2. Agenda
• Introduction to Artificial Intelligence
• Introduction to Machine Learning
• Impact on FinTech
• Global Initiatives
• Case Studies
• CreditVidya
• Winjit
• Applications @ SA Taxi
3. The Mother of All Demos
9th December 1968
A computer demonstration done by Douglas Engelbart at Fall Joint Computer
Conference, SF
While all scientists concentrated in making computers smart, Engelbart was
interested in how computers could make humans smarter
In his terms, it was “Augmented Intelligence”
Ref: https://en.wikipedia.org/wiki/The_Mother_of_All_Demos
4. Automation
By definition:
Automation is the use of machines and technology to make processes run on their
own without manpower
Every action in today’s human life is a consequence of automation over the years
In history, each age progressed into the next age on the basis of automation
Data Age and “DATA IS THE NEW OIL”
5. Automation and AI
Ref: https://www.jibe.com/wp-content/uploads/2015/06/automation-and-recruiting-chart.png
6. Artificial Intelligence
Development of computer systems able to do tasks which normally require human
intelligence
Term AI is applied when machine mimics “Cognitive” functions
Field of AI Research was born @ Dartmouth College in 1956
AI Caliber
Artificial Narrow Intelligence (ANI)
Artificial General Intelligence (AGI)
Artificial Super Intelligence (ASI)
7. Common Examples of AI
Cars – Anti-lock brakes, tuning parameters of fuel injection system
Mobile Phones – navigating maps, talking to Siri, Music recommendations
Emails – Spam filters
Amazon, Netflix – Recommended for you
Google Search and Voice
Apple Siri
Googles Self Driven Cars
Mr. Delivery
9. Machine Learning
Computer systems learning from data and experience
“A field of study that gives computers the ability to learn without being explicitly
programmed” – Arthur Samuel
If a computer program can improve how it performs a certain task based on its past
experience, then we can say the program is learning
Close resemblance with Data Mining
17. Impact on FinTech
Customer Segmentation
Scoring
Applicant – Credit Scoring
Application
Behavioural – Predict likely customer default
Collection – Predict likely debts recover
Robo Advisors
Fraud Detections
Identity Theft Prevention
AI Assistants and Bots
Blockchain and Bitcoins
Debt Collections
Market Research and Sentiment Analysis
Augmented Decision Making
18. Global Initiatives
Royal Bank of Scotland brings Robo Advisors (Link)
Ernest and Young provided digital wallets to all employees, accepts BitCoin as
payment for its consulting service, and installed a BitCoin ATM (Link)
KPMG using McLaren Applied Technologies (its AI) to speed up audits (Link)
Xero (SA) bring AI to its Online Accounting System (Link)
Clevva’s (Western Cape) product helps in companies call centre operations (Link)
MyBucks provides financial services to sub-Saharan Africa using AI (Link)
Apple Machintosh : Released on January 24, 1984
He showcases
Windows
HyerText
Graphics
Efficient Navigation and Command Input
Video Conferencing
Computer Mouse
Word Processing
Dynamic File Linking
Revision Control
Collaborative Work
We are looking at the Data Age from the Current Information Technology Age where Data is the new Oil
Reliance Industries put internet in the hands of 1 billion Indians with the only motive to capture data. With USD 0 in revenue, it is valued at USD 4.5 Billion
@SA Taxi we automated processes and reaping the results of these.
ANI – Very specific stream of intelligence. E.g. Siri – Voice to Text Conversion
AGI – Human like intelligence. Could reason and learn. E.g. Jarvis
ASI – Super Human Intelligence. Can control what should be done and can make things happen. E.g. Ultron / Vision from Avengers
To understand the concept of Machine Learning better, let us consider some of its applications:
Hand written character recognition
Political campaigns: Machine learning is used to determine which voters a campaign should target and what optimal actions can bring about their desired goals.
Predictive Policing - Police departments in the US have started to direct their focus to crime prevention. The idea is to use ML techniques to identify areas where crimes are likely to occur and allocate resources more effectively and enable proactive and preventative policing strategies.
Surveillance: Systems that put people on the no-fly list use Machine learning to classify people into suspicious or not.
Facial recognition: Companies like Facebook use learning algorithms to recognize a person by photo after they have been tagged a few times.
Self-driving vehicles: An autonomous vehicle is effectively equipped with dozens of pairs of eyes (sensors), all connected to a brain (a learner) – which is wholly focused on safe and efficient driving.
Recommendation systems – Co.’s like amazon learn and predict user’s potential preferences
Ads – ML is being applied to different aspects of the advertising industry- E.g. Intent prediction of an individual, response prediction to an ad
Personal assistant – Learning algorithms are used to learn from actual language used by customers, gathered from call logs, chat histories, search terms used, twitter feeds etc.
Now how do we do Machine Learning?
This diagram depicts high level ML process.
First we have raw data, this data comes from different sources and formats
In the pre-processing stage - we clean, format and sample the data. As Figure shows, it’s an iterative process, with several different data pre-processing modules applied to the raw data. In fact, choosing the best raw data to start with, then creating prepared data from that raw data frequently takes up the majority of the total time spent on a machine learning project.
Once our data is prepared, we need to apply one or more algorithms to our data. The goal is to determine what combination of machine learning algorithm and prepared data generates the most useful results.
The resulting product of our algorithm is called a ‘model’. These models are then used to provide a solution to our problem. For e.g. It could be used to answer questions like “Is this transaction fraudulent?”
Machine learning tasks are typically classified into three broad categories,
Supervised learning – We present the algorithm with example inputs and their desired outputs, and the goal of this is learn the mapping between inputs and outputs.
Unsupervised learning – The learning algorithm is not given any example outputs and the goal is to find structure or patterns in our input data.
Reinforcement learning – RL is concerned with agents choosing actions that maximize the expected reward over a given time. This is best achieved when the agent has a good policy in hand. Learning the best policy - remains to be the goal in reinforcement learning
Lets take a closer look at these techniques.
The supervised learning technique can be further divided into two subcategories - Classification and Regression.
In classification problem, inputs are divided into two or more classes, and the learner must produce a model that assigns unseen inputs to one or more (multi-label classification) of these classes. E.g. Identity fraud detection – classes ‘fraud’ or ‘not fraud’
Regression – the outputs are continuous rather than discrete. E.g. Forecasting population growth
Unsupervised learning techniques divided into two subcategories as Clustering and Dimensionality Reduction.
Clustering problem basically cluster samples according to similarities of these samples regardless of class information. E.g. Recommendation Systems are used to recommend something for the users. It can be a movie, music or something which is sold in the market place.
Dimension Reduction simplifies inputs by mapping them into a lower-dimensional space. E.g. Reducing the high dimensional or big data for visualization purposes
Self driving cars use Reinforcement learning to make decisions continuously – which route to take? what speed to drive on? are some of the questions which are decided after interacting with the environment
1.
2. Robo Advisors are programs guiding casual trader to mange there portfolios
3. Application are identifying behavioural and transactional patterns to identify and prevent frauds
4. Bots working with customer service agents
5. Blockchain : Distributed blocks of transaction encrypted and verified. Its basically a distributed ledge r
6. Debt Collections: Identify patterns in payment and impact of external factors (e.g. weather, news, policies,) on customers payment.
5. Xero’s Cloud accounting software is connected to bank feeds, and receipt scanning tools, can access a real-time, live ‘ledger’ which they can use to inform advice and recommendations. Focus on High Integrity Accounting. Coding would be automated and data entry would be redundant.
6. Clevva’s software claims to “clone” the organisation’s key experts, giving everyone access to their advice through software.
7. MyBucks – Credit assessment, credit decisioning and scoring technologies combined with alternate data points for easy fast and automated disbursement of loans