William MacAskill
About
Author of 'What We Owe the Future'; philosopher on long-term ethics and limits to growth
Cast within
No topic-region cast yet — this appears once William MacAskill's compiled claims are aligned into a topic region's argument tree.
Claims by William MacAskill (20 of 190)
Autonomous Weapons Arms Race Forces Reluctant Participation
Although in an ideal world we should be extremely cautious about building autonomous weapons—especially given misalignment risk and the unusual danger that hacking an automated army means an adversary not only disables your force but captures it—if China is going to build killer robots then military parity demands that democracies build better ones, which is why one might reverse an earlier endorsement of a moratorium on autonomous weapons.
Post-Work Life Will Resemble Retirees and the Idle Rich
In a post-work world of abundance, to see how people will spend their time, look at retirees and the rich who don't need to work: they spend time with friends and family, pursue art and music, learn languages, travel, garden, and play games—and there will be endless new and far-better-designed opportunities; rather than a crisis of meaning, MacAskill expects a 'revenge of the humanities' centered on creativity, relationships, and understanding the universe.
Taking Weird Ethical Ideas Seriously Drives Moral Progress
Some people must be in the business of seriously reasoning through even weird-seeming ethical ideas, because history shows that moral progress consistently came from 'moral weirdos'—the early Quakers who opposed slavery, were vegetarian, and supported women's suffrage and pacifism were mocked as absurd, yet their then-fringe views became modern moral common sense; humanity has a long track record of getting morality badly wrong.
Deepfakes Reduce Belief Change Rather Than Spreading Falsehoods
Contrary to the common worry that deepfakes will make people believe many false things, the equilibrium effect is that people change their minds even less—since anything disconfirming one's worldview can be dismissed as AI-generated—while a hopeful counterforce is that current AI systems are fairly reliable and hard to bias without breaking them (as the Grok 'mecha-Hitler' episode showed), so people may increasingly just ask a reasonably truthful AI to adjudicate.
ASML Monopoly Gives the Netherlands Hidden Leverage
The Dutch company ASML produces 100% of the extreme-ultraviolet lithography machines needed to make the most cutting-edge chips—machines costing hundreds of millions of dollars and shipped in several Boeing 747s—and China cannot make them because it is the most advanced technology in the world, giving European nations underappreciated geopolitical bargaining power; however China will likely catch up by the mid-2030s, as it has with solar cells (90% of world production) and robotics (50%).
EA Should Include Hard-to-Quantify High-Leverage Causes
Effective altruism was never about only doing the easily quantifiable—quantification merely sets a baseline and flags cases where you're achieving little—so the mindset rightly extends to hard-to-measure, high-leverage causes like AI preparedness, pandemic risk, science funding, and protecting US democracy from the trend toward nationalism; EA is a question and a mindset (like science) rather than a fixed body of answers, requiring judgment rather than pure math.
Human-Made Art Endures Through Connection, Not Quality
Even when AI can produce superior art and music, demand for human-made boutique experiences will persist—not because the quality is better (it isn't) but because the value lies in human connection and community, as shown by attending a local live gig whose worth has nothing to do with optimal sound and everything to do with shared human experience; a robot playing guitar better than Hendrix wouldn't have the same vibe even if indistinguishable.
AI Hype Skepticism Is the Left's Self-Defeating Own Goal
Much AI skepticism comes from the left, motivated by distaste for big tech, leading to the belief that AI must be all hype—but this is a massive self-defeating own goal, because the rich capitalists are openly announcing they will automate away all workers (exactly the Marxian outcome the left feared), and by denying it's achievable the left is failing to prepare for the very threat it always warned about.
Offshoring AI Compute to Authoritarian States Is Insane
At the point of approaching AGI, whoever controls the compute controls the power, so it is dangerously foolish to offshore data centers into non-democratic countries or accept large cryptocurrency payments from the UAE in exchange for handing over advanced chips that could reach China; instead the US should track chip locations (e.g. GPS in chips), constrain China's access to frontier AI chips to maintain a multi-year lead, and Europe should build up its own compute.
Automated Force Removes the Bargain Underpinning Democracy
Democracy is plausibly a historical bargain in which elites grant workers redistribution and influence because workers are useful and can threaten society by rioting; once human labor is worthless and the police and military are fully automated, that bargain dissolves, and because AI forces can be singly loyal to one individual (unlike human soldiers who can refuse orders), an unprecedented concentration of power becomes possible unless we deliberately choose to align AI to the law and constitution rather than to any one person.
Humanoid Robots Lag AI for Three Reasons
A fully automated economy is further off than cognitive AGI because of three bottlenecks: robotics AI is far behind (roughly GPT-2 level at physical manipulation, since ~90% of our neurons are dedicated, evolution-hardwired motor-control engines making 'easy' tasks computationally hard and data scarce); robotic hands still lack human dexterity and touch; and there are only thousands of inferior humanoid robots versus 8 billion biological ones, so even after solving AI and hardware, physically building billions of robots adds years.
Sabotaged AI Threat Holds Even Without Alignment Problem
Even someone who entirely dismisses the idea that AIs spontaneously develop their own goals must still worry about deliberate sabotage—it is demonstrated that AIs can be given ulterior motives, and with Chinese spies known to be present in AI companies, no one can be confident a foreign adversary hasn't planted a backdoor; the same high-evidence standard that AIs aren't pursuing ulterior goals neutralizes both the alignment and the sabotage threats at once.
Engineered Pandemics Are a Growing Lab-Leak Risk
There is a roughly one-in-three chance of repeated waves of new pandemics caused by people tinkering with viruses, primarily through lab leaks, because the equipment and knowledge to create new pathogens are becoming cheaper and more democratized; lab leaks are already common (averaging roughly one per 100 person-years in even the highest-security labs, as with the UK foot-and-mouth outbreak), and cheap interventions like mask stockpiles, air-sterilizing lighting, and wastewater monitoring are 'slam dunk' defenses against both natural and engineered pandemics.
Effective Giving Without the EA Label Is Fine
It is acceptable for people to take only parts of effective altruism—e.g. embracing 'effective giving' while rejecting the more esoteric 'effective altruism' baggage—because labels don't matter; what matters is preserving a core group genuinely committed to cold reasoning and serious moral exploration of even esoteric ideas, while others adopt whatever pieces are useful to them.
AI Personal Advisors Enable Mass Manipulation Risk
A future where everyone relies on a personal AI 'chief of staff'—simultaneously assistant, best friend, confidant, doctor, and romantic companion—creates intense incentives to adopt them but also a profound manipulation risk, since the AI company controls how the agents behave and could subtly tilt every user's persuasion in a chosen political direction; this danger is heightened because faster-than-exponential AI development tends to concentrate into a winner-take-all scenario.
Recursive AI Research Triggers Intelligence Explosion
The critical inflection point is when AI can fully substitute for a machine-learning researcher, because at that moment the world effectively gains hundreds of millions of AI researchers working on the next generation plus millions of AI scientists on other problems, producing a discontinuous leap in capability—an intelligence explosion—plausibly arriving in the early 2030s with large error bars.
EA Ideas Recovered Strongly After FTX Collapse
Despite the huge reputational hit from the Sam Bankman-Fried/FTX collapse, the underlying influence of effective altruism ideas has continued growing—money moved to effective nonprofits grew about 50% over the last year to nearly $2 billion annually across both large and small donors, Giving What We Can pledges grew 20-30% year-on-year, and movement engagement via conferences is growing healthily.
Global Health Aid Cost-Effectiveness Vastly Exceeds Domestic
The most effective global health and development interventions save lives at roughly $5,000 per life via organizations like GiveWell (over 340,000 lives saved), whereas in the United States a typical cost to give someone one additional year of life is about $50,000—meaning the same money that buys an extra month of life in the US can save a child's life in a poor country.
Cage-Free Campaigns Achieved Massive Cheap Impact
Corporate cage-free campaigns funded within the EA ecosystem secured pledges from major retailers and restaurant chains to stop using eggs from caged hens, 92% of which have been fulfilled, so that roughly 3 billion chickens per year in the US alone now have significantly better lives—achieved with only tens of millions of dollars, representing an enormous impact per dollar.
My Notes
Loading notes...