"Dedicated to our Future Selves. May we take care of one another and the world by making good use of Artificial Intelligence."
Excerpt From: Francesco Mosconi, “Zero to Deep Learning.”
Artificial Intelligence is the most powerful technology of the 21st century. It is the fourth, and possibly the last, technological revolution in the history of humanity and it is impacting all aspects of our society and civilization.
Like the agricultural revolution, the industrial revolution and the Internet revolution of the late 20th century, the AI revolution is powered by a set of core technologies that are enabling new applications, breakthroughs, and new insights.
In the case of AI, these enablers are cheap sensors, cheap data storage, and cheap computing power. Cheap sensors and storage ushered in the rst epoch of the AI revolution: the era of Big Data. In the rst decade of the 21st century, Big Data became a “must have” technology for modern business. The ability to record consumer activity and interactions, as well as supply chain and production data, allowed companies to gather real-time intelligence on an unprecedented scale. It also enabled data-driven applications, a new set of products with data at the core. Social media platforms, like LinkedIn and Facebook, are excellent examples of these.
The ubiquity of data paved the way for the second epoch in the AI revolution: the era of Machine Learning and Deep Learning. Both of these technologies have been around since the second half of the 20th century but had rarely seen a broad application. Historically, this was mainly because their performance on real-world tasks was not good enough, due to the absence of large training data sets.
Everything changed in 2012. the ImageNet challenge is a computer-science competition, founded in 2009, where research scientists try to design algorithms that achieve the highest accuracy in visual object recognition, on a data set of millions of images. Until 2011, recognition errors were above 25%. However, in 2012 an algorithm based on Deep Learning brought the error down to 16% for the first time. This incredible achievement is widely recognized as the start of the Deep Learning revolution, and since then, we have seen Deep Learning conquer an increasing number of domains.
What's interesting about this is not so much that a new algorithm was applied (in fact, artificial neural networks had been around for years, if not decades, already). It's the fact that new hardware breakthroughs and the availability of large data sets made it possible to exploit such an algorithm. After conquering image recognition, Deep Learning made breakthroughs in many other fields of machine intelligence including:
- machine translation (think: Google Translate)
- speech recognition (think: Amazon Alexa, Apple Siri, Google Now, etc.)
- recommendation systems (think: Netflix, Amazon, Spotify, etc.)
- search engines and information retrieval (think: Google, Baidu, Bing, etc.)
- medical diagnostics (think: applications in cancer screening and imaging)
- fraud detection (think: Visa, Mastercard, Stripe, PayPal, etc.)
- forecasting (think: hedge funds, utility companies etc.)
- robotics & automation (think: Tesla autopilot)
What exact is Deep learning?
Deep Learning is a technology capable of learning very complex relation between arbitrary inputs and outputs. It is based on a mathematical concept called an Artificial Neural Network (ANN), which is nothing more than a fancy name for a very complicated mathematical function. It had been relegated to a few research labs scattered across the world until the early 2010s. However, since then, every major technology company has released an open source framework that implements the core concepts of Deep Learning, the most famous being Tensorflow by Google.
Why should you care?
According to research published by Element AI, there are roughly 22,000 PhD-level researchers in the field of Artificial Intelligence. Tencent, a Chinese technology giant, estimated that there were 300,000 engineers equipped with AI knowledge at the end of 2017, while the demand for talent from companies is in the millions. To give an idea of the magnitude of the Deep Learning revolution, China envisions building a $1 trillion AI industry by 2030 and are investing billions of dollars to make this happen. Private companies compete for AI talent so fiercely that salaries can reach millions of dollars, and private equity and acquisition deals in the hundreds of millions are common.
The industrialization phase of the Deep Learning revolution is well underway.
In mathematical terms, we can think of the 1-D array as a vector, the 2-D array as a matrix and the 3-D array as a tensor of order 3.