Artificial Intelligence Applications
The goal of AI can be to create smart machines that think and act like humans, with the ability to simulate intelligence. In this blog, we explore 20+ new artificial intelligence(AI) applications used in 2019.
Big data mining
Using data mining and machine learning methods and algorithms, it is possible not only to detect unusual patterns or anomalies in data but also to predict them.
In order to acquire this kind of knowledge from big datasets, either supervised or unsupervised machine learning techniques may be used. Supervised machine learning can be thought of as roughly comparable to learning from the example in humans.
Using training data, where correct examples are labeled, a computer program develops a rule or algorithm for classifying new examples. This algorithm is checked using the test data.
In contrast, unsupervised learning algorithms use unlabelled input data and no target is given; they are designed to explore data and discover hidden patterns.
As an example let’s look at credit card fraud detection, and see how each method is used.
Credit card fraud detection
A lot of effort goes into detecting and preventing credit card fraud. If you have been unfortunate enough to receive a phone call from your credit card fraud detection office, you may be wondering how the decision was reached that the recently made purchase on your card had a good chance of being fraudulent.
Given the huge number of credit card transactions, it is no longer feasible to have humans checking transactions using traditional data analysis techniques, and so big data analytics are increasingly becoming necessary.
Understandably, financial institutions are unwilling to share details of their fraud detection methods since doing so would give cybercriminals the information they need to develop ways around it. However, the broad brush strokes present an interesting picture.
There are several possible scenarios but we can look at personal banking and consider the case in which a credit card has been stolen and used in conjunction with other stolen information, such as the card PIN (personal identification number). In this case, the card might show a sudden increase in expenditure—a fraud that is easily detected by the card issuing agency.
More often, a fraudster will first use a stolen card for a ‘test transaction’ in which something inexpensive is purchased. If this does not raise any alarms, then a bigger amount is taken.
Such transactions may or may not be fraudulent— maybe a cardholder bought something outside of their usual purchasing pattern, or maybe they actually just spent a lot that month.
So how do we detect which transactions are fraudulent? Let’s look first at an unsupervised technique, called clustering, and how it might be used in this situation.
Based on artificial intelligence algorithms, clustering methods can be used to detect anomalies in customer purchasing behavior. We are looking for patterns in transaction data and want to detect anything unusual or suspicious which may or may not be fraudulent.
A credit card company gathers lots of data and uses it to form profiles showing the purchasing behavior of their customers. Clusters of profiles with similar properties are then identified electronically using an iterative (i.e. repeating a process to generate a result) computer program.
For example, a cluster may be defined on accounts with a typical spending range or location, a customer’s upper spending limit, or on the kind of items purchased, each resulting in a separate cluster.
When data is collected by a credit card provider it does not carry any label indicating whether the transactions are genuine or fraudulent. Our task is to use this data as input and, using a suitable algorithm, accurately categorize transactions. To do this, we will need to find similar groups, or clusters, within the input data.
So, for example, we might group data according to the amount spent, the location where the transaction took place, the kind of purchase made, or the age of the cardholder.
When a new transaction is made, the cluster identification is computed for that transaction and if it is different from the existing cluster identification for that customer, it is treated as suspicious. Even if it falls within the usual cluster, if it is sufficiently far from the center of the cluster it may still arouse suspicion.
For example, say an 83-year-old grandmother living in Pasadena purchases a flashy sports car; if this does not cluster with her usual purchasing behavior of, say, groceries and visits to the hairdresser, it would be considered anomalous. Anything out of the ordinary, like this purchase, is considered worthy of further investigation, usually starting by contacting the card owner.
Classification, a supervised learning technique, requires prior knowledge of the groups involved. We start with a dataset in which each observation is already correctly labeled or classified.
This is divided into a training set, which enables us to build a classification model of the data, and a test set, which is used to check that the model is a good one. We can then use this model to classify new observations as they arise.
Storing big data
Computer technology progressed rapidly and by the start of the personal computer boom in the 1980s, the average hard drive on a PC was 5 Mb when one was included, which was not always the case.
This would hold one or two photos or images today. Computer storage capacity increased very quickly and although personal computer storage has not kept up with big data storage, it has increased dramatically in recent years.
Now, you can buy a PC with an 8 Tb hard drive or even bigger. Flash drives are now available with 1 Tb of storage, which is sufficient to store approximately 500 hours of movies or over 300,000 photos. This seems a lot until we contrast it with the estimated 2.5 Eb of new data being generated every day.
Once the change from valves to transistors took place in the 1960s the number of transistors that could be placed on a chip grew very rapidly, roughly in accordance with Moore’s Law, which we discuss in the next section. And despite predictions that the limit of miniaturization was about to be reached it continues to be a reasonable and useful approximation.
We can now cram billions of increasingly faster transistors onto a chip, which allows us to store ever greater quantities of data, while multi-core processors together with multi-threading software make it possible to process that data.
David House, a colleague at Intel, after taking into account the increasing speed of transistors, suggested that the performance of microchips would double every eighteen months, and it is currently the latter prediction that is most often used for Moore’s Law.
This prediction has proved remarkably accurate; computers have indeed become faster, cheaper, and more powerful since 1965, but Moore himself feels that this ‘law’ will soon cease to hold.
Moore’s Law is now also applicable to the rate of growth for data as the amount generated appears to approximately double every two years. Data increases as storage capacity increases and the capacity to process data increases.
We are all beneficiaries: Netflix, smartphones, the Internet of Things (IoT; a convenient way of referring to the vast numbers of electronic sensors connected to the Internet), and the Cloud computing, among others, have all become possible because of the exponential growth predicted by Moore’s Law.
‘Data scientist’ is the generic title given to those who work in the field of big data. The McKinsey Report of 2012 highlighted the lack of data scientists in the USA alone, estimating that by 2018 the shortage would reach 190,000.
The trend is apparent worldwide and even with government initiatives promoting data science skills training, the gap between available and required expertise seems to be widening.
Data science is becoming a popular study option in universities but graduates so far have been unable to meet the demands of commerce and industry, where positions in data science offer high salaries to experienced applicants.
Big data for commercial enterprises are concerned with profit, and disillusionment will set in quickly if an over-burdened data analyst with insufficient experience fails to deliver the expected positive results.
All too often, firms are asking for a one-size-fits-all model of data scientist who is expected to be competent in everything from statistical analysis to data storage and data security.
Data security is of crucial importance to any firm and big data creates its own security issues. In 2016, the Netflix Prize 2 initiative was canceled because of data security concerns.
Other recent data hacks include Adobe in 2013, eBay and JP Morgan Chase Bank in 2014, Anthem (a US health insurance company) and Carphone Warehouse in 2015, MySpace in 2016, and LinkedIn —a 2012 hack not discovered until 2016.
This is a small sample; many more companies have been hacked or suffered other types of security breaches leading to the unauthorized dissemination of sensitive data.
On 7 December 2016, Amazon announced that it had made its first commercial drone delivery using GPS (global positioning system) to find its way. The recipient, a man living in the countryside near Cambridge in the UK, received a package weighing 4.7 pounds.
Drone deliveries can currently be made to only two Amazon Prime Air customers, both living within 5.2 square miles of the fulfillment center near Cambridge. A video, referenced in the Further reading section, shows the flight. This seems likely to signal the start of big data collection for this program.
Amazon is not the first to make a successful commercial drone delivery. In November 2016, Flirtey Inc. started a drone delivery pizza service in a small area from their home base in New Zealand and there have been similar projects elsewhere. At present, it seems likely that drone delivery services will grow, particularly in remote areas where it might be possible to manage safety issues.
Of course, a cyber-attack or simply a breakdown in the computer systems could create havoc: if, for example, a small delivery drone were to malfunction, it could cause injury or death to humans or animals, as well as considerable damage to property.
This is what happened when the software controlling a car traveling along the road at 70 mph was taken over remotely. In 2015, two security experts, Charlie Miller, and Chris Valasek, working for Wired magazine, demonstrated on a willing victim that Uconnect, a dashboard computer used to connect a vehicle to the Internet, could be hacked remotely while the vehicle was in motion.
The report makes alarming reading; the two expert hackers were able to use a laptop Internet connection to control the steering, brakes, and transmission along with other less critical functions such as the air-conditioning and radio of a Jeep Cherokee.
The Jeep was traveling at 70 mph on a busy public road when suddenly all response to the accelerator failed, causing considerable alarm to the driver.
As a result of this test, the car manufacturer Chrysler issued a warning to the owners of 1.4 million vehicles and sent out USB drives containing software updates to be installed through a port on the dashboard.
The attack was made because of a vulnerability in the smartphone network that was subsequently fixed, but the story serves to illustrate the point that the potential for cyber-attacks on smart vehicles will need to be addressed before the technology becomes fully public.
The advent of autonomous vehicles, from cars to planes, seems inevitable. Planes can already fly themselves, including taking off and landing.
Although it’s a step away to think of drones being in widespread use for transporting human passengers, they are currently used in farming for intelligent crop spraying and also for military purposes. Smart vehicles are still in the early stages of development for general use but smart devices are already part of the modern home.
the term ‘Internet of Things’ (IoT) is a convenient way of referring to the vast numbers of electronic sensors connected to the Internet. For example, any electronic device that can be installed in a home and managed remotely—through a user interface displayed on the resident’s television screen, smartphone, or laptop—is a smart device and so part of the IoT.
Voice-activated central control points are installed in many homes that manage lighting, heating, garage doors, and many other household devices. Wi-Fi (which stands for ‘wireless fidelity’, or the capacity to connect with a network, like the Internet, using radio waves rather than wires) connectivity means that you can ask your smart speaker (by its name, which you will have given it) for the local weather or national news reports.
These devices provide Cloud-based services and are not without their drawbacks when it comes to privacy. As long as the device is switched on, everything you say is recorded and stored in a remote server.
In a recent murder investigation, police in the United States asked Amazon to release data from an Echo device (which is voice controlled and connects to the Alexa Voice Service to play music, provide information, news reports, etc.) that they believed would assist them in their inquiries.
Amazon was initially unwilling to do so, but the suspect has recently given his permission for them to release the recordings, hoping that they will help prove his innocence.
Further developments, based on Cloud computing, mean that electrical appliances such as washing machines, refrigerators, and home-cleaning robots will be part of the smart home and managed remotely through smartphones, laptops, or home speakers. Since all these systems are Internet controlled they are potentially at risk from hackers, and so security is a big area of research.
Even children’s toys are not immune. Named ‘2014 Innovative Toy of the Year’ by the London Toy Industry Association, a smart doll called ‘My Friend Cayla’ was subsequently hacked. Through an unsecured Bluetooth device hidden in the doll, a child can ask the doll questions and hear replies.
The German Federal Network Agency, responsible for monitoring Internet communications, has encouraged parents to destroy the doll, which has now been banned, because of the threat to privacy that it presents. Hackers have been able to show that it is fairly easy to listen to a child and provide inappropriate answers, including words from the manufacturer’s banned list.
Although the smart home is only just becoming a reality, the IoT together with multiple information and communication technologies (ICTs) are now predicted to make smart cities a reality. Many countries, including India, Ireland, the UK, South Korea, China, and Singapore, are already planning smart cities.
The idea is that of greater efficiency in a crowded world since cities are growing rapidly. The rural population is moving to the city at an ever-increasing rate.
In 2014, about 54 percent lived in cities and by 2050 the United Nations predicts that about 66 percent of the world’s population will be city dwellers.
The technology of smart cities is propelled by the separate but accumulating ideas from early implementations of the IoT and big data management techniques. For example, driverless cars, remote health monitoring, the smart home, and telecommuting would all be features of a smart city.
Such a city would depend on the management and analysis of the big data accumulated from the sum total of the city’s vast sensor array. Big data and the IoT working together are the keys to smart cities.
For the community as a whole, one of the benefits would be a smart energy system. This would regulate street lighting, monitor traffic, and even track garbage.
All this could be achieved by installing a huge array of radio-frequency identification (RFID) tags and wireless sensors across the city. These tags, which consist of a microchip and a tiny antenna, would send data from individual devices to a central location for analysis.
For example, the city government would monitor traffic by installing RFID tags on vehicles and digital cameras on the streets. Improved personal safety would also be a consideration, for example, children could be discretely tagged and monitored through their parents’ cell phones.
These sensors would create a huge amount of data which would need to be monitored and analyzed in real-time, through a central data processing unit.
It could then be used for a variety of purposes including gauging traffic flow, identifying congestion, and recommending alternative routes. Data security would clearly be of paramount importance in this context, as any major breakdown in the system or hacking would quickly affect public confidence.
Songdo International Business District in South Korea, scheduled for completion in 2020, has been purpose built as a smart city. One of the main features is that the entire city has fiber-optic broadband. This state-of-the-art technology is used to ensure the desired features of a smart city can be accessed quickly.
New smart cities are also being designed to minimize negative environmental effects, making them the sustainable cities of the future. While many smart cities have been planned and, like Songdo, are being purpose-built, existing cities will need to modernize their infrastructures gradually.
In May 2016, the United Nations Global Pulse, an initiative aimed at promoting big data research for global benefit, unveiled its open ‘Big Ideas Competition 2016: Sustainable Cities’ for the ten member states of the Association of Southeast Asian Nations (ASEAN) and the Republic of Korea.
The other side of the search coin is “organic” or unpaid results. This is the consequence of having just what the searcher is looking for in the eyes of the search engine, be it Google, Bing, or Yahoo.
This has given rise to the content marketing movement where marketers have become publishers. If you want people to find your Fluorescent Lamp and Ballast Tester Kit, you would be well served to post lots of information about it, answers to frequently asked questions, and generally, be the place on the Internet that Google points to first when asked.
The first concern is whether Google doubts your legitimacy. If Google thinks you’re spamming its engine, it will summarily boot you off its results page. To see how safe your content is, Safeco not offers a crawler that reviews your site and assigns a score on each page based on the risk it runs of being penalized.
Content Management—Image Matters
Past the obstacle of Google as adjudicator of authenticity is the issue of Google as judge of relevancy. While raking in money hand-over-fist selling keyword-related ads, Google keeps people coming back by serving up the world’s most applicable links.
To be deemed relevant, a company must become very adept at publishing everything there is to know about those fluorescent lamp and ballast tester kits. AI is at the ready to help out. A standard search through an enormous library of white papers, PDF fliers, web pages, and research documents might yield a list of relevant results.
But AI’s visual and audio recognition capabilities come in very handy here to sift through all that other, unstructured data stored in otherwise unfindable files.
The creative soul, asked to illustrate an article on the safety issues around rewiring a fluorescent lamp, can only hope to use the same, tired old images that were properly labeled by some other poor creative soul. Machine learning, however, might find and offer up a range of options.
Martin Jones, senior marketing manager at Cox Communications, gave IBM’s Watson Content Hub a spin at the end of 2016 to witness the practical use of machine learning in creating marketing campaigns.
Rather than just listing images that matched search criteria, Watson was able to understand the meaning of the marketing message Jones wanted to convey. Further, Watson assisted in creating a customized experience for each website or mobile app visitor.
Like the American Marketing Association’s efforts with Lucy, Jones first acquainted Watson with Cox’s images and other digital assets. That process included the automatic image resizing and cropping by classification profiles with Watson recommending tags as it recognized the contents of pictures, including product names.
Jones was able to identify conditions and variants for analyzing multiple images for real-time personalization, which makes use of that resizing and cropping for delivery across multiple devices.
Other technologies are in the picture to help you with your pictures—or even user-generated content. Olapic helps you find photos of your products out in the wild that might be useful in your ads or on your site. They also offer Photorank, which “evaluates multiple data points to accurately predict engagement and conversion per each image/video.”
Curalate can find good photos and then secure permission to use them with a hashtag-based rights management system.
Infinigraph will help you pick the best thumbnail for your video. Somatic can look at a picture and create a short description in different styles. Want your description to imitate a celebrity? They’ve got you covered.
Pinterest is working with visual recognition in photos to help you buy things you see in real life. Just snap a picture with the Pinterest app and it suggests things that look like the object of your affection. I offered Pinterest a simple challenge on my first try, the napping chair in my office.
Content Consumption Analysis
Behavioral analytics comes in handy here, too. From the first log file analysis tools, we’ve been minutely dissecting what pages, categories, and specific products interest individuals. But now, the machine can help us out.
Not only can the machine see what interests individuals, but it can also determine how interested they are and where they might be in the buying cycle. That generates probabilities on what content to offer next and triggers a next-best-offer.
This capability works across the Internet at large as well. Charlie Tarzian, Founder of The Big Willow, says that content bingeing in 15–20-minute bursts can tell you a lot about where someone is in a buying process.
The company uses pattern matching and machine learning to find people “across more than 10,000 B2B websites, blogs, and communities and engage them.”
SOCIAL MEDIA ENGAGEMENT
Monitoring social media to get a clear picture of what the market-place is saying about you is one thing. Engaging with social media is another. One is a passive, observational task; the other is reaching out—reactively or proactively—to engage people where they tweet.
eMarketer interviewed AJ Mazza, director of Marketing Communications, and Dedra DeLilli, director, Social Media Marketing and Corporate Sponsorships at TD Ameritrade, in an article in February 201715 about their use of AI.
Through a partnership with Havas Cognitive, we developed a social media promotion to support an NFL campaign that we put together last season. We scoured the social feeds of opted-in consumers to assess their confidence in their favorite NFL team.
It analyzed the tone and words that consumers used in their social posts to produce an aggregated confidence score. This score showed how confident a consumer was compared to other fans. Our goal was to drive engagement by giving fans the opportunity to increase their confidence score through specific actions, such as sharing our content, in an effort to win a prize—a trip to the Super Bowl.
The technology enabled us to put a fun twist on a typical investing questionnaire. When we match customers with a product or service, we provide questionnaires to evaluate risks, and Watson allowed us to make this more engaging.
It helped articulate TD Ameritrade’s brand message about providing tools and resources that make customers confident investors. The campaign also allowed us to test AI with minimal risk and delivered learnings that are being applied to benefit other areas of the business.
AI can help you find people, figure out what they’re up to, and what interests them. It can also help you communicate. Welcome to the world of bots. Google’s Smart Reply goes a step further and offers you a few suggestions on how you might like to respond.
Lots of bots answer straight questions with pre-written answers but hand over a Twitter account to a socialbot, train it well, and you have the equivalent of a voice response system at the call center. Train it even better and you have a brand representative.
For now, socialbots are only about as useful as voice response systems that have answers for 90 percent of the questions they usually get and no answers for the question you have whenever you call. Socialbots are starting to take on the Turning Test with modest success, but in time they will be able to represent your brand as the first line of defense.