The Age of AI

CHAPTER 1

WHERE WE ARE

IN LATE 2017, a quiet revolution occurred. AlphaZero, an artificial intelligence (AI) program developed by Google DeepMind, defeated Stockfish—until then, the most powerful chess program in the world. AlphaZero’s victory was decisive: it won twenty-eight games, drew seventy-two, and lost none. The following year, it confirmed its mastery: in one thousand games against Stockfish, it won 155, lost six, and drew the remainder.¹

Normally, the fact that a chess program beat another chess program would only matter to a handful of enthusiasts. But AlphaZero was no ordinary chess program. Prior programs had relied on moves conceived of, executed, and uploaded by humans—in other words, prior programs had relied on human experience, knowledge, and strategy. These early programs’ chief advantage against human opponents was not originality but superior processing power, enabling them to evaluate far more options in a given period of time. By contrast, AlphaZero had no preprogrammed moves, combinations, or strategies derived from human play. AlphaZero’s style was entirely the product of AI training: creators supplied it with the rules of chess, instructing it to develop a strategy to maximize its proportion of wins to losses. After training for just four hours by playing against itself, AlphaZero emerged as the world’s most effective chess program. As of this writing, no human has ever beaten it.

The tactics AlphaZero deployed were unorthodox—indeed, original. It sacrificed pieces human players considered vital, including its queen. It executed moves humans had not instructed it to consider and, in many cases, humans had not considered at all. It adopted such surprising tactics because, following its self-play of many games, it predicted they would maximize its probability of winning. AlphaZero did not have a strategy in a human sense (though its style has prompted further human study of the game). Instead, it had a logic of its own, informed by its ability to recognize patterns of moves across vast sets of possibilities human minds cannot fully digest or employ. At each stage of the game, AlphaZero assessed the alignment of pieces in light of what it had learned from patterns of chess possibilities and selected the move it concluded was most likely to lead to victory. After observing and analyzing its play, Garry Kasparov, grand master and world champion, declared: “chess has been shaken to its roots by AlphaZero.”² As AI probed the limits of the game they had spent their lives mastering, the world’s greatest players did what they could: watched and learned.

In early 2020, researchers at the Massachusetts Institute of Technology (MIT) announced the discovery of a novel antibiotic that was able to kill strains of bacteria that had, until then, been resistant to all known antibiotics. Standard research and development efforts for a new drug take years of expensive, painstaking work as researchers begin with thousands of possible molecules and, through trial and error and educated guessing, whittle them down to a handful of viable candidates.³ Either researchers make educated guesses among thousands of molecules or experts tinker with known molecules, hoping to get lucky by introducing tweaks into an existing drug’s molecular structure.

MIT did something else: it invited AI to participate in its process. First, researchers developed a “training set” of two thousand known molecules. The training set encoded data about each, ranging from its atomic weight to the types of bonds it contains to its ability to inhibit bacterial growth. From this training set, the AI “learned” the attributes of molecules predicted to be antibacterial. Curiously, it identified attributes that had not specifically been encoded—indeed, attributes that had eluded human conceptualization or categorization.

When it was done training, the researchers instructed the AI to survey a library of 61,000 molecules, FDA-approved drugs, and natural products for molecules that (1) the AI predicted would be effective as antibiotics, (2) did not look like any existing antibiotics, and (3) the AI predicted would be nontoxic. Of the 61,000, one molecule fit the criteria. The researchers named it halicin—a nod to the AI HAL in the film 2001: A Space Odyssey.⁴

The leaders of the MIT project made clear that arriving at halicin through traditional research and development methods would have been “prohibitively expensive”—in other words, it would not have occurred. Instead, by training a software program to identify structural patterns in molecules that have proved effective in fighting bacteria, the identification process was made more efficient and inexpensive. The program did not need to understand why the molecules worked—indeed, in some cases, no one knows why some of the molecules worked. Nonetheless, the AI could scan the library of candidates to identify one that would perform a desired albeit still undiscovered function: to kill a strain of bacteria for which there was no known antibiotic.

Halicin was a triumph. Compared to chess, the pharmaceutical field is radically complex. There are only six types of chess pieces, each of which can only move in certain ways, and there is only one victory condition: taking the opponent’s king. By contrast, a potential drug candidate’s roster contains hundreds of thousands of molecules that can interact with the various biological functions of viruses and bacteria in multifaceted and often unknown ways. Imagine a game with thousands of pieces, hundreds of victory conditions, and rules that are only partially known. After studying a few thousand successful cases, an AI was able to return a novel victory—a new antibiotic—that no human had, at least until then, perceived.

Most beguiling, though, is what the AI was able to identify. Chemists have devised concepts such as atomic weights and chemical bonds to capture the characteristics of molecules. But the AI identified relationships that had escaped human detection—or possibly even defied human description. The AI that MIT researchers trained did not simply recapitulate conclusions derived from the previously observed qualities of the molecules. Rather, it detected new molecular qualities—relationships between aspects of their structure and their antibiotic capacity that humans had neither perceived nor defined. Even after the antibiotic was discovered, humans could not articulate precisely why it worked. The AI did not just process data more quickly than humanly possible; it also detected aspects of reality humans have not detected, or perhaps cannot detect.

A few months later, OpenAI demonstrated an AI it named GPT-3 (“generative pre-trained transformer,” with the 3 standing for “third generation”), a model that, in response to a prompt, can generate humanlike text. Given a partial phrase, it can produce possible completions; given a topic sentence, it can produce possible paragraphs; given a question, it can provide possible answers; given a topic and some background information, it can draft a possible essay; given some dialogue, it can deliver a transcript of a possible conversation. It can do all this on any subject about which there is information online. It trains for its task by consuming that information.

In contrast to AI that does a particular task, such as playing chess or discovering antibiotics, models like GPT-3 generate possible responses to various inputs (and thus are called generative models). This makes them both widely applicable and, at the same time, difficult to evaluate; they do not solve specific problems. Sometimes, their results seem uncannily human. Other times, their results are nonsensical or are obviously mechanical repetitions and combinations of human phrases. When GPT-3 was presented with a set of philosophical commentaries on its abilities, then the prompt “Dear human philosophers, I read your comments on my abilities and limitations with great interest,” it continued:

Without comparing this text to the commentaries that were provided to GPT-3, one cannot judge how original or creative its response was, but it certainly appears sophisticated.

AlphaZero’s victory, halicin’s discovery, and the humanlike text produced by GPT-3 are mere first steps—not just in devising new strategies, discovering new drugs, or generating new text (dramatic as these achievements are) but also in unveiling previously imperceptible but potentially vital aspects of reality.

In each case, developers created a program, assigned it an objective (winning a game, killing a bacterium, or generating text in response to a prompt), and permitted it a period—brief by the standards of human cognition—to “train.” By the end of the period, each program had mastered its subject differently from humans. In some cases, it obtained results that were beyond the capacity of human minds—at least minds operating in practical time frames—to calculate. In other cases, it obtained results by methods that humans could, retrospectively, study and understand. In others, humans remain uncertain to this day how the programs achieved their goals.

THIS BOOK is about a class of technology that augurs a revolution in human affairs. AI—machines that can perform tasks that require human-level intelligence—has rapidly become a reality. Machine learning, the process the technology undergoes to acquire knowledge and capability—often in significantly briefer time frames than human learning processes require—has been continually expanding into applications in medicine, environmental protection, transportation, law enforcement, defense, and other fields. Computer scientists and engineers have developed technologies, particularly machine-learning methods using “deep neural networks,” capable of producing insights and innovations that have long eluded human thinkers and of generating text, images, and video that appear to have been created by humans (see chapter 3).

AI, powered by new algorithms and increasingly plentiful and inexpensive computing power, is becoming ubiquitous. Accordingly, humanity is developing a new and exceedingly powerful mechanism for exploring and organizing reality—one that remains, in many respects, inscrutable to us. AI accesses reality differently from the way humans access it. And if the feats it is performing are any guide, it may access different aspects of reality from the ones humans access. Its functioning portends progress toward the essence of things—progress that philosophers, theologians, and scientists have sought, with partial success, for millennia. Yet as with all technologies, AI is not only about its capabilities and promise but also about how it is used.

While the advancement of AI may be inevitable, its ultimate destination is not. Its advent, then, is both historically and philosophically significant. Attempts to halt its development will merely cede the future to the element of humanity courageous enough to face the implications of its own inventiveness. Humans are creating and proliferating nonhuman forms of logic with reach and acuity that, at least in the discrete settings in which they were designed to function, can exceed our own. But AI’s function is complex and inconsistent. In some tasks, AI achieves human—or superhuman—levels of performance; in others (or sometimes the same tasks), it makes errors even a child would avoid or produces results that are utterly nonsensical. AI’s mysteries may not yield a single answer or proceed straightforwardly in one direction, but they should prompt us to ask questions. When intangible software acquires logical capabilities and, as a result, assumes social roles once considered exclusively human (paired with those never experienced by humans), we must ask ourselves: How will AI’s evolution affect human perception, cognition, and interaction? What will AI’s impact be on our culture, our concept of humanity, and, in the end, our history?

FOR MILLENNIA, humanity has occupied itself with the exploration of reality and the quest for knowledge. The process has been based on the conviction that, with diligence and focus, applying human reason to problems can yield measurable results. When mysteries loomed—the changing of the seasons, the movements of the planets, the spread of disease—humanity was able to identify the right questions, collect the necessary data, and reason its way to an explanation. Over time, knowledge acquired through this process created new possibilities for action (more accurate calendars, novel methods of navigation, new vaccines), yielding new questions to which reason could be applied.

However halting and imperfect this process may have been, it has transformed our world and fostered confidence in our ability, as reasoning beings, to understand our condition and confront its challenges. Humanity has traditionally assigned what it does not comprehend to one of two categories: either a challenge for the future application of reason or an aspect of the divine, not subject to processes and explanations vouchsafed to our direct understanding.

The advent of AI obliges us to confront whether there is a form of logic that humans have not achieved or cannot achieve, exploring aspects of reality we have never known and may never directly know. When a computer that is training alone devises a chess strategy that has never occurred to any human in the game’s millennial history, what has it discovered, and how has it discovered it? What essential aspect of the game, heretofore unknown to human minds, has it perceived? When a human-designed software program, carrying out an objective assigned by its programmers—correcting bugs in software or refining the mechanisms of self-driving vehicles—learns and applies a model that no human recognizes or could understand, are we advancing toward knowledge? Or is knowledge receding from us?

Humanity has experienced technological change throughout history. Only rarely, however, has technology fundamentally transformed the social and political structure of our societies. More frequently, the preexisting frameworks through which we order our social world adapt and absorb new technology, evolving and innovating within recognizable categories. The car replaced the horse without forcing a total shift in social structure. The rifle replaced the musket, but the general paradigm of conventional military activity remained largely unaltered. Only very rarely have we encountered a technology that challenged our prevailing modes of explaining and ordering the world. But AI promises to transform all realms of human experience. And the core of its transformations will ultimately occur at the philosophical level, transforming how humans understand reality and our role within it.

The unprecedented nature of this process is both profound and perplexing; having entered it gradually, we are undergoing it passively, largely unaware of what it has done and is likely to do in the coming years. Its foundation was laid by computers and the internet. Its zenith will be AI that is ubiquitous, augmenting human thought and action in ways that are both obvious (such as new drugs and automatic language translations) and less consciously perceived (such as software processes that learn from our movements and choices and adjust to anticipate or shape our future needs). Now that the promise of AI and machine learning has been demonstrated, and the computing power needed to operate sophisticated AI is becoming readily available, few fields will remain unaffected.

Persistently, often imperceptibly, but now unavoidably, a web of software processes is unfolding across the world, driving and perceiving the pace and scope of events, overlaying aspects of our daily life—homes, transportation, news distribution, financial markets, military operations—our minds once traveled alone. As more software incorporates AI, and eventually operates in ways that humans did not directly create or may not fully understand, it will be a dynamic information-processing augmenter of our capabilities and experiences, both shaping and learning from our actions. Frequently, we will be aware that such programs are assisting us in ways that we intended. Yet at any given moment, we may not know what exactly they are doing or identifying or why they work. AI-powered technology will become a permanent companion in perceiving and processing information, albeit one that occupies a different “mental” plane from humans. Whether we consider it a tool, a partner, or a rival, it will alter our experience as reasoning beings and permanently change our relationship with reality.

The journey of the human mind to the central stage of history took many centuries. In the West, the advent of the printing press and the Protestant Reformation challenged official hierarchies and altered society’s frame of reference—from a quest to know the divine through scripture and its official interpretation to a search for knowledge and fulfillment through individual analysis and exploration. The Renaissance witnessed the rediscovery of classical writings and modes of inquiry that were used to make sense of a world whose horizons were expanding through global exploration. During the Enlightenment, René Descartes’s maxim, Cogito ergo sum (I think, therefore I am), enshrined the reasoning mind as humanity’s defining ability and claim to historical centrality. This notion also communicated the sense of possibility engendered by disrupting the established monopoly on information, which was largely in the hands of the church.

Now the partial end of the postulated superiority of human reason, together with the proliferation of machines that can match or surpass human intelligence, promises transformations potentially more profound than even those of the Enlightenment. Even if advances in AI do not produce artificial general intelligence (AGI)—that is, software capable of human-level performance of any intellectual task and capable of relating tasks and concepts to others across disciplines—the advent of AI will alter humanity’s concept of reality and therefore of itself. We are progressing toward great achievements, but those achievements should prompt philosophical reflection. Four centuries after Descartes promulgated his maxim, a question looms: If AI “thinks,” or approximates thinking, who are we?

AI will usher in a world in which decisions are made in three primary ways: by humans (which is familiar), by machines (which is becoming familiar), and by collaboration between humans and machines (which is not only unfamiliar but also unprecedented). AI is also in the process of transforming machines—which, until now, have been our tools—into our partners. We will begin to give AI fewer specific instructions about how exactly to achieve the goals we assign it. Much more frequently, we will present AI with ambiguous goals and ask: “How, based on your conclusions, should we proceed?”

This shift is neither inherently threatening nor inherently redemptive. Yet it is sufficiently different that it very likely will alter the trajectories of societies and the course of history. The continued integration of AI into our lives will bring about a world in which seemingly impossible human goals are achieved and where achievements once presumed to be exclusively human—writing a song, discovering a medical treatment—are generated by, or in collaboration with, machines. This development will transform entire fields by enveloping them in AI-assisted processes, with the lines between purely human, purely AI, and hybrid human-AI decision making sometimes becoming difficult to define.

In the political realm, the world is entering an era in which big data–driven AI systems are informing growing aspects: the design of political messages; the tailoring and distribution of those messages to various demographics; the crafting and application of disinformation by malicious actors aiming to sow social discord; and the design and deployment of algorithms to detect, identify, and counter disinformation and other forms of harmful data. As AI’s role in defining and shaping the “information space” grows, its role becomes more difficult to anticipate. In this space, as in others, AI sometimes operates in ways even its designers can only elaborate in general terms. As a result, the prospects for free society, even free will, may be altered. Even if these evolutions prove to be benign or reversible, it is incumbent on societies across the globe to understand these changes so they can reconcile them with their values, structures, and social contracts.

Defense establishments and commanders face evolutions no less profound. When multiple militaries adopt strategies and tactics shaped by machines that perceive patterns human soldiers and strategists cannot, power balances will be altered and potentially more difficult to calculate. If such machines are authorized to engage in autonomous targeting decisions, traditional concepts of defense and deterrence—and the laws of war as a whole—may deteriorate or, at the very least, require adaptation.