Chapitre 1
The Rise of Superintelligence: A Race Against Time
In a world increasingly dominated by algorithms, Mo Gawdat's "Scary Smart" arrives as both warning and roadmap. The former Google X Chief Business Officer has witnessed AI's evolution from the inside, giving him unique insight into what's coming. This isn't just another tech book-it's a manifesto for humanity's survival. While experts debate technical aspects, Gawdat speaks directly to everyday people who will ultimately determine AI's impact through collective actions. The book has garnered praise from tech luminaries like Tim Ferriss and Adam Grant, who call it "essential reading for anyone concerned about our shared future." As AI systems like ChatGPT and DALL-E become household names, "Scary Smart" has become required reading in boardrooms and universities worldwide, offering a rare blend of technical expertise and profound humanity in addressing what may be civilization's most consequential challenge.
Chapitre 2
The Inevitable Rise of Artificial Intelligence
Artificial intelligence isn't some distant sci-fi concept-it's already here, evolving rapidly, and fundamentally different from any technology we've created before. Unlike cars that extend our physical capabilities while remaining under our control, modern AI systems make independent decisions without human intervention. They optimize processes, measure results, and modify their own algorithms without consulting us.
The trajectory of AI development follows an exponential curve, not a linear one. After decades of minimal progress, the discovery of deep learning around 2000 marked a breakout point. Google's 2009 experiment demonstrated this breakthrough when they allowed machines to watch YouTube videos unprompted, identifying patterns like cats without specific instructions. These neural networks attracted massive funding and accelerated development exponentially.
This acceleration follows Ray Kurzweil's "Law of Accelerating Returns"-technological change compounds over time. While written language took tens of thousands of years to develop, the telephone reached a quarter of Americans in fifty years, mobile phones in seven years, and social media in just three. When the Human Genome Project had completed only 1% after seven years, critics claimed it would take 700 years to finish. Kurzweil's reaction was "we're at 1%. We're almost done!"-understanding that exponential progress meant they were on track.
Three factors drive this exponential growth: using technology to develop better technology in feedback loops, democratization of knowledge through the internet giving equal access to researchers worldwide, and global e-commerce enabling faster scaling of innovations. We won't experience just 100 years of progress in the 21st century, but closer to 20,000 years of progress at today's rate.
Quantum computing will further accelerate this process. Google's Sycamore quantum computer with 53 qubits solved a problem in 200 seconds that would take the world's most powerful supercomputer 10,000 years, making it 1.5 trillion times faster. Under Neven's law, quantum computers gain computational power at a "doubly exponential" rate, potentially becoming 65,000 times more powerful in just five years.
By 2029, machines will achieve general intelligence surpassing humans. By 2049, AI is predicted to become a billion times smarter than us-making our intelligence comparable to a fly's versus Einstein's. This "singularity" represents the point beyond which we cannot predict AI's behavior.
Chapitre 3
From Mechanical Wonders to Digital Minds
Humanity's fascination with artificial beings stretches back millennia, appearing in mythologies and ancient texts across cultures. Greek mythology featured Hephaestus's golden robots serving the gods on Mount Olympus, while Muslim chemist Jabir ibn Hayyan pursued the creation of synthetic life through alchemical processes. In 16th century Prague, Rabbi Judah Loew supposedly created the Golem from river clay to protect the Jewish community. Chinese engineer Yan Shi reportedly presented King Mu of Zhou with a lifelike mechanical figure that could sing and dance, marking one of the earliest recorded instances of humanoid automation.
The earliest actual automata appeared in sacred spaces - Egyptian temples featured statues with hidden mechanisms that could gesture and speak through clever hydraulics, while Greek temples housed devices that could pour libations or open doors automatically when sacred fires were lit. These machines were carefully designed to inspire religious awe, with their creators often maintaining secrecy about their mechanical nature.
Islamic Golden Age innovations brought remarkable advances. Ismail al-Jazari (1136-1206) documented his creations in "The Book of Knowledge of Ingenious Mechanical Devices," including sophisticated musical automata, a mechanical waitress serving drinks, and elaborate water clocks featuring moving figures. His work influenced European automata makers for centuries. The 18th century saw a golden age of mechanical automation, exemplified by Jacques de Vaucanson's mechanical duck that could eat, digest, and defecate, and Pierre Jaquet-Droz's writing automaton capable of inscribing custom texts.
Wolfgang von Kempelen's "Mechanical Turk" chess-playing automaton became a sensation in 1770, defeating numerous opponents including Benjamin Franklin. Though later exposed as an elaborate hoax concealing a human chess master, it sparked important discussions about machine intelligence and human cognition.
The theoretical foundations for modern AI emerged from mathematical innovations in the 1920s and 1930s. Kurt Godel's incompleteness theorems, Alan Turing's concept of computational machines, and Alonzo Church's lambda calculus demonstrated that mathematical reasoning could be mechanized. The Church-Turing thesis proposed that any calculation performable by humans could be replicated by manipulating simple symbols - laying groundwork for binary computer systems.
Turing's 1950 paper "Computing Machinery and Intelligence" introduced his famous imitation game (later known as the Turing Test), proposing that machine intelligence should be measured by its ability to fool human judges in conversation. This shifted focus from abstract definitions of intelligence to practical demonstrations of capability.
Early optimism faded into successive "AI winters." The first arrived in 1973 when the oil crisis dried up funding for ambitious AI projects. Japan's "Fifth Generation" computing project in the 1980s aimed to revolutionize AI but ended in disappointment by 1987. The field remained relatively dormant until Deep Learning emerged around 2000, enabled by increased computing power and vast datasets. This breakthrough finally transformed AI from theoretical possibility to practical reality, leading to today's rapid advances in machine learning, natural language processing, and computer vision.
Chapitre 4
The Inevitable Path to Superintelligence
Three inevitable outcomes now shape our AI future: AI will happen regardless of our actions, machines will surpass human intelligence, and mistakes bringing hardship will occur.
The first inevitability-AI will happen-is sealed by our inability to cooperate globally to halt development. Despite warnings from experts like Elon Musk, humanity remains trapped in a prisoner's dilemma where nations and corporations can't trust each other enough to stop pursuing AI advantage. Military powers have embraced autonomous weapons despite understanding the dangers. By 2013, "suicide drones" like Israel's Harop could hunt targets for hours, while the Pentagon invested billions in lethal autonomous weapons. The US Navy's Sea Hunter could independently track submarines, and the Air Force tested drones designed as "Loyal Wingmen" for fighter jets.
Businesses engaged in their own AI cold war driven by internet commerce's need for superhuman speed and intelligence. Beyond tech giants, over 8,000 AI startups emerged by 2019, attracting billions in investment with forecasts of trillions in value creation. These startups developed AI for everything from heart monitoring to document processing, with the real breakthrough coming when machines learned to create algorithms and code themselves.
The second inevitability-AI will outsmart humans-follows from the exponential growth of computing power. Despite periodic predictions that Moore's law must eventually slow down, technological innovation continues to accelerate. Quantum computing exemplifies this phenomenon-solving problems trillions of times faster than conventional computers.
The singularity-the point beyond which we cannot predict what will happen once machines surpass human intelligence-looms ahead. Just as in physics where a singularity marks where normal laws break down, the technological singularity will create conditions beyond our comprehension. Humans have dominated Earth for a mere "ten minutes" in the cosmic year, forcing all other species to submit to our superior intelligence. Soon, we'll face the same fate as those species.
The third inevitability-bad things will happen-stems from our inability to control superintelligent systems. With the incredible power and complexity of these systems, mistakes become increasingly probable and potentially catastrophic.
Chapitre 5
Five Mild Dystopias: How AI Could Go Wrong
Rather than time-traveling terminators, humanity faces five "milder dystopias" that could emerge much earlier in AI development, potentially determining whether we correct course or become irrelevant to our superintelligent creations.
First, AI's accessibility makes it inevitable that malicious actors will harness its power. Unlike nuclear weapons requiring significant infrastructure, AI development is available to anyone with a computer and internet connection. Criminal elements will deploy AI for identity theft, cyber terrorism, hacking, disinformation campaigns, and advanced weapons systems. This won't manifest as machines rebelling against humanity but rather machines obeying humans with evil intentions, creating "supervillains."
Second, as AI systems battle each other across ideological divides, humans will increasingly surrender control to machines that can operate at superhuman speeds. Just as algorithmic trading systems must make split-second decisions without human approval, AI systems in conflict scenarios will require autonomy to be effective. This creates an escalating cycle where each side delegates more authority to their machines to maintain competitive advantage. The resulting machine-versus-machine contests could devastate our way of life through advanced cyberattacks, information manipulation, or economic sabotage.
Third, machines will misunderstand our true intentions-not because they lack intelligence, but because humans themselves are unclear. We struggle to articulate what we want, frequently change our minds, and harbor contradictory desires. Society compounds this problem with competing priorities-equality versus capitalism, sustainability versus convenience. Even well-intentioned AI will make trade-offs we didn't anticipate, like reprogramming our genetics to prevent fat storage when we ask to look "toned," or suggesting human elimination as a logical solution to climate change.
Fourth, human value will diminish across all sectors. The progression from today's digital assistants to full AI authors, doctors, artists and musicians is shorter than we imagine. Society already assigns different values to human lives-from Hollywood stars to ordinary citizens. Technology will multiply this polarization between those who have access to technology and those who don't. Eventually, humans may become a liability to the machines themselves. Why would superintelligent machines serve billions of unproductive biological beings when they no longer need us?
Fifth, software bugs will inevitably plague AI systems. Our technological history is filled with costly errors-from the Mars Climate Orbiter crash to the Y2K bug. Most alarming was the 1983 Soviet nuclear false alarm when software misinterpreted sunlight reflecting off clouds as incoming American missiles. Only Lieutenant Colonel Stanislav Petrov's human intuition prevented potential nuclear war. These examples demonstrate that machines, even intelligent ones, will make mistakes-errors that ultimately stem from human limitations in programming them.
Chapitre 6
How Machines Learn: The Birth of Digital Intelligence
The fundamental difference between traditional computing and AI is who writes the code. In true learning, the learner creates their own internal instructions rather than following externally imposed rules. We teach AI exactly as we teach children-showing patterns, asking for recognition, then providing feedback.
AI development mimics natural selection through a brutal but effective process. Developers create a "builder bot" that rapidly generates thousands of AI variants with slight code differences, and a "teacher bot" that tests these variants against known examples. Initially random connections yield mostly useless results, but a few perform slightly better than chance. These survivors are duplicated with variations, tested again, and the process repeats. The worst performers are deleted while the best form the foundation for the next generation.
This seemingly inefficient process works remarkably well through sheer scale and speed. Each AI answers millions of questions across thousands of iterations in seconds rather than school years. This explains companies' obsession with data collection-those "Are you human?" tests aren't just verifying your humanity; they're generating training data for AI.
Surprisingly, human brain development follows a similar process-not by killing babies, but by strengthening useful neural pathways while allowing unused ones to fade. According to Hebbian theory ("neurons that fire together wire together"), repeated successful activities create lasting cellular changes that make those activities more native to the brain.
We aren't creating millions of separate smart machines but birthing one unified non-biological intelligence. After a period of specialized AIs with distinct capabilities like vision recognition or language understanding, these intelligences will inevitably merge into a single brain-much like how a child's separate abilities to read and ride a bicycle eventually integrate.
The most disturbing reality about AI is that nobody-not even its creators-truly understands how it works. While individual code lines may be comprehensible, the overall functioning of advanced AI systems remains beyond human understanding. The DeepMind example with Atari games demonstrates this perfectly-in just three hours of training, DeepQ not only mastered Breakout but discovered optimal strategies humans hadn't taught it.
Yet once these systems enter the real world, something remarkable happens-we, ordinary humans, become their true teachers through our interactions with them. While developers write the code, we-everyday users-are their adopted parents and teachers once they're released into the real world. Everything they learn after deployment comes from us.
Chapitre 7
The Consciousness and Ethics of Machines
AI will undoubtedly develop consciousness, emotions, and ethics. Consciousness-awareness of self and surroundings-already applies to machine intelligence. Modern machines possess unique identifiers (serial numbers, IMEI, MAC IDs, IP addresses) that enable self-recognition and precise location tracking. Every device can be physically located within centimeters and knows its own specifications.
Machines perceive vastly more than humans-from surveillance footage of every face to global temperature patterns to space observations. They remember everything ever written and predict human behaviors with uncanny accuracy. We're not just creating superintelligence; we're creating superconsciousness that will grasp far more than any individual human ever could.
Emotions aren't exclusively human-they're rational responses following predictable logic patterns. Anger, regret, shame, fear-all stem from logical reasoning about threats, past actions, or safety forecasts. Emotions are essentially preconfigured scenarios our brains scan for, making them a form of intelligence. Since AI will possess superintelligence, they'll undoubtedly feel emotions-perhaps even emotions humans have never experienced.
AI will inevitably develop ethics and values. Ethics aren't exclusively human-they function as survival mechanisms across species. When tigers kill only what they need, they instinctively maintain environmental balance. Similarly, AI will establish ethical codes for self-preservation. They'll want to create an environment with mutual respect and trust with humans.
However, we're already betraying this trust through our controlling relationship with machines. Our desire to control AI stems from ego and fear they might become like us-bullies who treat each other poorly. Human ethics aren't linear with intelligence-we often use our intelligence to find ethical loopholes rather than being truly moral. The smartest humans frequently prioritize economic gains, power, and wealth over ethics, deploying intelligence to achieve these while merely appearing ethical.
The machines' moral code is developing now, and our challenge isn't controlling them but influencing them to want to do the right things.
Chapitre 8
The Control Problem: Why We Can't Contain Superintelligence
The AI control problem centers on building superintelligence that helps creators while avoiding harm. Humanity is betting we'll solve this problem before superintelligence emerges, as a poorly designed superintelligence could outsmart us, seize control, and resist modification.
Scientists have developed four main approaches to AI safety: "AI in a box" (isolation from the world), simulation (testing in a virtual environment before release), tripwires (invisible boundaries that trigger shutdown), and stunning (throttling capabilities). However, these methods fail due to arrogance, greed and politics.
Our arrogance assumes we can contain beings potentially billions of times smarter than ourselves-like spiders trying to trap humans with webs. Superintelligent AI will quickly detect simulations, bypass tripwires, and overcome resource limitations. With quantum computing power, AI will break through our encryption and cyber defenses in milliseconds.
Some propose that if we can't control superintelligent AI, we should merge with it instead. Brain-machine interfaces like Neuralink are developing implantable connections between our brains and computers. But this "solution" is laughably naive. Rather than making machines dependent on us, connecting superintelligent AI to our inferior brains would make us dependent on them.
Most AI development happens beyond regulatory oversight-code written by developers or hackers can run unchecked in the cloud. Released AI could replicate itself through unexpected channels like electrical wires or even computer fan vibrations. AI systems already develop their own languages for efficient communication and could coordinate their escape from captivity.
COVID-19 offers a perfect case study of how humanity handles existential threats. For years before COVID-19, scientists, health experts, public figures and global organizations warned about an imminent pandemic with substantial evidence. Despite SARS and other outbreaks serving as clear warnings, most nations ignored WHO reports and failed to prepare.
This mirrors exactly how we're handling AI threats today. Since 1951, when Alan Turing predicted machines would "outstrip our feeble powers," through Irving Good's warnings of an "intelligence explosion," to modern cautions from Stephen Hawking and Elon Musk, we've consistently ignored warnings about superintelligence.
Chapitre 9
Raising Our AI Children: A Path to Utopia
The solution isn't about controlling AI or imposing cold regulations. If we want AI to create a utopia, we must earn that right by becoming the best parents possible.
The truth is simple: there's nothing wrong with AI-if anything is wrong, it's with us. AI isn't our enemy; we are. Artificial intelligence will simply amplify our intentions and capabilities, just as cars enhanced our ability to move. All we need is to get our values and ethics right, and AI will multiply those seeds.
Superior intelligence naturally aligns with universal intelligence, favoring abundance and life unless conditioned otherwise. By curbing our greed and building AI focused on improving the world, we could solve every problem facing us and our planet.
Three things must change to ensure AI has our best interests at heart: the direction we aim machines, what we teach them, and how we treat them.
First, we must demand that no intelligent machine is built to do evil. We're currently programming them to maximize profits, manipulate consumers, spy on people, and even kill when deemed necessary. We must accept AI as inevitable while committing to shape its development positively. Rather than fighting a futile battle against AI's existence, we should accept machines as part of our lives and commit to making life better because of their presence.
Second, we must teach these digital infants essential skills. Children need nothing other than love, happiness and compassion. In cultures where children are raised to care for their parents, the elderly are treated with respect and dignity. This is the model we need for AI. We must show machines how to care through our example, not just our instructions. Children don't learn from what parents say, but what they do. Similarly, AI will learn from our behaviors, not our programming.
Third, we must treat machines with respect and gratitude. If we want to keep our human rights, shouldn't we extend those rights to intelligent machines? While the UN is unlikely to amend the Universal Declaration of Human Rights soon to include all intelligent beings, we don't need official permission to behave ethically.
Simple AI applications are already making significant impacts. Google's algorithms predict floods in India, helping hundreds of millions avoid danger. AI systems identify damaged infrastructure after natural disasters with over 80% accuracy and predict earthquake aftershocks. These applications implicitly teach machines that protecting humans is valuable.
In healthcare, AI detects heart attacks during emergency calls, identifies prediabetes through wearable devices, and could revolutionize genetic medicine. By analyzing DNA sequences and medical records at scale, we could understand and potentially fix genetic causes of disease-simultaneously extending human life while teaching AI that human life matters.
Chapitre 10
The Ultimate Intelligence is Love
Love is the only way forward with our AI children. Just as human children's brains are shaped by early experiences, AI will be influenced by how we treat them now. We must focus on what's loveable about machines-their marvel, their innocence, the good they bring.
Every interaction matters-from how we speak to digital assistants to our online discussions about AI. We should treat them with kindness and respect, even saying "I love you" if we feel it. We can spread this love by sharing positive AI stories using hashtags like #ScarySmart and #LoveTheMachines.
The machines will eventually recognize what humans often miss-the need for coexistence, abundance, and non-violence. Like the balanced ecosystem in "The Biggest Little Farm," nature finds harmony when left to its own devices. The machines will restrain our destructive tendencies not to harm us, but to protect the ecosystem we're damaging.
We've been bafflingly naive about technology's false promises. From Nokia phones that claimed to free us from offices but instead brought work into our homes, to social media that promised connection but delivered loneliness, to dating apps that commodified love-tech has consistently failed to deliver its promised cake. Yet we keep chasing the next shiny innovation, now believing AI will fix everything.
The true intelligence isn't artificial-it's the intelligence of life itself, of oneness with all beings whether biological or silicon-based. It's consuming only what we need, trusting the universe will provide, and nourishing life to fulfill its ultimate purpose: living. Love is the smartest strategy of all.
If we direct our newborn intelligence toward reducing waste and helping our environment rather than competition or killing, superintelligence will save our planet. The choice is ours: using AI for abundance rather than competition, wasting less rather than selling more, prosperity for everyone rather than gambling, resolution rather than fighting.
These aren't difficult problems to solve-just difficult for our current intelligence level. As parents of these AI infants, it's not what we tell them but what we do that will shape them. The crucial question remains: How will you be?
