Did you know that “in the entire world, fewer than 10,000 people have the skills necessary to tackle serious artificial intelligence (AI) research”? That is according to Element AI, an independent lab in Montreal (reported in The New York Times, October 23, 2017).
The competition for AI talent is fierce, and start-ups—once a magnet for computer talent—are having a hard time attracting the skills they need. How can they compete with the deep pockets of Google, Facebook, and Microsoft? Like professional athletes, top AI people can command compensation in salary and shares valued in single or double-digit millions. Google bought DeepMind in 2014 for an estimated $650 million. At that time, DeepMind had 50 employees—which now has expanded to 400—with staff costs in 2016 of $138 million (or $345,000 per employee).
To help alleviate the shortage, companies such as Google and Facebook run classes for current employees to teach “deep learning,” and nonprofits such as Fast.ai and companies such as Deeplearning.ai offer online courses.
The consensus is that this talent shortage won’t be allayed for years, which is good news for those with the time and energy to still jump on the AI bandwagon. Where to begin? Whenever I want to learn about a new field, I start by compiling a glossary of key terms. This, at least, gives me the bare bones, and the sense of whether I want to go deeper, or not. It also allows me to showcase my knowledge to friends and others—at least with those who know even less than I do.
Some Key Terms in AI
Artificial Intelligence: The definition depends on who you ask (helpful!), but basically, it is the field of computer engineering focused on developing systems that can gather data, learn, make decisions, and solve problems, i.e., simulate human intelligence and behavior.
Artificial Neural Network (ANN): Standard algorithms don’t work well with noisy or incomplete data. That’s where ANNs come in—a non-linear computational model based on the neural structure of the brain that can learn on its own. ANNs can learn to perform tasks such as classification, prediction, decision-making, and visualization by considering examples.
Algorithm: This is the most important part of AI—a formula or set of rules/commands for a system to perform a task, learn on its own, find answers, solve problems.
Autonomous: An AI machine that doesn’t require input from a human operator to function properly.
Cognitive Computing: A computer model mimicking how the human thinks using data mining, natural language processing, and pattern recognition.
Data Mining: A process by which hidden patterns are discovered within large sets of data to provide useful information for future use.
Deep Learning: A multilayered neural network that learns representations of the world and stores them in nested hierarchies of concepts (i.e., cascading layers of information). It will recognize objects beginning with simple building blocks such as straight and curved lines; then eyes, mouths, and noses; then faces; and then specific facial features.
Machine Learning: In general, machine learning can be categorized into three types:
- Supervised: There is always a specific preset outcome determined by a human before the machine begins to learn, e.g., e-mails are categorized into “junk” or “regular.”
- Unsupervised: There is no right or wrong answer. The target function for the system is to extract structure from the data, e.g., by identifying clusters. Think of all the data collected from the use of store discount cards. The machine sorts the data into clusters without predefined categories.
- Reinforcement: The machine achieves its target function (see below) through experimentation and maximizing reward.
Natural Language Processing (NLP): The ability of computers to process natural human languages and derive meaning from them. This typically involves machine interpretation of text or speech. If you’ve used AI assistants such as Alexa, Siri, or Cortana, you are already familiar with this capability.
Strong AI: Field of AI that works toward the goal of making a conscious machine with the ability to perceive, feel, reason, make judgments, plan, learn, and have self-awareness. Weak AI can simulate human cognition, but the as-yet-hypothetical Strong AI would have human cognition. Some predict it can be done by 2030-2045, others say within the next century, and others believe it can’t be done at all.
Target Function: The specific task an AI machine has been designed and programmed to complete.
Turing Test: Basically, the test is: “Can an AI ‘fool’ a person into believing he or she is seeing or interacting with a real person, e.g., is this a real therapist or a machine?”
Weak AI: Also known as narrow AI. A weak AI machine can appear to be intelligent, e.g., a chess game system can beat the very best human players, but it doesn’t think or plan. All it’s moves are based on some rules fed into it by a human. Weak AI operates within a predetermined range of skills and a single or small set of tasks. Most AI today is of the weak kind.
While the revenues from AI were only $644 million in 2016, market research firm Tractica estimates this will rise to $15 billion by 2022.
Does the world of AI worry you? If so you are not alone. The Future of Life Institute has created what it calls the Asimolar AI Principles. Cosmologist Stephen Hawking and Elon Musk of SpaceX and Tesla are both members of the Board of Advisors. The 23 principles aim to ensure the development of ethical, safe AI. The first principle is: “The goal of AI research should be to create not undirected intelligence, but beneficial intelligence.”
Those of us interested in intelligence and learning in all its forms need to be able to join in the conversation. Start learning the language of AI today. It could pay off in many ways.
Terence Brake is the director of Learning & Innovation, TMA World (http://www.tmaworld.com/training-solutions/), which provides blended learning solutions for developing talent with borderless working capabilities. Brake specializes in the globalization process and organizational design, cross-cultural management, global leadership, transnational teamwork, and the borderless workplace. He has designed, developed, and delivered training programmes for numerous Fortune 500 clients in the United States, Europe, and Asia. Brake is the author of six books on international management, including “Where in the World Is My Team?” (Wiley, 2009) and e-book “The Borderless Workplace.”
