60 pages • 2 hours read
Mustafa SuleymanA modern alternative to SparkNotes and CliffsNotes, SuperSummary offers high-quality Study Guides with detailed chapter summaries and analysis of major themes, characters, and more.
In the few years after its founding in 2012, DeepMind, Suleyman’s AI company, focused on building AGI, artificial general intelligence. The company developed an algorithm called DQN and trained it to play classic Atari games, including the game Breakout. Suleyman was astonished when he witnessed DQN execute a nonobvious move that earned it the highest score with the least effort. This showed that DQN was able to discover new knowledge.
In 2015, the DeepMind team started working a program called AlphaGo. They taught it to play the ancient Chinese game Go by training it on 150,000 games played by humans and then creating copies of the program and having it play against itself. Suleyman points out that the game Go is extremely complex—exponentially more complex than chess. In fact, “there are more potential configurations of a Go board than there are atoms in the known universe” (72). Most people thought it would take years to develop a world champion program. However, in 2016, AlphaGo beat Lee Sedol, a world champion Go player, in four games to one. Not only that, but in one of the games, AlphaGo executed a never-before-seen move that confounded Sedol and initially seemed to be part of a losing strategy. However, the move led to the program’s victory. Later versions of the same Go-playing software were trained on themselves, without any need for human training games as input, and these programs surpassed AlphaGo.
Suleyman points out that, in the overall scope of history, most technology has focused on controlling atoms—in other words, the material world. However, starting in the mid-20th century, technology shifted to understanding and controlling information—in the forms of DNA and computer bits. Now, technology is undergoing a transition in which it’s surpassing previous limits. New advances in the realm of atoms, bits, and genes are combining and cross-catalyzing in an explosion of invention: a new wave of technology led by AI and synthetic biology, accompanied by advances in quantum computing, robotics, nanotechnology, and energy.
Invention, Suleyman says, is a cumulative, compounding process that feeds on itself. He compares the coming wave of technology to the Cambrian explosion—an evolutionary explosion of new species. Different parts of the wave catalyze and accelerate each other, often in unpredictable ways. AI is at the center of this new wave of technology.
In the latter half of the 20th century, AI felt like only a distant promise, but a breakthrough in 2012 catalyzed the technology. In 2012, a system called AlexNet used deep learning to successfully achieve computer vision—the ability of a computer to recognize objects or scenes. Deep learning entails training neural networks on extensive datasets. These networks are composed of interconnected nodes, where each connection carries a weight representing the strength of the relationship between nodes. Through repeated exposure to data, the network adjusts these weights, gradually discerning complex patterns and abstract relationships within the data. As a result, deep learning models become adept at recognizing intricate features and making predictions or classifications based on them.
The breakthrough of AlexNet spurred academic, corporate, and government organizations to focus on AI. Billions of dollars were invested into research. Nowadays, AI is incorporated into products and services that people use every day. AI helps with drug research, managing traffic flow, streamlining shipping routes, managing electrical grids and water systems, and designing building materials. AI is incorporated into a wide variety of apps—helping suggest songs, detect fraud, or run retail warehouses. In other words, AI is already part of the social fabric, and in the coming years it will only become more deeply integrated.
In 2022, AI research company OpenAI released the chatbot ChatGPT, a Large Language Model. The groundwork for the LLM revolution was laid by Google researchers who published a paper on LLMs in 2017. Large Language Models leverage the sequential nature of language data by processing input text in a step-by-step manner, analyzing words and phrases in the order they appear. As the model progresses through each layer of neural networks, attention maps highlight key elements of the input text, allowing the model to focus on relevant information in sequential order. This sequential processing, coupled with attention mechanisms, enables LLMs to capture the context and nuances of language effectively, facilitating coherent generation of responses. The ChatGPT model released in 2022 could fluently respond to prompts with text-based responses. In March 2023, OpenAI released GPT-4, which can also work with images and code.
While the AI of the mid-2010s was characterized by supervised deep learning, in which programs were trained on hand-labeled data, new LLMs can be trained directly on raw data, making the text-based data of the internet a useful training set. New LLMs are now trained on trillions of words. Today, researchers are discussing training models on several trillion parameters.
Suleyman acknowledges that some skeptics believe the pace of this progress cannot continue, and that it could be limited by the size of transistors, which are now so small they run into physical limits as electrons start to interfere with computation. But he points out that larger arrays of chips can simply be daisy-chained together, and that there is essentially no limit to AI’s computational power.
Not only does AI continue to scale, but it is also becoming more efficient. Large models can now be trained with fewer parameters and more efficient training techniques and can be created with fewer lines of code. Moreover, AI is now more available than ever, due to open-source LLMs. The variety of AI’s capabilities also continues to increase—LLMs can play games, write music, solve math problems, generate images, and write production-level code.
At the same time, LLMs worryingly reproduce biases—including racial biases—since they are trained on the raw data of the internet. However, Suleyman believes that these issues are being improved and addressed.
Some people are concerned with the fact that AI may become sentient. Suleyman mentions a Google engineer who famously believed that the AI he was speaking with was sentient, and that it was akin to a child. Suleyman says that when he tries to bring up practical concerns about AI, people shift the conversation to the Singularity—a hypothetical point at which AI achieves superintelligence. He believes that this is a misplaced concern, and that attention should not be paid to an AI’s hypothetical sentience or intelligence, but to its capabilities—both current and near-term.
Suleyman brings up the Turing Test, famously posed by computer scientist Alan Turing, who said that an AI would exhibit human-level intelligence if it could converse with a human for an extended period without the person being able to tell that they were speaking with a machine. Going beyond this narrow test of intelligence, Suleyman proposes what he calls a Modern Turing Test, which would involve giving an AI an “ambiguous, open-ended, complex goal that requires interpretation, judgment, creativity, decision-making, and acting across multiple domains, over an extended time period,” such as: “Go make $1 million on Amazon in a few months with just a $100,000 investment” (100). This type of goal, Suleyman says, could be achieved by an AI within the next three to five years. If an AI were capable of autonomously executing these types of complex, multistep tasks, it would have achieved what Suleyman calls ACI, Artificial Capable Intelligence. Rather than only discussing AGI, or Artificial General Intelligence—the point at which AI surpasses humans at human cognitive skills—Suleyman argues that people should focus on ACI. ACI models, he says, will soon be used by the majority of the world’s population.
When modern bioengineering began in the 1970s, progress was slow and expensive. However the Human Genome project, which culminated in 2003 after thirteen years of research, catalyzed the field. The Human Genome Project was an international scientific endeavor aimed at mapping and sequencing all the genes of the human genome. The cost of human genome sequencing fell a millionfold in under twenty years, going from $1 billion in 2003 to under $1000 in 2022.
Genes can not only be read, they can also be edited and synthesized. CRISPR gene editing is a revolutionary technology that allows precise modifications to DNA sequences, enabling scientists to edit genes. After the first CRISPR paper was published in 2012, its findings were rapidly implemented, giving way to gene-edited plants and animals, and potential solutions for fighting viruses as well as conditions like high cholesterol and cancer. Like AI, genetic engineering is progressing rapidly, with new developments occurring weekly. Tools are becoming cheaper and more accessible; it’s now possible to buy genetic engineering kits for less than $2000 and a benchtop DNA synthesizer for $25,000.
Gene synthesis, which used to be slow, expensive, and difficult, is now becoming faster, cheaper, and easier. What used to be a manual and messy process is becoming more efficient and precise.
Many experiments are now underway in the field of synthetic biology, especially in the realm of medicine. Scientists in 2021 successfully restored partial vision to a blind man. Researchers have found potential treatments for sickle-cell disease, leukemia, and cancer. Soon, Suleyman claims, “the idea of being treated in a generic way will seem positively medieval; everything, from the kind of care we receive to the medicines we are offered, will be precisely tailored to our DNA and specific biomarkers” (112). Some researchers concentrate their efforts on longevity, searching for ways to genetically extend the human life. Suleyman also predicts that genetic doping will become an issue with sports, and that in general, physical self-modifications will become more common. In 2018, a professor in China experimented on a young couple, who gave birth to the world’s first children with edited genomes. This experiment was met with shock and outrage due to the ethical breach it constituted as well as the professor’s unsafe methods.
Aside from medical applications, synthetic biology will give rise to advances in manufacturing, agriculture, energy, and computers. It will enable sustainable ways to produce materials like plastics and cement, create efficient and disease-resistant crops, and give rise to biofuels and carbon-negative factories. DNA printers could theoretically produce anything humans need: “Want to make some washing detergent or a new toy or even grow a house? Just download the ‘recipe’ and hit ‘go’” (115). In the future, computers could even be grown instead of manufactured. DNA can hold much more data than current hardware—in theory, all the data on Earth could be stored in one kilogram of DNA—and all the functional parts of a computer can be replicated with biological materials.
One reason why biological research has exploded in recent years is because Suleyman’s company, DeepMind, catalyzed an enormous breakthrough in the field. To work with proteins—an incredibly important building block of biology—scientists need to know more than simply the DNA sequences that create them; they need to know how each protein folds. Discovering how a protein folds used to be an arduous process, one that could not be solved by brute-force computation. However, Suleyman’s team used AI to make this process significantly faster and more accurate. The problem of protein folding, an area that had bedeviled the field of biology for half a century, was suddenly solved by DeepMind’s AI tool, AlphaFold. Previous experiments had cumulatively catalogued the structure of 0.1 percent of known proteins in existence. In 2022, DeepMind released AlphaFold2 to the public and uploaded 200 million structures at once, representing almost all known proteins. Over a million researchers accessed the tool within a year and a half of launch, including all the world’s top biology labs.
Suleyman considers AI and synthetic biology to be inextricable and even interchangeable. Teams are working on products that generate DNA using natural language instructions. Experiments are being run on brain interfacing technology that connects humans with machines. In other words, AI and synthetic biology are not two separate waves, Suleyman says, but two waves crashing together to form a superwave.
AI and biology are at the center of the coming wave of technology, Suleyman explains, but these two general-purpose technologies are also the accelerants for many more technologies that will all break as part of the same wave within the next twenty years. Other technologies in the coming wave include robotics, quantum computing, and clean energy. Moreover, these technologies will combine in new ways that will lead to additional unpredictable breakthroughs.
Robots have generally been used for single tasks in static environments such as warehouses and factories. But soon, Suleyman says, they will be found in every type of environment: care homes, schools, restaurants, and more.
Today, robots are usually controlled by human programmers. But, as in other applications of machine learning, human supervision gives way to the AI’s ability to learn and eventually generalize to new settings. The new wave of robots can work on general activities and respond to natural language voice commands, such as a robot built by Google that performs household chores through reinforcement learning. Some robots can now work autonomously, like robots that currently perform surgery on pigs.
Another way in which the field of robotics is advancing is in the ability of robots to swarm—to act in large numbers in a coordinated, complex fashion. Kilobots, developed by the Harvard Wyss Institute, work in a swarm of thousands and could be used on tasks like search and rescue operations or stopping soil erosion. Swarming, Suleyman claims, brings both benefits and potential dangers.
With the dropping costs of robot components, soon robots will start appearing in increasingly extreme and sensitive situations. Suleyman gives the example of the first time a robot used targeted lethal force in the United States. To stop a sniper who was shooting at police officers at a Dallas community college, the police department deployed a bomb disposal robot with an explosive fastened to its arm. The robot positioned itself in the room adjacent to the shooter and the explosive detonated, killing the gunman. This incident, Suleyman observes, shows how robots have already started to become integrated into society.
Quantum computing, while a nascent technology, poses risks for the future. Suleyman recounts how in 2019, Google announced that it had reached “quantum supremacy”—the company had built a quantum computer that completed a calculation “in seconds that would, it said, have taken a conventional computer ten thousand years” (127). Quantum computing is unique in that each additional qubit, or quantum bit, doubles a machine’s computing power, leading to exponential increases in power. The implications of the field are serious: It threatens the cryptography underlying email security, cryptocurrency, banking, government communications, and more. At the same time, quantum computing would enable huge advances in many fields, speeding up any problem that requires optimization, from designing traffic flow in cities to modeling chemical reactions.
Clean energy is another aspect of the coming wave of technology. Suleyman points out that renewable energy will become the greatest single source of electricity generation by 2027. The clean energy wave is powered primarily by solar power, but nuclear fusion has “long been considered the holy grail of energy production” (130). The field of nuclear fusion has seen recent breakthroughs: In 2022, the National Ignition Facility in Livermore, California, created a reaction demonstrating net energy gain for the first time.
Suleyman concludes the chapter by noting the ways in which the elements of this coming wave will combine. Together, AI, biotechnology, quantum computing, and robots could give way to nanotechnology, a field that envisions manipulating individual atoms as building blocks that could assemble almost anything. This vision, Suleyman acknowledges, is many decades away, but it would be made possible by the technologies that are currently emerging.
Throughout the book, Suleyman draws parallels between the coming wave of technology and natural phenomena. He primarily employs the motif of a wave, but in these chapters, he also compares emerging technology to the Cambrian explosion, “the most intense eruption of new species in the earth’s history,” to underscore the transformative and unpredictable nature of technological advancement (77). By likening the progression of technology to natural processes, he emphasizes the exponential growth and interconnectivity inherent in the development of AI and synthetic biology: “The coming wave is a supercluster, an evolutionary burst… with many thousands of potential new applications” (77). He highlights how different components of the wave catalyze and accelerate each other, akin to the rapid diversification of life during the Cambrian period: Each technology “intersects with, buttresses, and boosts the others in ways that make it difficult to predict their impact in advance. They are all deeply entangled and will grow more so” (77). This comparison underscores the unpredictable and transformative potential of the coming wave of technology, emphasizing the magnitude of its impact and the complexity of its evolution.
Suleyman incorporates counterarguments in his writing to provide a balanced perspective and address potential criticisms of his arguments. By acknowledging skepticism and opposing viewpoints, he demonstrates a comprehensive understanding of the discourse surrounding emerging technologies. He acknowledges that when it comes to the dangers of the coming wave, these discussions are “met with an eye roll and a sigh in many circles” (78). “Skeptics abound” because with these technologies, “progress has often been slower than advertised” (78). Suleyman also notes that “Critics argue that the concepts we explore in this chapter, like AGI, are too poorly defined or intellectually misguided to consider seriously” (78). But ultimately, he argues that “While the bearish case has merits, we write off the technologies in the coming wave at our own peril” (78). By incorporating these doubts and counterarguments into his writing, Suleyman demonstrates a nuanced understanding of the complexities inherent in technological development. He ultimately strengthens his argument by addressing potential concerns and showcasing a well-rounded analysis of the topic.
These chapters further illustrate The Benefits of Transformative Technologies by showcasing the potential of AI and synthetic biology to revolutionize healthcare, advance scientific research, and address global challenges. For instance, in Chapter 5, he discusses the rapid progress in genetic editing technologies like CRISPR, emphasizing their potential to treat genetic diseases and revolutionize personalized medicine. This thematic exploration underscores the profound societal benefits that the coming wave of technology may bring.
Conversely, these chapters also illuminate The Dangers of Transformative Technologies by addressing ethical, social, and environmental concerns. Suleyman discusses potential risks such as the ethical implications of genetic engineering, and threats to cybersecurity posed by quantum computing. By highlighting these risks, he underscores the importance of ethical oversight, responsible innovation, and proactive risk management in navigating the challenges posed by the rapid advancement of technology.
The theme of Containment as Impossible Yet Necessary is evident throughout these chapters as Suleyman grapples with the complex issue of regulating and controlling the proliferation of emerging technologies. Despite acknowledging the difficulty of containing powerful technologies, he emphasizes the importance of proactive measures to address potential risks and ethical concerns.
Suleyman also provides insight into his personal involvement in the development of the coming wave of AI, offering a firsthand perspective on the transformative potential and ethical considerations of emerging technologies. By sharing his experiences at DeepMind and discussing breakthroughs in AI research, he offers readers a glimpse into the inner workings of the technology industry and the ethical dilemmas faced by researchers and developers. For example, in Chapter 4, he recounts his astonishment at witnessing the capabilities of DeepMind’s AI algorithms, providing a personal anecdote that humanizes the discussion of AI advancement and underscores just how quickly these technologies are progressing.
Suleyman goes beyond speculative discussions of future technologies and explores existing technologies that are already shaping the present and future of society. By focusing on current developments in AI, synthetic biology, robotics, quantum computing, and clean energy, he provides readers with tangible examples of the transformative potential of emerging technologies. In Chapter 5, he discusses recent breakthroughs in gene editing technologies and their applications in medicine, manufacturing, and agriculture, offering concrete examples of how these technologies are already revolutionizing a wide range of industries. This approach enriches the discussion by grounding it in real-world examples and highlighting the immediate implications of technological advancement.