Artificial Intelligence in Practice

by Bernard Marr

  They also have to ensure a speedy response when passengers ask for a ride, so as not to lose custom to other companies or public transport providers.

  Additionally, drivers (particularly late at night) sometimes face problems themselves because they have to deal with passengers who may be drunk and abusive.

  These – along with a whole host of others – are some of the problems that Uber is tackling using AI.

  How Is Artificial Intelligence Used In Practice?

  Uber uses AI for its core business of despatching drivers to passengers who are waiting to be picked up, and calculating the most efficient routes to get them to their destinations.

  It also powers the company's “surge pricing” model, which increases fares when the service is in high demand to encourage more of its drivers to clock in, therefore reducing customer waiting times.

  Hotels, airlines and public transport operators have used this technique of balancing customer demand for a long time – which is why flights and hotel rooms are usually more expensive to book at peak times or during holiday periods.

  Uber's innovation is to use advanced predictive technology to adjust pricing in real time, meaning it can more efficiently respond to changes in supply and demand.

  With passengers increasingly using Uber in both their professional and personal lives, people may find themselves juggling two different accounts with the service, depending on the purpose of their journey.

  Uber is using AI to solve this, too. By analyzing the pick-up point and destination as well as the time of day, it attempts to predict whether a journey is for business or pleasure, and if a user has two accounts on their phone, it will automatically suggest which one should be used.2

  Uber also uses AI in its marketing efforts, using machine learning algorithms to segment customers. It categorizes them according to how likely they are to respond to certain types of promotional advertising, and even understands how the frequency with which they open messages relates to how likely they are to unsubscribe.3

  A recent patent application reveals that Uber has even developed technology designed to predict whether a customer may be drunk. Although it hasn't publicly stated how it will use this technology, there is speculation that it could be aimed at protecting its drivers from passengers who could possibly be abusive or dangerously intoxicated.

  Critics have said that this could amount to discrimination, and put customers in a dangerous situation if drivers decide to decline them service. It could even potentially lead to drivers with nefarious intentions targeting those that the system predicts are likely to be in a vulnerable state. However, Uber has not yet said that it has any plans to roll out this technology.4 Perhaps this is a good example of a situation where just because something is possible with AI, it doesn't mean it's necessarily a good idea.

  Uber uses machine learning within its Uber Eats food delivery platform, too. Here, the idea is to predict as accurately as possible how long it will take a customer's food to arrive. It has to take into account how long a meal will take to prepare, when a driver will be available to go to the restaurant and collect the food, and how long it will take the driver to reach the customer's home.5

  What Technology, Tools And Data Were Used?

  Uber uses GPS data from passengers’ and drivers’ phones as well as map data to plan routes between pickups and assign drivers to those in need of a ride.

  Data gathered from every one of the millions of journeys made is fed back into its learning algorithms with the aim of giving customers more accurate ETAs for their rides, and shortening waiting times.

  If a passenger is waiting on a particular street corner, and it is able to calculate that due to traffic restrictions or speed limits there would be a significantly shorter waiting time at a different, nearby location, then it will suggest that the customer move there instead to wait for their vehicle to arrive.6

  To complete all of these tasks, Uber has built its own machine learning platform called Michelangelo. As well as maintaining the company's data lake where it logs all of its transactional and customer behavior data, it takes care of training and evaluating algorithms, deploying the most efficient models, making predictions and monitoring those predictions to determine their effectiveness.7

  Uber's patented method of detecting whether passengers are likely to be drunk – Predicting User State Using Machine Learning – uses data, including customer walking patterns, errors made while typing and how they are interacting with the app, and draws inferences by comparing this data with how they usually use it.

  Uber's current AI research division, Uber AI Labs, was formed with the proceeds of its acquisition of Geometric Intelligence in 2016. AI Labs conducts research on applications of deep learning and neural networks that go beyond Uber's own business cases.

  Recent research topics at Uber AI Labs include subjects as diverse as machine learning techniques for identifying wild animals photographed in the Serengeti for conservation purposes, to developing and open sourcing Pyro, its own AI programming language.8

  What Were The Results?

  Shorter waiting time for rides and more efficiently routed journeys mean improved customer satisfaction, and a higher likelihood that a user will become a regular customer, with a high lifetime value to the business.

  The success it has had with machine learning and predictive models means that Uber has been able to efficiently scale from a San Francisco start-up to a worldwide ride hailing network.

  Uber claims an 80% success rate at using AI to predict whether a passenger should pay for their ride using their business or personal account.

  Key Challenges, Learning Points And Takeaways

  Machine learning algorithms mean that customer wait times and journey times, as well as routes, can be predicted with a high degree of accuracy.

  Uber looks at machine learning as a tool that can be applied to any area of its own business, to generate efficiencies and improve customer experience.

  Uber has emerged as an “AI first” company with the aim of competing with the key players in the global AI revolution, such as Google, Facebook and Amazon.

  Uber's rapid success has disrupted the traditional taxi hire business around the world.

  Notes

  1Forbes, Uber Might Be The First AI-First Company, Which Is Why They “Don't Even Think About It Anymore”: https://www.forbes.com/sites/ johnkoetsier/2018/08/22/uber-might-be-the-first-ai-first-company- which-is-why-they-dont-even-think-about-it-anymore/#49b54a165b62

  2BGR, Uber to use Artificial Intelligence to help differentiate between personal and business rides: https://www.bgr.in/news/uber-to-use-artificial-intelligence-to-help-differentiate-between-personal-and-business-rides/

  3Techwire Asia, How does Uber use AI and ML for marketing?: https://techwireasia.com/2018/06/how-does-uber-use-ai-and-ml-for-marketing/

  4Independent, Uber Patent uses Artificial Intelligence to Tell if You're Drunk: https://www.independent.co.uk/life-style/gadgets-and-tech/news/uber-patent-drunk-passenger-ai-artificial-intelligence-app-a839 5086.html

  5Uber Engineering, Meet Michelangelo: Uber's Machine Learning Platform: https://eng.uber.com/michelangelo/

  6Tech Republic, How data and machine learning are “part of Uber's DNA”: https://www.techrepublic.com/article/how-data-and-machine-learning- are-part-of-ubers-dna/

  7Uber Engineering, Meet Michelangelo: Uber's Machine Learning Platform: https://eng.uber.com/michelangelo/

  8Uber AI Labs: http://uber.ai/

  Part 5

  Manufacturing, Automotive, Aerospace and Industry 4.0 Companies

  41

  BMW: Using Artificial Intelligence To Build And Drive The Cars Of Tomorrow

  German automobile manufacturer BMW builds and sells 2.5 million cars every year all around the world, badged with its BMW, Mini and Rolls Royce brands.

  With a reputation for excellence and early adoption of new technology, its vehicles are some of the most sophisticated on the roads, and like
its rival Daimler, it is a frontrunner in the race to make self-driving cars an everyday reality.

  BMW operates more than 30 assembly facilities spread across 15 countries, involving a vast logistics operation in which efficiency is key to remaining profitable and competitive with old rivals such as Daimler and new challengers like Tesla.

  What Problems Is Artificial Intelligence Helping To Solve?

  Car manufacturing is a hugely costly and labour-intensive industry where millions are spent each year on research and development, production and marketing. This can drive revenues of billions, but mistakes at any stage of the fast-moving, complex process can be extremely costly – particularly if they are not discovered before vehicles hit the roads.

  On top of this, more than 100,000 people around the world die in road accidents every year, with the majority caused by driver error. Self-driving cars are being developed as the answer to this needless waste of human life, but first they must be trained to understand how to navigate and interact with other cars, either automated or driven by humans.

  How Is Artificial Intelligence Used In Practice?

  Like most businesses that are investing in artificial intelligence (AI), there are two main threads to BMW's activities. One is integrating autonomy throughout their own business processes to streamline them, drive efficiency and discover new opportunities. The other is to integrate AI into their products and services to create more compelling offerings for their customers.

  Back in 2016, a partnership with IBM saw four BMW i8 vehicles connected to the IBM Watson cognitive computing platform through their Bluemix cloud service. The idea was that the car could learn how to improve its understanding of driver behavior and then adapt its system to suit personal preferences. By uploading all the data it gathers to the cloud, the system is able to build a vast database of user behaviors, and then use machine learning to anticipate the needs and preferences of other drivers.

  Following the trial, the system was rolled out to users of BMW's ConnectedDrive app in Germany in 2017. Examples of how it can be used include quicker and more accurate diagnosis of vehicle faults and access to cheaper insurance premiums if drivers are willing to share their data with insurance companies.

  Another partnership has seen BMW teaming up with Intel, which recently acquired computer vision specialists Mobileye. Computer vision is essential to cars being able to operate autonomously – essentially it is the process that allows cars to “see” by analyzing image data from onboard cameras and therefore react to the world around it.

  This technology uses machine learning to categorize images as they are captured – allowing the car to decide in milliseconds how it should react to objects such as other vehicles or even pedestrians stepping into the road in its path. By analyzing a sequence of images from a video feed, it can not only determine what an object is, but how far away it is, what direction it is travelling in and at what speed. These are all processes that our human brains have learned through evolution to carry out subconsciously. Within limited parameters, humans are very efficient at doing so – however, natural evolution doesn't happen quickly enough to keep pace with the speed of technological progress in a world that has gone from horses and carts to 100 kph motor cars in a little over 100 years. Hence, the large number of casualties caused by underestimating speeds, overestimating distances or simply lack of attention. Computer-driven cars will not make those same errors.

  One challenge is that developing autonomous driving systems requires huge amounts of data to train vehicles to tackle every permutation of situations they may come across on the roads. According to Sam Huang, of BMW iVentures, an autonomous system may have to drive around 6 billion miles before it is fully trained.

  BMW's solution here is that these miles don't all have to be driven in the “real world”. To this end, it is investing €100 million in what it says will be the world's most advanced driving simulation center in Munich, Germany. BMW describes this as “bringing the road into the lab” and it will allow them to gather data for training autonomous vehicles far more quickly, cheaply and safely.

  What Were The Results?

  Although a few years away from becoming part of our everyday lives, BMW has unveiled several concept models that represent the fruits of its research into AI-assisted cars and autonomous vehicles.

  Likely to be mostly of interest to the super-rich is the Rolls Royce concept model 103EX, controlled by an AI known as Eleanor – named after the actress who inspired the marque's famous Spirit of Ecstasy head ornament. In fitting with the brand's image, Eleanor is styled as an AI chauffeur rather than simply an AI assistant as used in other connected cars. The car is even capable of generating a “virtual red carpet” using LED projectors to ensure the occupants always alight in style.

  Perhaps of more relevance to most of us, the Mini Vision Next 100 is designed for a future where car sharing is forecast to play an increasing part in our lives. Here, technology is purposed to recognize different drivers as they enter different vehicles in a fleet and adapt to our preferences. It will also autonomously drive to its servicing hub between users, to be cleaned and prepped for the next user.

  Autonomous Mini concepts have also been set out, and here the aim is to produce a consumer product that interacts seamlessly with our lives, just as smartphone and other connected device manufacturers hope to do.

  For other applications that are likely to enjoy a wider exposure among its customers, BMW has partnered with a number of leaders in the AI field, including IBM. Its Watson cognitive computing platform was used in prototype i8 hybrid vehicles to learn about how drivers and their cars’ systems can interact more comfortably and naturally.

  Areas where IBM has said that Watson will be put to work are self-diagnosis of faults and issues that are limiting car performance, management of communications with other autonomous cars and detecting and adapting to drivers’ preferences.

  What Technology, Tools And Data Were Used?

  BMW is using computer vision technology created by Intel, through its work with MobilEye, to train autonomous cars to navigate urban and rural roads. It also works with IBM's Watson cognitive computing platform and BlueMix cloud platform to gather and analyze driver data, which includes using Watson's natural language processing abilities to interpret and react to voice commands.

  GPS data is provided through the Here location data service, co-owned (along with Volkswagen and Daimler) since its acquisition from Nokia, allowing BMW to understand where and how its vehicles are driven.

  Data is collected from onboard cameras, as well as machine data such as braking force applied and use of peripheral systems such as wipers, headlights and airbags.

  In its manufacturing and production operations, data is gathered across design, production, logistics, distribution and servicing departments, and here BMW works with Teradata to automate operational decision making. Their systems allow the journey of any part to be tracked from when it is manufactured to when it is fitted to a vehicle and when the vehicle is sold, helping create efficiencies in its logistics and ensure every part is in the right place at the time it is needed. Its production lines operate using predictive maintenance, which means worn machinery parts are replaced before they break down, driving further efficiencies.

  Key Challenges, Learning Points And Takeaways

  Self-driving cars are seen by every major automobile manufacturer as the future of personal transportation.

  The autonomous cars of the future will be safer and more efficient thanks to their use of AI to anticipate and react to unexpected circumstances on the road.

  Traditional automobile manufacturers are partnering with tech companies to bring in the expertise needed when it comes to integrating cutting-edge cognitive software with large-scale vehicle production.

  Before autonomous cars are commonplace, we are likely to see far greater integration of AI in manually driven cars in the form of virtual assistant AIs, changing the way we operate and interact w
ith vehicles.

  Sources

  BMW, Driving Simulation Centre: https://www.press.bmwgroup. com/global/article/detail/T0284380EN/bmw-group-builds-new-dri ving-simulation-centre-in-munich

  Cognilytica, 6 billion miles, Sam Huang: https://www.cognilytica. com/2017/11/15/ai-today-podcast-011-bmw-investing-ai-interview- sam-huang-bmw-iventures/

  42

  GE: Using Artificial Intelligence To Build The Internet Of Energy

  GE was founded by Thomas Edison and today operates globally across the power, manufacturing, healthcare, aviation, oil and gas and financial sectors. Thirty per cent of the world's electricity is supplied by GE Power's turbines and generators.1

  In many ways, the energy industry has continued to follow the basic pattern set by Edison in the 18th century. The core process involves the generation of electrons that are transmitted one way between their source and destination.

  Today, two pressures – the data-driven, digital revolution and the need for a switch to more sustainable sources of energy – have combined to create unique challenges, but also great opportunities.

  To meet these challenges and capitalize on these opportunities, GE has spent five years and a billion dollars2 transforming itself from an industrial company to a software and analytics company, with a focus on building smart, self-learning machines.

  What Problem Is Artificial Intelligence Helping To Solve?

  The growth of the world population and industry means there is greater demand than ever before for electricity, and that demand is only going to increase as more of the developing world becomes industrialized.