Right now, as you’re reading this blog post, Waymo (Google’s offshoot) is testing its self-driving cars and it’s nearly ready for mass production!
Quite soon, you can tap a button on your smartphone to summon your car that can drive you around the neighborhood or get you through the morning traffic while you sip your coffee and relax listening to your favorite music!
Are you a Star Wars fan? If so, do you want your own R2D2 to take care of your household chores?
Well, researchers at MIT are training robots by creating simulations of common household chores. In a few years from now, you can say goodbye to those boring mundane tasks!
By 2050, robots may take over all mundane, monotonous, and assembly-line jobs and provide goods very cheaply or nearly free! In such a case, you’ll be free to work on ‘creative’ jobs that are actually fun while your government provides you income!
All these scenarios will be possible only due to two vowels that hold the key to our future: AI!
Artificial Intelligence (AI) is at the heart of all such future changes that’s going to propel humanity to new heights for the better! Imagine C3PO and R2D2 in Star Wars and Ava in Ex Machina and JARVIS (Just a Rather Very Intelligent System) in Iron Man… All these examples are going to be part of our everyday lives in the next few decades!
And, in case, you were pondering about Terminator-like situations, AI researchers are working hard to prevent such events, so don’t worry about that.
I am going to write a series of blog posts on AI but before you take a deep dive, have a look at a brief overview of the important technical ideas and terminologies associated with AI in this post.
Artificial intelligence is the ability of the computer to think and perform actions without human intervention. An AI system will be able to make decisions and act on its own without any explicit instruction.
You probably want to apply AI to solve your problems. But AI is a huge field with multiple approaches and applications. Therefore, before you begin to apply AI to any problem, you first need to understand the different approaches or sub-fields of AI.
Types of AI
AI is classified into two types based on its scope:
- Strong or general AI
- Weak or narrow AI
What we now have in common use across most industries is called ‘Weak AI’ or ‘Artificial Narrow Intelligence’. It’s this AI that automatically performs specific tasks. It’s restricted to a particular problem’s domain.
This is the general-purpose AI that can perform any task. The AI you see in Star Wars and Westworld that has consciousness and can ‘think’ on its own, is called ‘Strong AI’ or ‘Artificial General Intelligence’ and that’s what scientists and engineers are working on.
OpenAI, a non-profit AI research startup founded by Elon Musk and Sam Altman, is actively working on such a general-purpose, friendly AI to help humanity.
Origins of AI
AI has been part of our imagination for a long time and research on it has been going on since the 1950s.
The first work in AI (even before the term was coined) was done by Alan Turing, who proposed that “If a human couldn’t distinguish between responses from a machine and a human, the machine could be considered ‘intelligent’.” The systems that satisfy this rule are called Turing-Complete.
It all began in 1956 at the Dartmouth Conference where scientists from MIT and CMU coined the term and gave birth to AI. They became the founders of AI. They developed several algorithms to play checkers, solve algebra, and speak English and were funded by the Department of Defense, USA.
But they faced several difficulties with the hardware and AI lost focus in the following decades beyond a few sci-fi books and movies. Then, as computation power began to increase rapidly in the 90s, AI began to rise again and was used in logistics and data mining.
After 2012, the rise of fast and cheap Graphic Processing Units (GPUs) and new neural network algorithms and access to large data caused the current AI boom.
Sub-fields of AI
The major sub-fields are:
- Machine Learning (ML)
- Computer Vision (CV)
- Reinforcement Learning (RL)
- Natural Language Processing (NLP)
Note: These sub-fields often blend with one another, may overlap, and share algorithms and techniques.
Machine Learning is the name of the game!
It’s a sub-field of AI where the machine learns by processing all the data provided to form a statistical model or pattern for the data, and based on the model, performs actions or makes predictions.
The initial data is called training data and the model is the trained model. Then, that model can be applied on new data for predictions and classifications.
For instance, you’re already using products powered by ML in your daily lives:
- Product recommendations on eCommerce websites, such as Amazon, and on music and video streaming services, such as Spotify and Netflix
- Google Search
- Virtual assistants, such as Siri and Alexa
- Targeted advertising done by Amazon, Facebook, and Google
ML systems are also accelerating research on:
- Self-driving and driver-less cars (yes, there’s a difference between the two!)
- Medical diagnoses and personalized medicines, including cancer detections and cures
- Identifying financial crises and recessions
Computer Vision makes machines ‘see’!
This sub-field deals with computers learning to read images and videos and generating images on their own. In other words, CV algorithms allow machines to recognize and understand images and videos.
These algorithms enable machines to:
- Identify texts in images and understand them (for instance, postal codes and addresses on packages)
- Identify and automate inspections on production lines
- Self-drive cars
- Identify cancers based on images of tumors
Reinforcement Learning allows machines to make decisions and move about!
RL is a sub-field of machine learning, which is concerned with decision-making and motor control. Watch an AI agent learning to walk by itself!
These algorithms work without any training data. Given an environment, these algorithms explore it and perform some actions, and based on the result, train themselves. These algorithms work by using the trial-and-error method.
The RL agent learns a policy that gives it a reward based on the action it performs and tries to improve the policy to maximize the reward.
RL is a very general-purpose algorithm that can be applied to any situation. It encompasses all problems that involve making a sequence of decisions. RL algorithms have started to achieve good results in many difficult environments.
Deep Blue, the machine that beat Gary Kasparov in the historic chess match in 1997 was based on an RL algorithm. Also, the latest robotic vacuum cleaners available in the market is an RL agent.
The OpenAI initiative, set up by Elon Musk, and DeepMind, started by Google, have created several environments for RL algorithms. They have made several game-changing breakthroughs and designed systems that can play traditional games, such as GO and chess and computer games, such as Dota.
Natural Language Processing enables machines to learn languages!
The NLP sub-field deals with learning and understanding human languages. These algorithms work with text and speech data and try to understand the speech and its context. This algorithm learns the semantics and grammar of the language and produces a set of rules based on which it creates its own content. It improves its rules by continuously learning based on interaction.
For instance, Google Translate, Google Assistant, Siri, and Alexa all work by using NLPs. Microsoft has even created a real-time language translator for skype by using NLP.
Artificial Neural Networks
All these sub-fields work together to build AI systems. One of the most important algorithm or technique used by all of these sub-fields is called Artificial Neural Networks (ANNs). ANNs are systems inspired by human brains. It is a collection of nodes called neurons.
Each neuron receives a signal, processes it, and signals other neurons. Thus, all the neurons collectively make a decision based on the input signal.
They are the most commonly used algorithms right now and there is hope that they’ll lead to Artificial General Intelligence. Deep learning is an implementation of ANN that uses several layers of neural networks.
Ahem, <coughs> ahem, I know this is getting to be a long post and I maybe frying your neural network… So, I’ll end this post with a short introduction on AI-based applications in FinTech! 😊
AI applications in FinTech
A mortgage model can be built using an ML algorithm to find the price of a house based on several parameters, such as sq. ft. size, area, facilities, etc.
Machine learning can be used to automatically find the best mortgage loan for a person and how likely is it that he/she will repay the loan or default on it.
ML can be used to build credit-risk models by capturing the data of borrowers and automatically audit the loans based on credit history and other parameters.
General financial outlook
ML systems can be used to monitor economic situations based on several parameters, such as interest rate, inflation, GDP, and build risk models and stress tests that will flag any anomaly or identify potential cause for recessions or financial downtrends.
For instance, BlackRock is using AI to predict the future of the economy. ML systems can ensure that events like the 2008 subprime mortgage collapse can be prevented.
ML systems are already being used for fraud detection. Every time you make a credit card transaction, AI systems are monitoring the transaction and deciding whether it is fraudulent or not. They monitor all user transactions to identify credit frauds and alert the banks for potential frauds.
Zensed, an AI-based startup, has developed a product to monitor and prevent frauds.
RL-based systems can be used to build high-frequency trading algorithms that’ll update themselves based on current market trends to predict stock prices and trends. RL algorithms can help in both technical analysis and fundamental analysis of a company as well as to learn trading strategies for various kinds of trading.
So, that’s all my friends… Hope you had a satisfying read. Bricks and bouquets are welcome! Leave a comment and I shall surely respond. Meanwhile, to whet your appetite on AI until I return with my next post, check out the following case-studies and research papers:
Stanford paper on the future of AI: https://ai100.stanford.edu/sites/default/files/ai_100_report_0831fnl.pdf
Hitachi case-study on AI in FinTech: http://www.hitachi.com/rev/archive/2016/r2016_06/pdf/r2016_06_104.pdf
A book published by Microsoft covering short stories on AI and ML among others: https://news.microsoft.com/futurevisions/
Please note that the views, thoughts, and opinions expressed in this blog post belong solely to the author, and not necessarily to the author’s employer, organization, committee or other group or individual.