by Bernard Marr
Using traditional computer storage infrastructure, it just wouldn't be possible to access enough historical transactional data quickly enough to make an accurate prediction in an acceptably short timeframe.
American Express's fraud detection systems used a combination of supervised and unsupervised learning techniques to become increasingly efficient at raising flags when data indicative of fraudulent transactions is encountered.
What Were The Results?
Analyzing transactions in real time using machine learning algorithms means that there is a greater chance that fraudulent transactions will be detected.
It also means there is less chance of false positives occurring, which are inconvenient for customers and may make them less willing to trust the American Express payment system for making their transactions.
More fraudulent transactions will be blocked as they occur, meaning it is less likely that remedial action, such as recovering spent funds, will have to be taken after the fact, drastically reducing the costs involved in dealing with fraud.
Key Challenges, Learning Points And Takeaways
Machine learning models for detecting fraud need to constantly adapt and update themselves in real time, meaning they need a consistent flow of data to learn from.
Distributed storage and large amounts of compute power are needed to handle the amount of data that is needed to make accurate predictions in real time.
The large number of transactions processed by American Express means that small increases in efficiency can make big improvements to overall security.
As well as fraud detection, financial services businesses are looking to AI for the added value it can give to customers, leading to changes in the way they can access their services.
Notes
1American Express, Company 2018 Investor Day: http://ir.american- express.com/Cache/1001233287.PDF?O=PDF&T=&Y=&D=&FID= 1001233287&iid=102700
2Forbes, The World's Most Valuable Brands: https://www.forbes.com/ powerful-brands/list/
3The Nilson Report, Card Fraud Losses: https://nilsonreport.com/upload/ content_promo/The_Nilson_Report_Issue_1118.pdf
4Mapr, New Age Fraud Analytics: Machine Learning on Hadoop: https:// mapr.com/blog/new-age-fraud-analytics-machine-learning-hadoop/
5American Express, American Express Acquires Mezi: https://about. americanexpress.com/press-release/american-express-acquires-mezi
6Mapr, Machine Learning at American Express: Benefits and Requirements: https://mapr.com/blog/machine-learning-american-express- benefits-and-requirements/
32
Elsevier: Using Artificial Intelligence To Improve Medical Decisions And Scientific Research
Elsevier is a global multimedia publishing business that offers more than 20,000 products for educational and professional science and healthcare communities, including leading research publications such as The Lancet and Cell.
Stage one of the company's ongoing digital transformation has involved the digitization of the huge amount of data published in reports and journals during the company's 140-year history.
Now it is building artificial intelligence (AI) tools that will draw new insights from this data, as well as combining it with other big data sources such as anonymized patient data and data from insurance claims.
What Problem Is Artificial Intelligence Helping To Solve?
In the United States, it can often be the case that two patients of the same age and gender will present to their primary healthcare practitioner with the same symptoms, and yet there will be a huge variation in the outcome, and cost, of the treatment they receive.1
This is because diagnosing and treating are done by different healthcare staff with different levels of knowledge and experience, as well as personal feelings about which treatments are more effective and how to achieve preferred outcomes.
By developing AI-derived “pathways” from initial presentation and examination to treatment procedures and prescribing of medication, patients are more likely to get better quicker, and the cost of providing healthcare is reduced.
How Is Artificial Intelligence Used In Practice?
Elsevier is building what it calls its advanced clinical decision support platform, which uses natural language processing and machine learning to suggest the optimal treatment pathway for patients.
The system builds on its Via Oncology platform, which is currently deployed in leading cancer centers around the United States. It is able to correlate data from patient records as well as Elsevier's vast archive of research published through its medical journals.
It then looks for previous cases where patients have reported the same sets of symptoms, and analyzes what outcomes were achieved. It is then able to suggest the treatment that is most likely, based on the data, to have a positive outcome for the patient.
I spoke to John Danaher, president of clinical solutions at Elsevier, who told me that moving forward with development of platforms that combine AI analytics with their vast data sources is currently a key business priority.
Elsevier used all of its content – books, journal articles, etc. – to map diseases to symptoms, which allowed it to create predictive models. The company then trained its neural network models against large patient databases to create models that can generate a differential diagnosis. The model can then give weighted predictions that these particular symptoms in a person of this age and gender give you a 70% chance it's disease A or a 35% chance it's disease B.
What Tools, Technology And Data Were Used?
Elsevier's platform uses anonymized patient data, including medical histories, treatment histories and outcomes. It also uses a database of 5 million medical insurance claims. Then it throws in all of the articles and research published in its journals over the last 140 years.
To carry out analysis on this data it has built its own proprietary analytics tools. These tools utilize natural language processing to understand the contents of the medical literature in its database, as well as the patient records.
The company is evaluating commercial big data and AI solutions such as those offered by Microsoft and Amazon for the next stage of its AI rollout, Danaher tells me.
As well as its advanced clinical decision support platform, Elsevier applies AI to research solutions outside of the healthcare ecosphere, such as its Science Direct tool. This tool also uses the published corpus of scientific literature, and supports researchers by pointing them in the direction of relevant publications and articles that the AI predicts will be relevant to their work.2
What Were The Results?
The best indicator of the results, says Danaher, is the adherence rate of 85% among clinical staff to the treatment pathways suggested by its Via Oncology platform.
He says: “You want to know that if you go to MD Anderson, you're going to get the most current care given to you by the smartest tools – oh, and by the way, that you're going to get the best outcomes.
“So, we get over 85% adherence to our pathways by our clinicians, and when they do go off pathway, which happens sometimes – patients may have allergies to certain medicines – we review and look at the reasons for them going off pathway and if necessary review our decision making.
“You can see the ramifications for how people will do clinical research in the future too – it's all going to be driven by these analytics.”
Key Challenges, Learning Points And Takeaways
Elsevier amalgamates patient medical records, insurance claims and billing data and published medical literature to predict which treatment pathways are most likely to be effective.
Elsevier owns 25% of the US output of published scientific and medical research. It has looked to AI to develop new methods of drawing value from this information.
Treatment can be standardized if machines are used to determine optimal treatment paths dependent on the patient's details, medical history and the symptoms they present with.
Standardized treatments lead to better patient outcomes if
they can be optimized according to data, and also help healthcare providers to reduce overall costs.
Notes
1Wall Street Journal, Mayo Clinic’s Unusual Challenge: Overhaul a Business That’s Working: https://www.wsj.com/articles/mayo-clinics-unusual-challenge-overhaul-a-business-thats-working-1496415044
2LinkedIn, Artificial Intelligence And Big Data: The Amazing Digital Transformation Of Elsevier From Publisher To Tech Company: https:// www.linkedin.com/pulse/artificial-intelligence-big-data-amazing-digital -elsevier-marr/
33
Entrupy: Using Artificial Intelligence To Combat The $450 Billion Counterfeit Industry
Launched in 2016, Entrupy uses artificial intelligence (AI) to combat counterfeit goods. The company provides its platform as a service to brands that want to minimize revenue lost to counterfeiters, as well as resellers who want to ensure they aren't inadvertently breaking the law by selling fake goods.
CEO and co-founder Vidyuth Srinivasan said he settled on counterfeiting as the focus of his machine learning development, following a battery failure on an apparently new and genuine battery while motorbiking across the country.1
What Problem Is Artificial Intelligence Helping To Solve?
Aside from dodgy batteries, sales of counterfeit goods total close to half a trillion dollars per year globally.2 As well as revenue lost to the brands whose IP is being stolen, this dilutes brand identities – something fashion brands in particular are more than willing to spend money to protect.
Counterfeiting also eats into the business of genuine resellers and wholesalers and sometimes leaves them out of pocket when stock purchased in good faith turns out to be fake. A widely used style of fraud known as return fraud involves dishonestly returning counterfeit items to retailers in place of genuine ones. This is only possible because many retailers don't have the time or technological means to check every item that comes back to them.
How Is Artificial Intelligence Used In Practice?
Entrupy has developed scanning technology that uses machine learning and deep learning techniques3 to detect whether items are genuine. Clothes, accessories, jewellery, electrical goods and even automobile parts can be “fingerprinted” in minute detail. Service users can then use either a phone app or a dedicated handheld scanner to check if their purchases, or inventory, are the real deal.
This is possible thanks to microscope cameras that are able to record the tiniest details of a product's construction, such as the direction of tiny microfibres during the weaving process and the use of deep learning algorithms to assess a potentially counterfeited product. Any scanned image can be compared in real time to the reference images stored in the cloud to provide an immediate assessment as to whether it is genuine or not.
Entrupy claims that its technology is able to discern even “super fakes” – very high-quality replicas that are impossible for a human to differentiate from a genuine product.
What Technology, Tools And Data Were Used?
Entrupy has a database of millions of images4 of products sold by the brands it covers, which include Chanel, Dior, Burberry, Gucci, Louis Vuitton and Prada – some of the most counterfeited names in the world.
Specialized microscope lenses were needed for capturing the training images, as most microscopes aren't capable of capturing both the level of detail and the required amount of surface area for the data to be usable.
The images of the genuine products are used to train convolutional neural network algorithms to classify images based on texture, differences in threads and grains and marks made on products during manufacturing.5 On second-hand goods it can even distinguish differences in wear and tear to identify a fake product.
What Were The Results?
Entrupy say that its system has a 98.5%6 rate of correctly identifying counterfeit merchandise.
This means that as the technology becomes more widely used, retailers and customers will be able to buy more confidently, and the lives of counterfeiters will become more difficult.
Key Challenges, Learning Points And Takeaways
AI can parse image data in incredibly high detail far more quickly than a human eye could, and determine between counterfeit products and genuine items.
Brands are happy to help with putting this technology into the hands of customers and resellers if it helps protect their revenue and perceived value.
Counterfeiting has existed throughout human history and is unlikely ever to be completely eradicated. As with other types of fraudsters, counterfeiters are likely to step up their own technology game in response.
AI start-ups that build their business models on creating unique datasets will increasingly be seen as valuable allies to large corporations that can find value in that data.
Notes
1Tech.co, How AI Is Powering the Fight Against the $900B Counterfeit Industry: https://tech.co/ai-counterfeits-2017-08
2OECD, 2018 – Trade in Counterfeit and Pirated Goods.
3Entrupy.com: https://www.entrupy.com/technology/
4Tech Crunch, Machine learning can tell if you're wearing swap-meet Louie: https://techcrunch.com/2017/08/11/machine-learning-can-tell-if-youre-wearing-swap-meet-louie/
5KDD, The Fake vs Real Goods Problem: Microscopy and Machine Learning to the Rescue, Ashlesh Sharma, Vidyuth Srinivasan, Vishal Kanchan, Lakshminarayanan Subramanian: http://delivery.acm.org/10.1145/ 3100000/3098186/p2011-sharma.pdf
6Entrupy.com: https://www.entrupy.com/
34
Experian: Using Artificial Intelligence To Make Mortgages Simpler
Experian is one of the world's largest consumer credit reference agencies, which means that businesses, banks and financial institutions rely on its help to decide, for example, whether we are a safe bet to lend money to.
It also means it holds a huge amount of data on us and our spending habits. Now it is applying artificial intelligence (AI) to this data to make more accurate predictions, but also to make our lives easier when it comes to complex financial transactions.
One area where it is concentrating its efforts is mortgages, and it hopes that by using machine learning it will cut down the time it takes to complete the lengthy process of applying for mortgages, leading to less stress and lower fees.
What Problem Is Artificial Intelligence Helping To Solve?
Applying for a mortgage is a time-consuming and complex process. The average application involves coordinating information between a large number of agencies – buyers, sellers, surveyors, estate agents, solicitors, underwriters, mortgage brokers and lenders.1
This is the reason why buying a property is often listed as one of the most stressful life events that we deal with.
Often work is duplicated between agencies due to inconsistencies in the way information is transferred, and of course that leads to higher fees, adding to the overall expense of the process to us as consumers.
Although the process has been streamlined to some extent due to adoption of digital technology in recent decades, the reality is that it will still take weeks, even months, to get a mortgage approved,2 involving several days’ worth of activity and visits to numerous offices and agencies.
How Is Artificial Intelligence Used In Practice?
Experian is trialling an AI system that will work by analyzing thousands of mortgage applications to determine where efficiencies can be made by reducing duplicate workloads and streamlining workflows between different parties.3
The system will be trained to look at each data element, assess how frequently it is used during the process and categorize it so it can be quickly located and passed to where it is needed.
This type of work would be practically impossible for a human to carry out on anything other than a small historic sample dataset. Machines, on the other hand, can work on fast-flowing real-time data, which is updated every time a new mortgage application is completed.
It's also possible that this form of predictive technology will make it easier for those with limit
ed credit histories to obtain mortgages or personal loans. Lenders will be able to assess applications based on data of other customers that fits the applicant's profile, and come to more accurate and trustworthy decisions about their ability to keep up payments.4
What Technology, Tools And Data Were Used?
Machine learning is used to process data across the workflow. As it learns more about where data is used, and in some cases isn't used, it can build accurate models about what data is valuable, and what is surplus to requirements, at each stage of the process.
As Experian CIO Barry Libenson put it when I talked to him: “Over time we may find out that we don't need to care about five years’ worth of tax returns – what we need is five years of credit payments.”
Experian has built a platform that it calls Analytic Sandbox, which allows it to produce on-demand data-driven insights. It used the open source H20 machine learning and deep learning framework to drive its analytics algorithms.5
It also uses Cloudera's Enterprise platform to enable quick access to big data, helping it to make more accurate decisions based on consumers' credit histories.6