In a post, we highlighted the current state of AI and explored the first steps your company should take to prepare for AI. This final installment will peruse the top six promising commercial AI use cases.
Netflix' recommendation system is predicated on the fact that humans are very bad at choosing between many options, quickly getting overwhelmed or making poor choices. Research suggests that a typical Netflix member loses interest after 60 to 90 seconds of choosing, having reviewed 10 to 20 titles (perhaps three in detail) on one or two screens. The recommendation algorithm is to make sure that on those two screens each member in Netflix’s diverse pool will find something compelling to watch, and will understand why it might be of interest.
The main components of the recommendation system contain a collection of different machine learning algorithms that define the Netflix experience. Most of them come together on the Netflix homepage. Through machine learning, Netflix can access a large data set —what each member watches and when, the place on the Netflix screen where the customer found the video, recommendations customer didn’t prefer, and the popularity of videos in the catalog. Some of the algorithms being used include: video-to video similarity (Sims) which makes recommendations in the “Because You Watched” row; Top N ranker algorithm which makes recommendations in the “Top Picks” row; Or “evidence” algorithms, which focus on what information to show a viewer about a movie (e.g. if it won an Oscar).
The goal of the Top N algorithm is to find the best few personalized recommendations in the entire catalog for each user, focusing only on the head of the ranking, an element that the Personalized Video Ranker (PVR) does not have since it is used to rank arbitrary subsets of the catalog. The Top N ranker is optimized and evaluated using metrics and algorithms that look only at the head of the catalog ranking that the algorithm produces, rather than at the ranking for the entire catalog (like in the case of PVR). Otherwise, the Top N ranker and PVR share similar attributes, for example, combining personalization with popularity, and identifying and incorporating viewing trends over different time windows ranging from a day to a year. The recommendation system produces $1 billion a year in value from customer retention.
Expert decision support - CAD Workstation
Machine learning is also used to compensate, in some cases, radiologists’ oversight in interpreting mammograms. One study explored using computer-aided diagnosis (CAD) to review early mammograms of women who eventually developed breast cancer. The CAD workstation detected 52% of the cancers a year before they were officially diagnosed. With a CAD workstation, a laser scanner first transforms the mammography film into a detailed matrix of digital data. Microcalcifications show as tiny white spots, and masses appear as round or irregular shapes. The system’s computer vision and AI algorithms scan the digital matrix, sift out background findings and normal soft tissue, and then highlight patterns that are likely to represent lesions. Areas interpreted as suspicious are flagged on the digital mammogram with arrows. After reviewing the mammograms and the computer output, a radiologist prepares a negative or positive report. The results showed that CAD may serve as a second opinion for traditional screening mammograms.
To prevent money laundering, PayPal uses AI technology that is able to accurately detect possible fraud. Algorithms mine data from the customer’s purchasing history— besides reviewing patterns of likely fraud stored in PayPal’s databases—and can tell whether, any suspect transactions were only innocent actions of a globe-hopping pilot accessing an account from different countries. Fraud is always possible via theft of consumer data in breaches like “phishing” emails. That’s why the payments giant relies on intensive, real-time analysis of transactions. When patterns are discovered—like sudden strings of many small purchases at convenience stores that turn out to be fraud—they are turned into a “feature,” or a rule that can be applied in real time to stop purchases that fit this profile. This way, PayPal can tell the difference between friends buying concert tickets together and a thief making similar purchases with a list of stolen accounts. Artificial intelligence approaches have helped keep PayPal’s fraud rate extremely low, at 0.32 percent of revenue.
An Israeli startup, Deep Instinct, has applied deep learning to spot malware. Deep learning involves training a network with many layers of simulated neurons using massive amounts of data. When fed a large number of examples, such a network will correctly identify new examples that seem different on a basic level. A deep learning system can, for example, be trained to recognize a particular person’s face using thousands of images, and then spot that person in new photos, even ones taken in poor lighting or from an odd angle. Research shows that the best commercial antivirus can catch around 87 percent of all new threats several months after the software was last updated. As a result of using deep learning, Deep Instinct’s software was able to detect 20 percent more new malware than existing antivirus software.
Given the complexity of recent advances in AI like cognitive computing, machine learning, and deep learning, companies need to be aware that it is possible to deploy simpler yet effective solutions without breaking the bank. With fewer technical requirements and less time and money, they can get in a position of advantage on AI’s future. Enter “AI Lite” systems, which can be smart about sorting and distributing information very fast. This is what Allstate Business Insurance, a division of Allstate Insurance, used to develop a virtual assistant called ABIe (pronounced “Abby” for Allstate Business Insurance Expert) to answer questions from its 12,000 agents. Since the costs of expanding the call centers were prohibitive, ABIe saved unnecessary spending by using an avatar-driven interface that offers accurate answers to policy questions while streamlining the quote process. Thanks to this tool, in three years the number of queries answered exploded from a few thousand a month to 100,000. A new version of ABIe will be answering queries directly from the customers. However, ABIe is not a ‘one-size-fits-all’ solution. It took Allstate Business almost a year to design and implement ABIe by putting together a team of experts who were responsible for figuring out the “taxonomy” of the words, phrases, and data that ABIe would need to answer all queries. Answers are pulled from a data warehouse, where all the knowledge relating to the company’s products and processes has been organized. The benefit of all these efforts is that by marrying the right data to the right vocabulary and terminology, a company’s information capabilities will soar.
Is also one of the reasons why Hong Kong has one of the world’s best subway systems with a staggering on-time rate of 99.9%. Owned and run by MTR Corporation, it carries around 1.6 billion passengers per year, and it is by far the most popular transportation mode in Hong Kong. Although most tasks are performed by humans, they are scheduled and managed by AI. The AI algorithm works with a simulated model of the entire system to find the best schedule for necessary nightly engineering works. From its all-wise view, it can spot chances to combine work and share resources that no human can. The generated schedule is still subject to human approval. Urgent and unexpected repairs can be added manually – the system only reschedules less important tasks. It also checks the maintenance it plans for compliance with local regulations. The team of experts encoded into machine-readable language 200 rules that the engineers must follow when working at night, like keeping noise below a certain level in residential areas. The AI techniques also make use of several innovative techniques, such as a non-intrusive XML rule-engine, intelligent self-healing coding, and combining heuristic search with a genetic algorithm. This is the first AI system to be deployed in Asia Pacific that uses such innovative techniques as well as a modern service-oriented architecture. Page 9 of 9 The AI solution saves MTR two days a week of squabbling over the repair schedule. And MTR’s repair teams have 30 minutes longer to finish their night’s work – a small-time boost that saves MTR millions per year.
The lack of resources and the gap between research and actual AI usage are the main factors that make many companies hold back when it comes to applying AI. However, the interest in AI is growing and there is a lot of room to reduce this gap with strong data definitions, systemized business processes and ERP systems. Getting a grip on data, ensuring your ERP systems are powerful, and optimizing your business processes are the main steps to follow before implementing any AI-based solution. Data is the foundation of the decision-making process. To maintain a high level of data quality, you should focus on ease-of-use, leaving aside everything that does not add value, good governance and smart automation. AI has been successfully used in many industries from healthcare to fraud detection to intelligent recommendation and data security. Given its wide versatility and applicability, there is no doubt 2017 will be a big year for AI with huge potential for organizations.