Artificial Intelligence is invisible and everywhere
A startup’s journey towards AI
Unless you’re a technophobe, most of us are consumers of AI products today— whether we like it or not. Those who have embraced AI technologies are likely to wake up talking to their Smart Assistant (Amazon Alexa, Google Assistant or Apple Siri) listening to the latest weather or daily news, have Google app read an article while preparing breakfast, be interrupted by your Smart Home console informing you that your neighbor is at your front door, or crawling into a Tesla with AutoPilot to drive back home after having one too many drinks (note: absolutely not recommended). The rest of the population are likely to be unaware of the “invisible” AI technologies that prevents email spam, the digital butler that recommends products as we shop online, facial recognition deployed in various states and cities, or the various algorithms that determine our insurance or credit risk. Whichever category you fall in, AI and its underlying technologies, are already here and everywhere.
The AI renaissance
Growing up as a geek in Silicon Valley and consuming lots of sci-fi movies, I became enamored with the vision of machines that can learn and demonstrate intelligence or Artificial Intelligence (AI). Fast forward several years beyond the decades of AI Winters, the equivalent of AI’s Great Depression, I felt like I woke up in an AI renaissance. Among the tell-tale signs was the announcement of the Nobel Awards of Computer Science, the Turing Award “won by 3 pioneers in AI.” Over the past decade, several branches of AI evolved such as Machine Learning (ML), Natural Language Understanding, Computer Vision, and Deep Learning. In fact, you’re likely to find all of these AI and STEM disciplines applied to Tesla’s electric vehicle platform which can be gleaned from a summary of AutoPilotAI which serves as a recruitment page with one question to applicants: “What exceptional work have you done in software, hardware or AI?”
Other popular applications of AI include Smart Assistants from Amazon, Google, Microsoft, and Apple — whose products have brought AI consciousness to our mobile phones and homes. There are other less known use cases such as DeepMind beating professional gamers of Go and Chess and accurately modeling the structure of proteins which could lead to significantly reduced time to produce life-saving drugs.
You may be asking, “that’s all well and good, but how can our company take advantage of AI?” The good news is that many of these AI technologies are within reach of much smaller companies thanks to dozens of open-source AI projects and pricing models based on consumption. As these AI Research labs drive the pace of innovation, Google’s DeepMind and Tesla / Microsoft’s OpenAI continue to publish 100s of papers that benefit the entire ecosystem.
This further motivated me to embark on a personal journey to learn more about AI technologies with a mission to drive competitive advantage for our small startup with a handful of engineers in our analytics team. I hope this blog and the dozens of links to other articles spark interest with other business leaders to take a similar journey, so that we may share experiences — and more importantly, plug into the fast evolving AI ecosystem (note: 2 of the AI Turing Awards recipients spent much of their time in Canada hence the Canadian AI ecosystem is pictured below).
Taking the first steps towards AI
As we embarked on our journey several years ago, we started with a couple of pilot projects building our own AI-enabled apps using Machine Learning such as Natural Language Understanding to enhance chatbot conversations and streamline user registration by feeding camera images to Vision AI. On a personal note, I thought my computer science background would enable me to lead my team beyond bolt-on applications. However, I soon realized that I had vastly underestimated the scope of the opportunities. Anyone who has taken a similar journey would have discovered that AI combines the STEM disciplines (science, technology, engineering, and mathematics) and many more including: statistics and data science. In short, I discovered we had just scratched the surface of the iceberg and the AI opportunities that lay underneath was so immense that it was a struggle just figuring out where to start. The good news is that we made the right decision several years ago to focus on a key ingredient to AI — data. Since, there is no AI without data, this gave us further momentum towards a data first strategy and culture.
Starting our AI journey
If you started your journey by searching for an Introduction to AI course, you’d probably wished you’d taken some time learning the pre-requisite courses. In short, the AI journey is as short or long as you’re willing to invest — both time, money, and brainpower. It’s a marathon without a finish line. However, if you’ve read this far, you’re likely to be interested in how your company (or you personally) can leverage the AI opportunity. It’s a cost vs benefit analysis which continues to be a positive benefit-cost ratio for us and contributed to our leadership position in markets despite larger competitors.
Years ago, our progressive CEO spent some time at Singularity University, an event that sparked our AI journey. He later sent our entire executive team to attend a similar program aimed at Achieving 10x Growth by leveraging rapidly accelerating technologies. In addition to being mostly entrepreneurs and technologists used to taking risks, having an executive team with a digital first mindset enabled me to take the first steps towards building our analytics team. I also benefited from a data science executive education program which contributed to my appreciation of Statistics and the science of data. This program included sessions with Data Scientists from global AI champions such as Facebook, Uber, Citibank, AirBnB, Netflix, Singtel. Some of these companies even hosted visits to their HQ in Silicon Valley to meet with their data science teams. Our startup could not match the investment in talent that these companies have spent over several years, but it gave us a peek into the AI opportunity with the diversity of applications from various industries.
How does AI work?
What is the difference between AI, ML, and Deep Learning? How do I apply it in my business? Below is a diagram from Singularity University since it provides a high-level concept of the 3-phase AI cycle and simplest answer to the question: How does AI work?
Think of AI as a general purpose tool that is configured to search for well-defined “patterns.” It sifts through huge volumes of data to formulate answers … while iteratively learning from its mistakes!
1. Perceive Environment — this starts with collecting data as diverse as transactional data from IT systems or IoT devices, video stream, recording of human voices, or digitized images of MRIs or X-Rays
2. Detect Patterns — given a hypothesis or business problem, use data to identify patterns using various AI algorithms
3. Learn from patterns — update experiential memory and repeat the cycle
Begin with the pattern (aka the problem)
While data is critical to the AI cycle, the first question that needs to be answered is “what problem are we trying to solve?” I’ll use an example that’s foremost in everybody’s mind since it has to do with our health. Some of us may have paid a visit to our doctor with a lingering question — “do I have cancer?” For the past several years, research labs, doctors, and radiologists at University of California, Los Angeles (UCLA) have been utilizing AI to improve cancer diagnostics and concluded that AI can spot subtle patterns that can easily be missed by humans. In the example below, it may be a pattern invisible to the human eye that an AI system can be trained to detect (see left diagram). This was compared to a scan of the same patient 4 years later (right diagram) showing an early stage cancer. This AI system called FocalNet has been trained to detect other patterns such as prostate cancer and was also intelligent enough to predict its aggressiveness.
And the most amazing thing is that there’s virtually no limit to its ability to learn as it is fed more data and computing power. Having said this, I should emphasize that current AI tools are designed to enhance decisions such as the case of doctors and radiologists above, while the most advanced are designed for complete autonomy such as full self-driving vehicles. If you’re still not convinced of the AI opportunity, then I recommend you research the value and impact of AI across industries (e.g. Applications and value of deep learning by industry sector, McKinsey 2018).
As a small business, the problems are more commercially driven than the life-saving or industry-wide goals exemplified by the deep science problems. Below are some examples of business value (stated as a question) which could be a proof of concept (PoC) for newly formed teams. Note that each problem will have its own definition of patterns, algorithms, and data representations (see links to articles):
- How do we prevent fraudulent transactions and reduce our costs?
- How do we increase revenue by getting consumers to buy more products per order?
- How much should I sell my used Mercedes car? (see Mercedes Phoenix Pricing project)
- What are some of the challenges that AI can tackle in Marketing? (see top use cases for AI in marketing)
Data for defining patterns and training
The 3-phase AI cycle from Singularity above is a highly simplified concept meant to jumpstart the understanding of the general principles of AI. However, vast amounts of sophistication and algorithms lay below what is visible in our AI iceberg. One of this is data science which is a personal passion and is essential to (Phase 1) perceiving environment and (Phase 3) learning from patterns. In order to maintain brevity, I’ve reserved this topic for a future blog, along with Algorithms (used for detecting patterns), and the different approaches machines learn. Think of it as part 2 of our AI Introduction designed for those that want to roll-up their sleeves and “open the hood” to see how the machine consumes data, spits out answers, and learns.
However, if you’re motivated to get some “hands-on” introduction to data science, machine learning or AI, you can view this free online book Python Data Science Handbook by Jake VanderPlas or Hands-On Machine Learning with Scikit-Learn and TensorFlow (not free) which takes you through the toolkits used by Data Scientists such as Python programming language and several machine learning libraries and platforms. There’s also Harvard University’s CS50’s Introduction to Artificial Intelligence with Python (free until Dec 2021). I highly recommend these for those who want to be AI practitioners and those tasked to build a team to help them understand the skills needed to recruit and interview candidates.
AI for Good
You’ve probably watched some of the sci-fi movies of humans vs machines in a post-apocalyptic world (e.g. Terminator The Movie) given the domination of super-intelligent machines. Or if you’re a gaming enthusiast, you may have experienced playing an android character in a PS4 game called Detroit: Become Human (top photo). This imagines a future where 25% of human jobs are performed by androids. Unlike Terminator, this timeline occurs before the technological singularity, when machines reach the irreversible point of super-intelligence. This spectrum in AI is sometimes referred to as Artificial General Intelligence (AGI) and some brave souls have published fearless forecasts of when this may be realized in the future.
Whichever future you are willing to imagine for our world, AI has demonstrated its ability to enhance our lives along with other countless possibilities across industries like Healthcare. It’s true that AI is unmatched by previous technological innovations, but we should also be aware of its dark side especially in military applications. Fortunately, there’s growing focus on AI for Good and awareness of risks can keep Dark AI in check. Large companies with billions of dollars in AI Research have made AI ethics and principles transparent. In the meantime, as our world evolves, let’s start our journey and take our place in the AI ecosystem.
