The concept of ordering a taxi has been dramatically transformed in recent years. Gone are the days when it was necessary to pre-book a taxi home from your Christmas party, or when you needed to call your local taxi service hours in advance of your ride. That’s not even to mention the inconvenience of needing cash to get yourself home. Thanks to companies such as Uber, getting around is as quick and easy as reaching for your smart phone and pressing a couple of buttons.
With the introduction of connected cars and the progression towards fully autonomous vehicles, the industry is set to be further disrupted. Not only are the increase of these vehicles saving time and resources, the eventual roll-out of driverless cars will help in reducing our carbon footprint, make the roads a safer place, and allow less able people the freedom of personal transport.
Uber has many strands to its development and in order for the company to stay ahead as market leaders in such a complex sector requires scrutiny in every area of the business. At the Deep Learning Summit in San Francisco this January 25 & 26, Vicente Silveira, Head of Fraud Data Science will share his work in maintaining the security of Uber through his use of deep learning to fight fraudsters. Generally, fraud models are based on narrow data streams processed by traditional machine learning models. Uber have taken this one step further and are improving this by applying deep learning to extract complex feature relationships from high-dimensional datasets such as tapstream and location data.
Upon speaking with Vicente, he explained that his team is focusing ‘on preventing fraudsters from exploiting the Uber platform and Uber’s users.’ The team uses machine learning to predict whether an Uber transaction will be fraudulent, for example, machine learning models and features help us detect credit card fraud, gps spoofing, promotion abuse, and fraudulent collusion.
With a background in software engineering, Vicente began working with machine learning products and services eleven years ago, and entered into the AI space through fraud and security work. Both of these areas have collected vast amounts of data which make the application of AI and more specifically machine learning ideal to improve the quality and speed of decisions, leading to a better user experience.
‘AI and machine learning have had exponential adoption in the last few years, but it is still a developing space.’
Vicente explains that the industry is going through a major transition between the world of software and the world of AI: ‘this change means that we now care a lot more about the quality and true meaning of data because the data will effectively help the machine write the software.’ His team at Uber are working to create better mechanisms and processes to manage data and need better ways to understand how data is affecting model decisions. As the field grows, it’s important to encourage more people into the field, and we need to make sure that large corporations are investing time in training the next generation of experts rather than taking talented university professors to become researchers of their own.
'Right now, there is a tremendous shortage of talent in this space. Someone with curiosity, common sense and an understanding of basic math can make a huge impact in this field. It is a great opportunity for everyone, not just PhDs.'
Uber's mission is to bring transportation for everyone, everywhere. ‘Transportation is a bedrock of the world's economy, so if we succeed in our mission we will be able to positively impact human development levels across the planet.’ The potential for collecting data through driving is huge, and a key ingredient to solving the above problem is leveraging such data and AI automated decisions to ‘drive efficiency, improve safety, and build a reliable marketplace’. This means that almost every department of Uber is employing AI,including anti-fraud, safety, production service monitoring, demand forecasting, ETA calculation, map making, driver dispatching, etc.
As Uber has such diverse uses for AI, we asked Vicente a couple of questions about AI progressions more generally:
Wow. It is difficult to pick just a few because there is so much happening in the field, you basically need to update your list on a weekly basis. I think deep learning will expand in a major way from the currently hot areas such as computer vision to a range of other domains and problems. Recent progress on deep learning understanding such as feature visualizations are helping us start to understand these algorithms. I also think that reinforcement learning advances, like the ones we see today with game playing AIs, will be broadly applicable especially as we increase the deployment of AI on robots and other tangible objects. In terms of industries, as it was with other major advances such as electrification and software, it is hard for me to think of one area that will not be transformed by AI. Naturally there is a curve of adoption for AI, with many practical applications today and even more in the coming years. Industry-wise, I'm very excited to work in the transportation sector where we are taking the lead in helping society benefit from this technology.
I am an optimist and I think that both the historical record and current AI trends point to a future with a lot more global wealth and a higher quality of life for most people, where AI-augmented humans will be able to accomplish amazing things. That said, it is also clear that AI will have major impact in the types of jobs that will be in demand. For example, the automation of farming has eliminated a huge part of the farming jobs that existed in the beginning of the last century. Also phone switchboard operators basically disappeared as we automated telephony switching. On the other hand, the introduction of new technologies has also created new jobs: computer specialist roles grew 95% in a period of just 40 years. Also, through the AI-augmentation built into smartphones and applications like Uber, today we can unlock the value of a skill like driving at a massive scale and allow 2 million people to have a flexible way to earn income. In the end, and just like previous technology revolutions, AI will introduce new types of occupations, some of them because of new types of businesses that AI will enable, some of them because of direct need of tapping into human cognition to label and curate data that will be used for AI training.
To learn more from Vicente about his work at Uber, join us at the Deep Learning Summit in San Francisco this January 25 & 26. Additional confirmed speakers include Ian Goodfellow, Research Scientist, Google Brain; Andrej Karpathy, Director of AI, Tesla; Daphne Koller, CCO, Calico; Clement Farabet, VP of AI Infrastructure, NVIDIA; Yves Raimond, Director of Machine Learning, Netflix, and many more who you can view here.