“One word: Plastics”. The famous quote in the first ten minutes of 1967’s The Graduate epitomizes what artificial intelligence has become in this new millennium. Plastics were the wave of the future in the 1960s. Much in the same way, AI has become a mainstream staple of twenty-first-century society. Self-driving cars, Amazon’s Alexa, Apple’s Siri, Google’s intelligent search and language translation tools, Facebook and Instagram’s face recognition are just a few expressions of this ever-changing and evolving field that has moved from the backend of computer science research labs to the frontline of society, corporations, and popular culture.
Modern artificial intelligence is deeply rooted in machine learning, a discipline through which computer software, in the form of algorithms, attains the ability to learn and improve from previous experience automatically.
Learning from experience means learning from data, and data is something of which humanity has plenty. The entire amount of digital data is expected to reach 44 zettabytes in 2020. This is 40 times more bytes than the number of stars in the observable universe, or if you remember your chemistry 101, only 14 times less than the Avogadro number. Add to this that the internet has 4.4 billion internet users as of June 2019, and that, as an IBM marketing cloud study points out, 90% of its data has been created since 2016, and you will get a glimpse of the magnitude of the digital phenomenon. It is this massive volume of varied types of data, commonly labeled as Big Data, and the velocity with which it is created, that feeds the machine learning algorithms of modern, data-driven AI.
"Modern artificial intelligence is deeply rooted in machine learning, a discipline through which computer software in the form of algorithms, attains the ability to automatically learn and improve from previous experience"
Of course, Big Data creation is only half of the story of modern AI’s success. The other half is due to the prodigious advances in machine learning algorithm development in the last twenty years, in particular, the skyrocketing resurgence of neural network technology in the last decade commonly referred to as deep learning. Most of the AI breakthroughs that we hear about in the media are due to deep learning. The ability of current neural network architectures to learn representations from data has revolutionized the fields of computer vision, and natural language and speech processing. If you have been paying attention to your interactions with Siri and other digital assistants, you have probably marveled at their progress. There is no magic behind it; it is just deep learning and robust engineering.
These machine learning algorithms, trained from data, produce insights. Data-driven AI has allowed us to see things that we could not see before, from human behavior in the form of sentiment and interest to prediction of weather patterns, worker and student performance, and detection of cancerous tumors with greater precision than human experts(to mention just a few applications). Data-driven AI is, as Michael Mahoney (Berkeley) has suggested, the digital era version of the microscope and the telescope.
Andrew Ng (Stanford, deeplearning.ai, AI fund) describes modern AI as the new electricity. He tells the story of how manufacturing was done before electricity, relying on steam or water. When electricity came along, companies started using electricity in a piecemeal manner, sprinkling some electricity in their factories. But then companies realized that to harness the full advantage of electricity, they needed to redesign their manufacturing plants. Nowadays, we are still in the early days of sprinkling AI, but the trend is moving quickly in the direction of the widespread use of AI. This will mean redesigning our processes and our products. Just think of one artificial intelligence product: the self-driving car. We are not fully there yet, but when we are, it won’t be enough to scatter a few self-driving cars in the highways. We will need a full redesign of our land transport system. A time will come when we humans will relinquish the driver’s seat for good.
This takes us to disruption: disruption by AI and disruption by AI-driven organizations. I often tell my students that if you work in any organization today, no matter what size, or whether it’s retail, entertainment, hospitality, or even higher education, you are going to see your organization disrupted by AI-driven companies like Amazon, Netflix, Apple, Google, Airbnb, Coursera, or any of the myriad AI startups that mushroom every year. Data-driven AI is an enabler, a game-changer, and an equalizer.
But despite its negative connotations, the word ‘disruption’ need not scare us. Disruption is a driver of progress. Joseph Schumpeter, one of the most influential economic thinkers of the 20th century, described creative destruction as “the process of industrial mutation that incessantly revolutionizes the economic structure from within, incessantly destroying the old one, incessantly creating a new one.” That’s the promise delivered by data-driven AI, and this is a good thing for the advancement of civilization.
We must, however, beware of the cautionary trajectory of plastics, from technological wonder to environmental catastrophe. Humanity will not be taken down by the terminator robots at Skynet, but it can fall victim to the siren songs of data-driven AI. As Shoshana Zuboff argues in her latest work, The Age of Surveillance Capitalism, free will may be at risk, as we become puppets in a market that trades behavior as an asset made for profit by predicting -and spurring- all our needs.
We need ethical, responsible AI. Good AI. It is up to us to demand it.