In my previous post I talked about how to identify the opportunities for AI to solve key operational challenges and deliver on organization goals. As I described there, a good place to start is by understanding the categories of AI and ML use cases.
The first half-dozen categories were classification, prediction, optimization, robotics and control, monitoring and anomaly detection / situational awareness, and decision making. Here in Part II, I round out the remaining six use case categories.
7. Discovery and diagnosis: Some of the most exciting and important work being done in AI today is happening in the field of health care, especially in the data-intensive area of medical diagnostics. (Note that this area substantially overlaps with classification, prediction, and monitoring methods, as described in the previous post.) Diagnostic applications look at patient imaging or diagnostic data to decide, for instance, whether an image is normal or abnormal. About 90 percent of all healthcare data comes from imaging technology. In his book, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, Eric Topol talks about how the precision and accuracy of new machine learning technologies will improve disease diagnosis and better identify optimal therapies. If used correctly, this technology can bring more humanity back to the practice of medicine.
As a few examples, AI start-ups like Freenome are applying advanced machine learning techniques to recent breakthroughs in genomic science to develop noninvasive blood tests for detection of early-stage cancers, and to improve precision oncology treatments for patients. Heartflow analyzes a CT scan of a patient’s heart to determine whether they are high enough risk to warrant opening up their chest for an invasive look at their heart health. Traige.AI looks at an image of a persons skin taken with the cellphone and classifies the into 120 different known skin conditions.
Diagnosis is also extremely important in complex engineered systems, such as cars, planes, and spacecraft, where there is an observed problem that could be caused by any of a large set of potential failures and engineers need to decide what to look for and what to fix.
8. Search and question answering: This is a use case category that many people associate with AI: think Google search with questions. Essentially it is the field of information retrieval and natural language processing (NLP) focused on building systems that automatically answer questions that humans pose in a natural language.
9. Information extraction, annotation, and summarization: This category is about processing human-generated language text by extracting structured information from unstructured and/or semi-structured machine-readable documents. Using natural language processing (NLP), a typical example is an HR application that matches resumes to job applicants, or matches information about a court case to find the best lawyer to prosecute it.
With the proliferation of multimedia content, this category also includes automatic annotation and content extraction from images, audio, video, and documents. For instance, SAP’s Brand Impact identifies logos in video images for purposes of validating marketing contracts.
10. Translation: Air travel may have made the world a smaller place, but real-time, text-, audio-, and image-based translation apps are now closing the comprehension gap by bringing faster and easier foreign-language translation to the smartphone-carrying masses. Beyond familiar consumer AI applications like Google Translate, Unbabel is creating a platform that combines the speed and scale of machine translation with the authenticity of native speakers. Unbabel helps businesses to filter their content through its “localized branding engines” for tailored, multilingual support at scale and at enterprise-level standards.
11. Matching and recommendation: If you’ve experienced a playlist of videos or music served up by the likes of Netflix, YouTube, Pandora, or Spotify; received a recommendation from Amazon for a certain book or product based on a recent purchase or browsing history; or had varied news or other content pop up on your social media feed, chances are it’s because of an AI matching or recommendation engine. These systems put people and things together using a single input, like music, or multiple inputs within and across platforms like news, books, and search queries. Online dating, travel, and restaurant sites also represent the growing list of users of recommender systems. Healthcare providers are also increasingly using AI-enabled matching apps like Deep 6 AI to match patients to clinical trials by applying AI to medical records to find more, better-matching patients for clinical trials in a matter of minutes, rather than months.
12. Workflow Automation: Business processes help organizations get things done. Supporting them, AI process automation tools simplify workflows. Robotic Process Automation (RPA), in particular, automates repetitive, rules-based, and high-volume transactions. In many cases RPA applications allow organizations to upskill their employees and shift them to more high-value tasks. Companies like Blueprism and UiPath have developed RPA platforms across multiple verticals to automate business processes and to drive higher productivity and accuracy to very low or zero error rates for routine, manual tasks. These tasks include invoice, receipt, and expense processing, new customer sign-up and onboarding, and supply chain management. As the routine and easy cases get tackled, the frontier of this field is Intelligent Process Automation (IPA), as described in this recent Gartner report.Kyndi’s IPA offering, for instance, consists of providing bots that analyze text and automate inefficient workflows in industries such as Insurance by analyzing medical records for class action lawsuits or understanding drug interactions for pharmacovigilance applications.
