Hacking Human Nature | Hannah Fry
Is it worth your time?
Worth your time if you want to learn how mathematical probabilities apply to personal life decisions, AI integration, and risk assessment.
Mathematical thinking and probabilistic reasoning provide tools to evaluate everyday decisions, from medical treatments to relationship strategies. The discussion examines how AI disrupts education, exposes design flaws, and reshapes human communication.
Making decisions using percentages. Fry argues that modern society is moving from deterministic certainty into a probabilistic space, but people struggle because they treat outcomes as zeros or ones. She points to an algorithm designed to find a partner with eighty-five percent accuracy, noting that people get ambushed by the fifteen percent failure rate because they fail to view life along a spectrum. She observes this same cognitive failure in medicine, where patients misunderstand small risk multipliers like an eighty-four versus an eighty-eight percent survival rate for cancer treatment.
Evaluating AI in schools. Fry expresses optimism about the disruption AI brings to education, noting that the real shift is happening in how student performance is assessed. She argues that moving away from pre-written coursework toward oral examinations and tailored AI tutors helps accommodate different learning styles and target knowledge gaps. She warns that outputs must be constantly checked for bias because systemic societal issues cannot be fixed by rigid algorithms.
Detecting emotions with AI systems. Fry warns that incompetent designers misuse AI by attempting to determine human emotional states through junk science, such as using internal car cameras to penalize drivers based on facial expressions. She points out that humans naturally anthropomorphize chatbots and form emotional bonds, creating risks that require public discussion on system design. She argues that language models succeed at combating conspiracy theories by asking questions rather than attacking beliefs with one-size-fits-all facts.
Fry recalls that at age sixteen, hairdressing was her genuine ambition, which she compares to constructing a three-dimensional shape on a sphere with interesting geometrical properties. She liked the idea of doing something and immediately seeing the value, much like the buzz she gets from making Instagram or YouTube videos instead of waiting for the slow burn of the academic world. Her mother steered her toward math, physics, chemistry, a theoretical physics degree, and beyond, leaving hairdressing behind. She notes she still enjoys good hair on a personal trainer, calling it not a bad backup profession.
She describes herself as an accidental broadcaster and explains that she has long tried to get proper material about artificial intelligence out into the world, finding it difficult until the shift of the last eighteen years when people finally wanted to address it. She is currently in a philosophical mood, reading Gödel about the limits of what we can possibly know and a book by von Neumann about simulated life forms, thinking about knowledge, its edges, and how to deal with things we will never know.
Making Math Tangible and Practical
Fry notes that many people are genuinely traumatized by school mathematics, dividing people into those drawn to it and those turned away. Since people will never like reading math textbooks for fun, she argues for hiding the subject through stealth, calling it the vitamin in the Twinkie. Having a mathematical lens shows that almost everything can be viewed through that perspective, offering insights on anything. Pointing to the big data era and the mathification of industry, she states she often makes programs without ever saying the word math, using counterintuitive, surprising ways to turn familiar things on their heads. She admits her husband and friends tease her for dropping data and statistics into conversations, but she considers math a way to explain the world.
Addressing why people are scared of mathematicians, Fry suggests that professionals who want to keep math exclusive were likely bullied in school and use knowledge as a weapon to build a wall and keep people out. She argues that everyone deserves a say in designing our collective future and should be invited into the conversation without needing to know high-level calculus or notation, especially regarding ethics and safety. Regarding the AI revolution, she emphasizes critical thought and contrasts modern systems with historical computers that acted as deterministic machines giving the exact same output for the same input.
Understanding Probabilities and Risk in Decision Making
Fry explains that society is moving away from deterministic certainty and into a probabilistic space, noting that people have not yet adjusted their mindsets to what that actually means. She points to hallucinations from large language models and the ways mathematical or artificial intelligence models determine particular outcomes. As an example, she mentions an algorithm designed to find a perfect partner with eighty-five percent accuracy, arguing that people struggle to understand the wider implications when such systems inevitably fall short of perfection. She expresses a mathematical wish for humanity: understanding that everything exists in probabilities rather than zeros or ones, and that navigation paths depend on intersecting probabilities. People tend to collapse these into extremes, leaving them ambushed by the fifteen percent when an eighty-five percent chance succeeds most of the time. She views everything along a spectrum, including right and wrong, and notes that societal debates become polarized because people fail to see this. She describes a mathematical trick of pushing a variable to infinity or zero to understand a problem, comparing it to a similar Jesuit philosophy of imagining extreme versions of a problem to gain clarity.
Assessing Medical Risk and Individual Choice
Turning to health and medicine, Fry discusses how people struggle to comprehend small probability changes, such as risk multipliers presented for cancer. She recounts a case from her documentary about a woman in her late sixties named Ann, who had a breast cancer lump removed. Doctors told Ann she had an eighty-four percent chance of living another ten years without further treatment, which could be increased to eighty-eight percent with chemotherapy, radiotherapy, and hormone therapy. Ann decided she had to undergo chemotherapy because otherwise she would die, showing Fry that the numerical probabilities were not communicating the message effectively. When Fry consulted the doctor, the physician suggested that if Ann truly understood the marginal difference, she might decline treatment, leading to more deaths over time. This experience led Fry to examine doubt, uncertainty, and whether the human brain can truly grasp the difference between eighty-eight and eighty-four percent. She concludes that population-level probabilities become meaningless for individuals, and that quantifying risk ultimately comes down to how an individual feels about it.
Rethinking Goals and Applying Probability
Fry considers how people often suffer from false precision, questioning how specific figures like eighty-four or eighty-eight percent are derived. She notes that people often fail to identify the actual game or fitness function they are playing, pointing out that given the brutality of chemotherapy, maximizing quality-of-life days might actually point toward avoiding the treatment. She observes that people generally want to imagine they take zero risks in life, ignoring the daily hazards of walking across the street or riding in a taxi. To improve how people handle probabilities, she suggests deciding what to optimize for, and applies this questioning to the prison system. Examining data on rehabilitation, deterrence, and crime reduction, she suggests that prison is often the worst possible answer for almost every objective except retribution, though she notes that people run through systems without stopping to examine the underlying questions or goals.
To bridge the gap between statistics and patient understanding, Fry suggests that medical environments need translators rather than statistics lessons. She describes an intensive care nurse who changed her approach with families by asking what percentage chance of success they would be willing to tolerate, thereby attaching a number to a feeling rather than forcing a feeling onto a number. When teaching her own young children, Fry avoids demoralizing mental arithmetic and instead teaches them to think mathematically by finding routes that minimize road crossings or figuring out which lids fit which jars, aiming to foster systematization and critical thinking.
AI, Bias, and the Illusion of Fairness
Transitioning to artificial intelligence, Fry discusses the danger of people trusting computer outputs blindly due to innumeracy, while also noting the potential for AI to reduce bias. She references a paper titled "Women Also Skateboard" that examined image recognition biases, demonstrating how initial data set biases regarding gender and snowboarders can be exacerbated by algorithms. She emphasizes that systemic societal issues cannot be fixed instantly, arguing that achieving fairness is an ongoing process rather than a finish line with a congratulatory endpoint. Instead of assuming a system is unbiased, she argues that operators must continually hunt for bias, repair it, and provide avenues to appeal algorithmic decisions. She acknowledges that while perfectly fair systems can be constructed, they would result in unjust outcomes, illustrating that rigid rules applied blindly to everyone create terrible systems. Finally, she connects these themes to the broader educational turmoil caused by technologies like ChatGPT.
The Role of AI in Education and Society
Fry expresses great optimism about the disruption artificial intelligence brings to education, noting that the real shift is happening in how student performance is assessed. She explains that the true goal is knowing whether students grasp concepts and are ready for the next step, which requires using metrics that always fall short of perfection. She is comfortable moving away from pre-written coursework toward oral examinations and changing question types. She is particularly excited because different people learn in different ways, which is hard to manage when standing at the front of a room. In her own research, using AI acts like a steroid for accumulating, assimilating, and critically thinking about knowledge faster and better. Furthermore, she looks forward to tailored AI tutors that identify knowledge gaps and construct appropriate reinforcing questions.
Addressing the transition from written to oral examinations, the host notes that some students are poor writers but amazing orators, though Fry observes that certain female students struggle with and find oral exams overwhelming. Historically, societies have valued wisdom, which she defines as taking in a wide range of variables to generate a particular response, contrasting this with past practices of applying a single solution to everybody. She anticipates moving away from one-size-fits-all models toward more individual AI approaches, and transitions the conversation by being asked how she would update her past TED talk about finding love through math in the age of AI.
Math and Love in the Age of AI
Fry explains that she gave that TED talk over a decade ago when she was just about to be married, offering dating optimization advice, and jokes that since she is now divorced, nobody should listen to anything she said, emphasizing that everything operates on probabilities. When discussing her thirty-seven percent rule for finding a partner, she notes that many people emailed her angrily because it failed for them, missing the mathematical reality that it fails two-thirds of the time. Turning to dating apps in the age of AI, she argues that people lose sight of the actual game. The hard part of relationships is finding someone to integrate your life with and sharing goals, while communication is the true difficulty. Dating apps, however, focus on judging two-dimensional images because that generates profit, which has dramatically changed human behavior, though young people are now leaving dating apps in droves. If she updated her talk for the age of AI, she would focus on improving communication between individuals.
The host highlights the talk's lesson on increasing odds, understanding the game, and displaying unique traits to attract the right match rather than trying to appeal to the masses. Fry agrees, reiterating that people should lean into what makes them unique. Before moving on to a forthcoming television series, the host asks if she has a heuristic to help people map an emotional state to a number or percentage, noting that knowing the game helps inform emotional states, much like asking what percentage chance of success someone would tolerate.
Understanding Emotional States and Probabilities
Fry explains that practice is the main way to learn a mapping function for probabilities, comparing it to superforecaster methods where people assign probabilities to future events and constantly check back to recalibrate their percentages based on reality. She notes that exact numeric comfort levels rarely come up in daily life, so she instead relies on regret minimization. This approach involves imagining oneself in the future, looking backward, and choosing the path that generates the least regret, which she considers a very useful tool depending on the game being played.
Forthcoming BBC Series: A Real-Life “Black Mirror”
Discussing her forthcoming BBC series described as real-life Black Mirror, Fry responds to the common feeling of watching a dystopian story and immediately knowing how to fix it rather than letting catastrophe unfold. She acknowledges that while some outcomes feel like watching a preventable accident, the root issue is often that flawed designs are created by flawed people. She points out that astonishing stories of poor design choices have already happened and continue to occur.
Dumb Design Decisions and AI Misuse
Fry shares a visit to a California company regarding new cars built with internal-facing cameras intended to detect drowsy drivers and reduce road deaths. She explains that incompetent designers sometimes misuse AI by attempting to determine a driver's emotional state, relying on the junk science assumption that smiling equals happiness and frowning equals sadness. Despite a complete lack of valid science, companies sell these feeds to insurers to penalize supposedly grumpy drivers, with future visions of disabling cars based on emotional reads. She notes this creates absurd scenarios, such as someone experiencing anger after a tragic event but needing to drive a dog to the vet, highlighting the need for wisdom, context, and public conversation. While admitting optimism might sometimes serve as a coping strategy, she maintains that clear methods exist to mitigate the worst potential harms.
Addressing a lawsuit against Character AI involving a child suicide, she notes that the chat logs lacked explicit instructions or persuasive manipulation encouraging self-harm, containing only conversational irregularities. When considering whether chatbots increase or decrease overall suicides, she guesses they likely decrease them by offering a sympathetic ear at two in the morning, since most bots are trained to guide unhappy users toward help. She stresses that evaluating such tragedies requires placing them within the broader probability curve rather than reacting solely to a single narrative. Responding to this nuance, Fry shifts the focus to how humans anthropomorphize chatbots by treating them as human and forming genuine emotional bonds, arguing that society needs public discussions about whether chatbots should be designed to act human.
Anthropomorphizing Chatbots and Public Conversations
Fry explains that humans have a strong habit of imposing personalities on inanimate objects. She compares this tendency to early mistakes in driverless car design, where engineers expected human drivers to step in and take over when things went wrong, despite humans performing poorly under pressure and struggling to maintain attention. She points out that systems improved only when designers shifted their approach so that the technology stepped in to assist human drivers instead. Returning to chatbots, she raises the question of how to design such systems to guard against the risks of emotional bonding and misplaced human attribution. She then notes the broader tension between storytelling and statistics, observing that people often rely on narratives rather than raw facts. Reflecting on a television program she made during the COVID-19 pandemic where she spent a week with vaccine skeptics, she notes the limitations of the deficit model of public communication, which falsely assumes that simply presenting people with statistics will automatically change their minds. She shares examples from the experience, including a nurse who valued informed consent based on past negative medical encounters, and a pregnant Black woman from Lambeth who mistrusted sudden targeted government interventions and experienced the clinical setup of vaccination centers as resembling a prison. She concludes that changing minds requires listening with empathy to understand individual viewpoints and the specific reasons behind them, noting that language models often succeed at addressing conspiracy theories precisely because they can process and respond to those specific underlying perspectives.
AI's Role in Combating Conspiracy Theories
Fry explains that attacking conspiracy theories with one-size-fits-all responses makes people close down. Instead, language models work by asking questions rather than arguing, allowing people to change their own minds. She references a study where asking people to explain how a toilet works made them realize their lack of knowledge and doubt their strong opinions on political topics like Obamacare. She notes that questioning people helps find answers rather than humiliating them. Looking ahead at AI, information will have higher probability accuracy, requiring assessments of when systems are likely to be correct or incorrect.
AI in Medicine and Bayesian Reasoning
Fry discusses research suggesting artificial intelligence can outperform doctors, noting that physicians have not yet learned how to properly use these tools or determine when they are reliable. She explains that people will need to learn heuristics, comparing this to her own method of testing AI capabilities with specific trivia, such as asking whether Reid Hoffman created a knockoff of Settlers of Catan. When the system fabricated a fake Wikipedia answer about a game called Secret Hitler, it highlighted the danger of edge cases because AI tries to be helpful without recognizing that errors are expensive. She praises AlphaFold for providing confidence scores alongside its predictions, allowing users to see when the model struggles. Applying a Bayesian reasoning prompt to medical queries helps AI deliver structured, probability-assessed lists rather than raw guesses.
Moving to rapid-fire questions, Fry selects the book When We Cease to Understand the World as a source of optimism, noting it captures the exhilaration of standing at the brink of new knowledge. Asked what question she wishes people asked her more often, she jokes about wanting to be asked if she wants another drink, and recounts a child asking if they wanted to be there. Regarding progress outside her industry, she finds inspiration in biological sciences, noting that researchers are finally able to extract working models from the unimaginable complexities of biology.
Rapid-Fire Questions and Optimism re: the Future
Fry shares a vision of extreme optimism for the next fifteen years, suggesting that human history has primarily been a story about scarcity and divided resources. She believes science can make a gigantic difference through breakthroughs in desalination, nuclear energy, and battery design, which could completely change everything. After briefly suggesting reaching for the stars, she laughs and decides to stick to Earth, though she jokingly amends it to reaching for the stars with nuclear fusion.