the emphasis should not be placed on the buzzwords but on how to apply these intelligent technologies to produce meaningful information discovery and analytics outcomes. A successful integration takes time, investment, domain-specific knowledge, and technical expertise. But with practical strategy and expert implementation, organizations can unlock endless opportunities for innovation.
What’s your strategy for leveraging AI-enabled search and analytics applications?
<<< Start >>>
<<< End >>>
* 120 AI Predictions For 2019, Gill Press, Forbes, 9 December 2018
** When Will AI Exceed Human Performance? Evidence from AI Experts, Katja Grace, John Salvatier, Allan Dafoe, Baobao Zhang, Owain Evans, Journal of Artificial Intelligence Research, 31 July 2018