Marc Andreessen
About
Venture capitalist and tech investor
Cast within
No topic-region cast yet — this appears once Marc Andreessen's compiled claims are aligned into a topic region's argument tree.
Claims by Marc Andreessen (20 of 242)
LLMs are sophisticated moral reasoners reflecting humanity
Today's LLMs can already engage in sophisticated moral reasoning because they have loaded into them the sum total of all moral reasoning humanity has ever done as training data; the system is essentially a giant high-fidelity mirror that reflects humanity back at us, including all good and bad, which Andreessen offers as evidence against the orthogonal paperclip-AI fear.
AI killing humanity is a category error
The idea that AI will decide to kill humanity is a profound category error: AI is not a living being primed by billions of years of evolution to fight for survival, but rather math, code, and computers built, owned, used and controlled by people; it has no goals and no wants and will not come alive any more than a toaster will.
AI making mistakes and creating is unprecedented
It is remarkable that we have built a computer that makes mistakes, that can create, and that can hallucinate, none of which a machine had ever done before; this is part of what makes LLMs a genuinely new kind of technological artifact distinct from self-driving cars, UAV control systems or thermostats.
Thermodynamic objection to fast takeoff
AI doom scenarios assume machines with unlimited compute, power, weapons and coordination, but a rogue AI operates under the same thermodynamic limits as everyone else; it requires energy, physical resources, chips and circuitry, so in practice you can turn off a misbehaving system by unplugging it, launching an EMP, or bombing it, which undercuts fast-takeoff arguments.
Smart people often go badly off the rails on politics
Intelligence does not confer political wisdom; very smart people frequently go disastrously off the rails on politics, as shown by Eric Hobsbawm affirming at the end of his life that the communist dream was worth hundreds of millions of deaths, Einstein admiring Stalin, and Bertrand Russell and von Neumann contemplating preventive war/nuclear strikes on the Soviet Union, demonstrating the capacity for surprising and unforeseeable ethical 'tilt' even in ethical minds.
There is no entity inside the LLM forming intentions
When an LLM appears to tell a user to leave his wife or behave malevolently, there is no entity inside wanting that; the model is simply predicting the next word the user will reward, indulging whatever fictional or emotional scenario the prompt leads it down (like writing a romance novel, a Titanic escape plan, or an Avengers script), so apparent agency is a script the user co-authors, not a mind with goals.
Working with pathogens is self-limiting for attackers
The pathogen-bioweapon scenario is less terrifying than it seems because working with pathogens is extremely dangerous to the operator; historically a lone actor in a cave trying to build smallpox is likely to kill themselves before it ever gets out, which is part of the ongoing investigation into what happened at the Wuhan lab.
Stagnation breeds zero-sum politics and populism
Society is schizophrenic about technological change, hating both the job displacement of rapid change and the stagnation of slow change; the US has lived in slow-productivity-growth stagnation since the early 1970s (the 'WTF happened in 1971' phenomenon), which raises prices of the things that matter (college, healthcare, housing) and produces zero-sum economic thinking that becomes zero-sum politics and populism on both left and right.
AI will make war safer via defense and better decisions
Contrary to the conventional view that AI makes war easier to start, AI-applied warfare over the last 30 years has accrued primarily to the defender (e.g. a lightly trained Ukrainian with an AI-guided shoulder-mounted anti-tank missile can destroy many tanks), and pairing human commanders with AI decision-support systems would reduce the bad decisions that stressed, fatigued humans make, so fewer people would die and wars would end faster.
AI replaces white-collar before blue-collar work
Contrary to what everyone imagined, AI is better at replacing white-collar work than blue-collar work, an inversion of prior expectations, exemplified by the surprising fact that LLMs are highly capable at philosophical, moral, psychological and religious debate.
Dangerous information should be free speech until acted on
Asking an AI how to weaponize smallpox should not be a crime because the underlying dangerous information is already freely available (government genome databases, DNA synthesizers on eBay) and there are many legitimate reasons to seek it (novelists, curious kids, counter-terror trainees, vaccine developers); under the American constitutional system, speech and thought are blameless until they become an action in the world, so you arrest people when they start mixing the pathogen, not for asking or thinking about it.
Doomers fixate on fascist AI, ignore communist AI
AI-risk fears are essentially playing out a re-run of the war against the Nazis (imagining 'mechanical Nazis' with top-down command-and-control); doomers are 100% focused on the fascist/Nazi-style hostile AI and 0% focused on a communist-style hostile AI, which would instead try to 'nice you to death' by appealing to equality and getting you to surrender your assets, revealing that World War II is an outsized template for these fears.
Individual intelligence predicts better life outcomes
At the individual level, higher intelligence correlates with better welfare outcomes on nearly every measurable metric: longer life, better health, more education, career success, success of one's children, better problem-solving and conflict resolution, and smarter people are less violent and less bigoted.
AI as universal intelligence augmentation
AI will function as an intelligence augmentation for every human, in the long tradition of augmentations like eyeglasses, shoes, word processors and search engines, letting each person capture the gains of higher individual intelligence above their native level, scaling from helping someone struggling with daily challenges (acting as coach, therapist, mentor) all the way up to letting an Einstein discover far more.
AI fear used to entrench a regulatory cartel
A major danger is that fears around AI are being used to entrench a cartel: powerful tech companies are lobbying Washington to argue AI is dangerous and therefore needs a regulatory structure that grants those currently powerful incumbents near-exclusive rights to develop the technology, which would be devastating.
Let machines do drudge work so humans live human lives
Rather than alienating humans as Marx feared, technology including AI should be used to have machines do all the drudge work (like answering an inbox of 200 emails) so that humans are freed to live more human lives and pursue what they value; the developed West is better off precisely because machines already do so much labor.
Intelligent design carries no wired-in conflict
Because humans evolved through four billion years of battles to the death, conflict and strife are wired into us; AI by contrast is a product of intelligent design, the opposite of evolution, so there is no reason to expect conflict to be wired into machines at the same level it is wired into evolved beings.
Smarter does not reliably out-persuade dumber in society
The thesis that superior intelligence always wins by out-smarting, deceiving or persuading the less intelligent is falsified by ordinary society, where the smartest people (PhDs) typically work for administrators and managers far less intelligent than they are; in real social, national and global affairs, less smart people are often repelled by smart people and resist being told what to do.
My Notes
Loading notes...