The Machines Are Taking Our Jobs - Thank God? Emad Mostaque’s Guide to the next 1000 Days

Episode Summary

Executive Summary: Ahmad Mustak argues that AI could make intelligence and many forms of production abundant while an economy built around scarcity, wages, and GDP registers displacement and inequality as distress. He proposes measuring intelligence, networks, and resilience alongside material output, and building collectively governed, open AI services and new financial mechanisms to support human flourishing. The hosts probe his economic theory, evidence, risks, and proposed transition.

Main Topics: Intelligence theory and economic foundations: Mustak grounds a proposed new economics in complex adaptive systems that persist by minimizing prediction error, model complexity, and the cost of updating. He draws analogies between these dynamics and generative AI, arguing that traditional economics rests on assumptions that may no longer hold when AI becomes a major producer. AI abundance, labor displacement, and the abundance trap: AI could provide low-cost expertise and create digital goods at sharply reduced costs, but workers may lose income and purpose as machines take over tasks. Mustak calls the resulting mismatch the abundance trap: society may experience economic hardship even as useful intelligence and services become more plentiful. Limits of GDP and a broader measure of economic health: The speakers question whether GDP captures the value of cheap or freely available AI services. Mustak proposes tracking four capitals—material, intelligence, network, and diversity—and emphasizes flows, openness, and resilience as conditions for healthy living systems. Human purpose, care, and social connection: Mustak argues that human relationships, meaning, and community matter even if AI can perform some caring or mentoring tasks well. The challenge is to create a social contract that lets people pursue these activities without relying on scarce jobs for survival, identity, and structure. Possible AI futures and concentration of power: The conversation contrasts digital feudalism, where a few firms control critical AI, with fragmentation into competing national systems, and a preferred path of symbiosis. Mustak supports open, decentralized, collectively governed AI for high-stakes areas such as health, education, finance, and government. Foundation Coin and universal AI access: Mustak describes a proposed digital asset whose compute would fund open models, datasets, and public-interest supercomputing, including for health and education. He also explores a dual-currency idea and income linked to being human, while acknowledging that key design details remain unresolved. Timing, governance, and AI safety: Mustak says the coming years are important for establishing ownership, access, and governance defaults before AI systems become more entrenched. He argues that collective incentives, diverse and culturally grounded datasets, and distributed systems could support safer, more human-aligned AI than profit maximization alone.

Key Arguments: Mustak's intelligence theory treats successful adaptive agents as systems that reduce the mismatch between their internal models and reality while managing the costs of computation and learning; he argues that similar dynamics can inform economics. Even without fully general AI, capable agents and inexpensive models could automate substantial knowledge work. Mustak emphasizes reliable execution of ordinary tasks—the 'cooks'—as potentially more disruptive to employment than rare, novel breakthroughs. AI-created abundance may not translate into broadly shared prosperity: machines do not need wages, housing, or food, while existing systems distribute purchasing power and social status largely through employment. GDP and scarcity-based economics can miss non-rivalrous intelligence and intangible value. Mustak proposes assessing material, intelligence, network, and diversity capital, supported by the principles of flow, openness, and resilience. Human care and connection remain meaningful, but cannot automatically absorb everyone displaced from work. Communities, purpose, structure, and a viable way to distribute resources are necessary to make a caring economy feasible. Concentrated ownership of advanced models could give a small number of firms disproportionate economic and political power; Mustak favors collective control of AI used in essential services while allowing diverse models for other applications. Mustak proposes using a digital asset to direct compute toward public-benefit AI, arguing that voluntary, measurable contributions to useful services could create stronger incentives than relying only on taxation or conventional corporate ownership. He argues that AI safety is shaped not only by model capabilities but also by training data, incentives, ownership, and deployment structure; culturally informed datasets and distributed systems could offer safeguards, though the proposal is not proven.

Data Points: Time since ChatGPT's release, as stated in the episode: Just over 1,000 days - Mustak uses the rapid pace of AI development to argue that societies have limited time to adapt. Corporate AI deployments reportedly not working: 95% - Mustak cites a recent MIT study, attributing the poor results partly to companies using older-generation models. Active parameters in a cited Alibaba model: 3 billion active parameters, with 30 billion total in a mixture-of-experts model - Mustak says the model performs strongly on benchmarks and could run on a CPU with 16 GB of RAM; these are claims made in the conversation. Parameters and memory cited for a medical model: 8 billion parameters and about 8 GB of RAM - Mustak says the model outperforms human doctors, while predicting further improvement; the performance claim is his. Illustrative U.S. basic income and tax figures: $16,000 per adult per year; about $5 trillion total; about $5 trillion in total U.S. taxes; $0.9 trillion in corporate taxes - Mustak uses these figures to argue that a poverty-level universal basic income would be difficult to finance through current tax revenues alone. Estimated value of adding intangible value to GDP: $96 trillion - The host and guest refer to Eric Brynjolfsson's GDP-B concept as an example of potentially large value not captured in conventional GDP. Illustrative AI-token cost comparison: GPT-3 at about $60 per million input tokens versus GPT-5 at about $1.25–$1.50 - The host cites this as an example of dramatic cost decline alongside model improvements. Estimated cost of a $20,000 robot at continuous use: About $1.50 per hour - Mustak uses the estimate to illustrate how the cost of automated physical labor could fall; it is a forward-looking example.

Pivotal Quotes: "The machines are taking our jobs. Thank God. Now we can get to our real work." — Ahmad Mustak, quoted by the host from the book: Captures the book's optimistic thesis that automation could free people for meaningful activities, provided society handles the transition. "The real impact on the economy is not a polymath coming up with a brand new thing. ... It's basically someone just blooming following instructions." — Ahmad Mustak: Explains why reliable AI execution of routine tasks may have broad economic consequences even without a dramatic leap to general intelligence. "We should be the guiders of anti-entropy, particularly if capital comes from us, you know, from being human." — Ahmad Mustak: Summarizes his vision of human value as cultivating connection and flourishing, rather than being measured solely by economic output.

Implications: If AI makes expertise and production far cheaper, access, ownership, income, and purpose may need redesign—not just workforce retraining. The proposals are ambitious and unproven, but the discussion highlights urgent choices about public-interest AI, concentrated power, economic measurement, and preserving human agency and connection.