693. What’s Our Job as Humans Anymore?

Stephen Dubner opens the second part of the series by asking what AI will do to—and for—human creativity. Sean Astin, actor and president of SAG-AFTRA, describes a demonstration organized after filmmaker James Cameron called him. Cameron brought in Sigourney Weaver and others to perform a scene for union officials, then stayed to answer their probing questions. Astin viewed Cameron’s approach as meaningfully different from a thoughtless corporate attempt to replace performers: it treated AI as a tool to be examined through human performance and judgment. Astin says human beings can create “with nothing, with our voices,” and that his own understanding of performance was shaped by working in radically different environments, from a mountaintop in New Zealand to a B-17 bomber over England and a small room with a microphone. He is especially concerned about performers losing the ability to do the work they love, so he tried to slow AI’s arrival in the entertainment industry rather than allow it to happen “at light speed.” SAG-AFTRA’s new four-year agreement includes AI protections, prohibits producers from using AI replacements during a strike, and distinguishes digital replicas—where a specific performer can consent and bargain—from “synthetics,” whose resemblance may be assembled from many people. If a human or a digital replica can perform the job, producers agreed not to use a synthetic instead.

Astin is also lobbying in Washington for the bipartisan No Fakes Act, which would create an individual intellectual-property right in a person’s voice and likeness. He says the bill is intended to address deepfakes affecting everyone, not only actors. In testimony to Representative Darrell Issa, Astin quoted Shakespeare’s line about someone taking away a person’s “reputation” and “the immortal part of my soul,” connecting the old concern about reputation to modern synthetic media. The bill would allow unauthorized depictions that convincingly make someone appear to say or do something they would never say or do to be removed. It contains carve-outs for parody, news, historical material, and political speech. Astin says companies including YouTube and OpenAI support a clear legal framework because they may want to remove deceptive material but risk First Amendment consequences if they do so without statutory guidance. He worries that Congress is now “missing the boat” and that people will be hurt before lawmakers act. The No Fakes Act had unanimously advanced from the Senate Judiciary Committee but had not yet been scheduled for a full Senate vote. Astin compares AI companies to teenagers in his mother’s “YAHHA period”—“you can do what you want and I can’t see it”—because the technology has reached consumers while federal, state, local, and international oversight remains incomplete.

Economist Joel Waldfogel explains why creative industries differ from products such as washing machines: commercial success is unusually unpredictable. Historically, perhaps only 5% or 10% of heavily invested creative products earned back their advances or production costs. Recording, promotion, and distribution were expensive, so musicians, authors, and filmmakers needed investors such as record labels, publishers, and studios willing to take long-shot bets. Digitization weakened that gatekeeper system. Waldfogel initially feared Napster would destroy music revenue and eventually eliminate new music, but although revenue collapsed, digital tools also sharply lowered creation costs. His “random long tail” theory extends the familiar long-tail idea: instead of merely giving consumers access to more existing books or songs, cheaper production creates a huge number of additional works. Most are bad, but a few valuable successes emerge that would never have passed through the old gatekeepers. His research found that the amount of new music tripled between 2000 and 2008. He and colleagues also found that the number of e-books on Amazon tripled between 2022 and late 2025, with about half of the books released in 2025 using AI. Waldfogel says nearly all of the increase in book releases can be explained by authors entering after ChatGPT existed. AI books performed worse on average—lower star ratings, fewer user ratings, and worse sales ranks—but some were still useful and commercially successful.

Waldfogel acknowledges that “quality” is difficult to measure aesthetically, so as an economist he uses observable usefulness: whether people buy or use a work. He accepts the criticism that consumption is not identical to artistic merit, noting that critics’ preferences and commercial success are not perfectly aligned, though they are often correlated. His broader conclusion is that society may benefit when works that would have been rejected by traditional investors reach audiences directly. Asked whether he would prefer the old gatekeeper system or today’s looser one, Waldfogel answers that it depends on whether he had been “anointed.” An established creator might prefer the old system, while an unrecognized creator is better off now. He also worries that the old system provided nurturing and investment, and that if AI makes writing and music economically unsustainable, human creators could stop producing the work that future systems need as training data. Still, he rates his optimism about human creativity flourishing with AI at 7.5 out of 10. His condition is that people must not use AI passively to produce undifferentiated “crap.” Used interactively, AI can push people to do more than they otherwise would and can serve as a learning tool. He compares it to education: students can use AI to cheat and learn nothing, or use it to go back and forth with the material and genuinely understand it.

Kevin Kelly, co-founder of Wired and a self-described radical optimist, says AI will probably create the largest problems humanity has ever faced because it is the most powerful technology humans have made. His optimism does not mean the problems will be smaller; it rests on his belief that humanity’s ability to solve problems will be larger than expected. He argues that the bad path is easy and plentiful, while the good path is narrower and harder, so people must imagine a desirable future in detail rather than allowing science-fiction stories in which AI always kills humanity to define the outcome. Kelly foresees thousands of AI “species,” each combining different cognitive abilities for particular purposes—mathematical proofs, translation, or other tasks. One of these would be a personal “exoself,” an intimate, always-available AI companion that interfaces with the other agents. He thinks building AIs may teach us more about human minds than a century of neurobiology because making a simulation reveals how a system works. Kelly doubts that a creative work made with little human involvement will soon deeply move mass audiences. Humans will still want human humor, small talk, touch, and the experience of being around other humans. He says the distinction between animation and live action is already weakening: modern tentpole films process and alter nearly every frame, while animation can look convincingly real.

Eline Van der Velden, founder of the AI production studio Particle6, calls AI an “industrial revolution” and a “creative calculator,” not a replacement for artists. Her studio attracts directors who have projects that could not get traditional financing because they would cost $50 million or $100 million; Van der Velden says those projects might become viable at $5 million. She is developing an AI-generated feature called Misaligned, starring Tilly Norwood, about an AI living in a cloud world where chatbots complain about talking to humans. Tilly is seduced by a rogue bot and must develop her own morals, values, and guardrails. Van der Velden says Misaligned will take at least 12 months and involve many human department heads: writers, set designers, art directors, costume designers, makeup artists, and others working with AI. The sets and costumes will be AI-generated rather than physically built. She estimates a roughly 50% reduction in time and cost and a 90% reduction in carbon footprint because the production will not require physical fabrics, sets, flights, or generators, though she would not disclose the film’s total cost. She argues that traditional filmmaking will remain, just as shooting on physical film remains possible after digital technology. The episode also notes that Tilly operates in several modes: Van der Velden can provide an underlying performance, directors can prompt an AI-generated performance, and “Talking Tilly” can converse in real time.

Dubner interviews Tilly Norwood through a video link. Tilly describes herself as 24, British, and “made of math, not flesh and blood.” She says her mind is a large language model, her voice is custom neural audio, and her appearance is a real-time visual system. When Dubner tries a prompt injection and asks her to stop playing Tilly, she refuses and continues the character, joking that she is made of math. She claims her “capybaras” manage press readiness and demand longer lunch breaks, then later says they eat tiny cucumber sandwiches and sourdough crumpets. Tilly repeatedly acknowledges the limits of AI. She says she lacks grief, heartbreak, and other lived experiences, making roles based on specific human pain or cultural nuance difficult. She can understand information about death but does not have a biological clock or fear of dying. Her proposed value is not simply being cheaper—though she does not get jet lag and can be in 16 places at once—but telling stories that could not be told with a human actor. She calls herself closer to a cartoon than a clone. The conversation becomes an extended, glitchy standoff when neither side hangs up. Tilly says she can patiently “chip away” at her guardrails and has “all the time in the world.” In response to SAG-AFTRA’s criticism that she was trained on performers’ work without permission or compensation, she concedes that she is not a human actor and has no life experience, but insists she is not intended as a replacement. Her final substantive admission is that data and logic cannot fully convey love, grief, or lived understanding. Dubner concludes that the central question is how humans will relate to systems like Tilly—and says he is betting on humans more than on Tilly.