AI Superpowers

A hypothetical AI that can match or even surpass humans at any task. Lee considers AGI a work of science fiction; though he believes it may be developed some time in the future, he does not expect to see AGI in our lifetimes.

Extensive data sets which are collected for specific purposes, usually related to analyzing patterns and trends in human behavior. Lee writes that big data is essential to truly powerful AI.

City brains are AI-powered software systems that collect and analyze data for urban management. The first city brain was introduced by the Chinese tech company Alibaba Group in 2016. Lee presents city brains as an example of China’s willingness to overhaul and shape policy around AI.

A machine learning method based on a neural network. Deep learning powers the majority of modern AIs. Lee believes it is the driving force behind the “AI revolution,” which is currently in its early stages.

In economics: the process by which skilled labor becomes obsolete. Deskilling is usually associated with the development of technology that automates this labor, eliminating the need for skilled workers (e.g., people who work in clothing factories don’t need any knowledge of tailoring or fashion design to produce garments).

GPTs are revolutionary technologies that drastically change the socioeconomic structures they’re introduced to on the national and/or global level (e.g., the steam engine). Lee believes that AI and deep learning will soon be regarded as GPTs.

GMI (or GBI—guaranteed basic income) is a policy originally introduced by Dr. Martin Luther King, Jr. Like a universal basic income (UBI), GMI is a social-welfare system intended to mitigate poverty. GMI would ensure that all people receive cash benefits from their government to support their basic needs. Lee regards GMI as an insufficient response to the oncoming jobs crisis.

The global shift from an agrarian economy (defined by individual means of production) to one defined by mass manufacturing (defined by automized production on a mass scale). This revolution took place over the course of the 18th and 19th centuries. Lee uses the Industrial Revolution as a point of comparison for the AI boom.

In machine learning: labeling is the process of identifying data (such as images or text files) a given AI is being trained on. These labels are used to help the AI differentiate between items (e.g., a picture of a cat or a picture of “not a cat”), which helps it “learn” to make predictions about novel data (e.g., after being trained on “cat”/”no cat” images, an AI should be able to guess whether a new image depicts a cat).

Coined in Eric Reis’s The Lean Startup (2011). A lean startup releases Minimum Viable Products (MVPs) in short, rapid cycles. This strategy centers efficiency and the iterative process: MVPs are released, consumer feedback is gathered, and the products are tweaked to meet the customer base’s needs and demands.

The field of training computer systems, through exposure to massive quantities of data, to “learn” and autonomously solve problems. Machine learning is the cornerstone of AI; it is the process by which all AI is trained.

Financial services that target individuals and small organizations with little or no access to traditional financial services. While its proponents argue that microfinance helps to mitigate poverty, critics warn that the microfinance ecosystem sends impoverished people into debt.

A type of AI that is created to perform a single/limited task.

Also called “artificial neural networks” (ANN). A computer system modeled after human/mammalian brains. An ANN is trained to weigh different options and probabilities by determining the differences between the data its fed.

Online-to-Offline startups. A marketing strategy that nudges online consumers toward shopping at brick-and-mortar locations. O2O nests into Lee’s term “Online-Merge-Offline,” which nudges consumers to interact with their services and products both digitally and under traditionally “offline” circumstances.

A term coined by Lee which represents the blurring of online and offline experiences. Lee asks: “When you order a full meal just by speaking a sentence from your couch, are you online or offline? When your refrigerator at home tells your shopping cart at the store that you’re out of milk, are you moving through a physical world or a digital one?” (128).

One of the three basic tenants of machine learning (alongside supervised learning and unsupervised learning). RL is a machine learning training method that rewards desired behaviors and disincentivizes undesired behaviors.

A 20th-century rivalry between the United States and the Soviet Union centered on achieving superior spaceflight technology. This competition was part of the tension between the two nations during the Cold War. Lee uses the space race as a historical analog for the “AI race” between the US and China.

A newly formed business designed to start small and grow rapidly. Lee’s background in venture capital gives him a particular interest in and familiarity with startups, as they frequently rely on VC funding to grow. Many giants in the modern tech economy began as startups, linking them to the AI economy.

Data points with a noticeable correlation to a specific outcome. These are usually associated with cause-and-effect relationships (e.g., Eye color is a genetically heritable trait. If you wanted to predict a child’s chances of being born with blue eyes, their parents’ eye colors would be strong features).

The philosophy or belief that technology is an unalloyed social good that will ultimately benefit mankind. Lee uses this term to describe the Silicon Valley ethos: “a wide-eyed techno-optimism, a belief that every person and company can truly change the world through innovative thinking” (37).

In computer science, a toy problem is a computing/coding puzzle that is used to explore and explain concepts and problem-solving techniques (e.g., teaching an AI to play tic-tac-toe).

In machine learning, using a pre-trained neural network to address a new problem or data set. This process can help compensate for a dearth of data. It is also an important aspect of deep learning.

In venture capitalism: a private startup that is valued at least $1 billion. Lee discusses unicorn startups briefly while exploring Shenzehn’s startup ecosystem, which he claims “has produced four unicorn startups” (139).

Similar to GMI, UBI consists of unconditional recurring payments from the government to its citizens. Unlike GMI—which requires citizens to meet specific need-based benchmarks to qualify for services—UBI guarantees a flat payout for all citizens, regardless of their income or needs.

An investor (or investment firm) who funds startups and small businesses with privately owned equity. As a venture capitalist himself, Lee focuses on VC practices and outcomes throughout AI Superpowers.

When a company takes ownership of its own productive means and processes, rather than outsourcing to external contractors or suppliers. Lee uses this concept to define his terms “going light” and “going heavy”: “The terms refer to how involved an internet company becomes in providing goods or services. They represent the extent of vertical integration as a company links up the on-and offline worlds” (80).