Some Disparate Notes on AI and the Left
Socialist strategy at the beginning of recursive self-improvement
Some actors on the left have woken up to the near-future possibility, downside, and upside of transformative artificial intelligence, including One Thousand Means, Transformer, Matt Bruenig, David Shor, and New Consensus/Saikat Chakrabarti. There has been some convergence on rejecting several components of the AI-skeptic complex on the left:
The stochastic parrot view of artificial intelligence, promoted by ideological entrepreneurs on the left, has been falsified by the progression of the technology.
Even if there is an AI asset bubble, such a financial downturn would be the result of individual firms over-investing in compute scaling in an unsustainable fashion, overshooting capabilities growth by 1 to 2 years. This would not have significant implications on the probability of transformative artificial intelligence and the automation of most knowledge work within the next ten years.
It is likely that the energy constraints on compute are real and have significant climate ramifications unless the supply restrictions on solar, wind, and nuclear in the United States are lifted. However, said climate ramifications are likely crowded out in importance by takeoff.
The above views appear to be highly correlated with a bundle of other undesirable left-wing positions, such as left-NIMBYism, degrowth-sympathetic environmentalism, a generic skepticism of technological progress, and so on. The mission of One Thousand Means is to counter these views and present a Promethean alternative.
Bernie Sanders and other progressives have decided to adopt a moratorium on new data centers as the central progressive AI policy.
Such a policy, imposed on the state level, may be strategic in that it will slow down the labs and decrease the probability of dangerous recursive self-improvement before January 2029. There must be modeling work done to account for the effect of displacing data center construction to red states and the Gulf.
However, this is the only plausible reason to prefer such a policy. Should a progressive be elected in 2028, a national moratorium would be undesirable.
Even if AI systems are misaligned and unsafe, additional compute could assist with alignment research, such as through automated mechanistic interpretability research.
Even if AI systems are misaligned and unsafe, a national moratorium without a slowdown treaty with China ensures that the PRC overtakes the United States and possibly develops an existentially dangerous system.
Data center water use concerns are overstated. Utilizing such concerns strategically may reinforce blue-state NIMBYism on housing, etc. in ways that are undesirable.
Left-wing organizations, especially the DSA, should seek to nationalize local struggles over data center siting into larger questions about capital and labor. Whether left-wing candidates should do this is a pragmatic question that I do not have the data or inside view to answer, but my intuition is that for candidates focusing on the electricity affordability dimension, and connecting it to automation, is optimal for 2026, but by 2028 capabilities may have advanced to the point that such arguments would not meet the moment.
There are many alternatives to moratoria that should be explored, including renewable energy mandates combined with significant state-level reductions in land use restrictions on solar, wind, and nuclear, state-level requirements that labs spend some % of their compute in the state on safety research, compute taxes, or permitting in exchange for public equity stakes.
As capabilities progress, AI-driven automation and the resultant wealth and power concentration considerations may present one of the most beneficial issues for left-wing politics in the history of competitive partisan democracy.
The automation of labor by capital, a ‘Permanent Underclass’, resistance to regulation, agents outside of human control, a ‘Final Offshoring’. Rhetorically, transformative AI could create political cleavages reflecting the central questions posed by the left. The abdication of this issue due to technological skepticism and overriding environmentalism is leaving something potentially wildly beneficial on the table.
The GOP, especially J. D. Vance is extremely vulnerable to the issue. Vance’s background in venture capital and abandonment of any regulatory posture may be fatal if deployed by a populist Democratic candidate, especially in the context of massive job displacement or a warning-shot event. The close relationship between lab leadership and the administration has made it such that a pivot is untenable.
The Democratic political establishment is vulnerable to primary attacks on AI due to their appeasement of the sector’s lobbying efforts and their standard-bearer, Gavin Newsom’s, resistance to sufficient policy responses in the state government with the highest leverage. Progressive candidates and staffers should prepare to mobilize arguments against Newsom, and plausibly Harris, on the basis that candidates in bed with the AI industry could not possibly serve as counterweights to the labs.
There are specific research questions to be worked on here. Blue Rose Research and Searchlight (?) are the only ones working seriously in this area to my knowledge.
Many actors in the AI safety community are fearful of political polarization on AI. I understand this fear, especially in the short term re: policies to reduce existential risk. However, the administration has already shown its total unwillingness to even consider moderate AI safety interventions, and in fact tried to implement the absurdly radical policy of a national moratorium on all regulations. The AI lobby is aligned with the GOP and the administration at a deep level, and following the likely election of Chris Gober in TX-10 will have a champion with a vast network in Republican politics. At the level of policymakers themselves, AI is already quite polarized.
Regardless, should Vance be elected President, I believe the AI safety community, primarily composed of people with cosmopolitan and broadly liberal values, should be extremely concerned about the shape of the future. Extreme power concentration by malevolent actors is one of the paths to some of the worst possible outcomes. While it remains useful to have a diversified portfolio of ideological interlocutors, especially on safety, the question of disempowerment and power concentration is irreducibly political and I expect that the AI safety community, or even Anthropic, will eventually be targeted as a political enemy by the GOP. In many ways, this has already begun.
We should expect any progressive challenger to face a deluge of spending by the AI lobby.
This presents several avenues of research and preparedness, including best rhetorical strategies for neutralizing AI lobbying, the potential for mobilizing funds from Anthropic or other interests to counter OpenAI, xAI, Meta, and Google, and modeling of the effects of the maximal saturation of campaign spending. What happens when you take money in politics to the limit when the payoff of victory for the labs is extremely large?
The Alex Bores campaign in NY-12, is a viable testing ground for rhetorical and field tactics to counter the AI lobby. Even if socialists in New York do not have strong preference between Bores and other candidates, they ought to volunteer for the Bores campaign to build up skills interfacing with what may soon be the most salient policy issue. Regardless, the Bores campaign is independently worthwhile and he is the candidate in the race whose election leads to the most progressive outcomes in expectation.
Most schemes of redistribution promoted by Silicon Valley are insufficient, and it is wild that the left has not proposed its own.
Some in Silicon Valley, such as Dwarkesh Patel, and in a previous life Sam Altman, have written about redistributive mechanisms, mostly predicated on compute taxes or a tax on market capitalization capitalizing a fund that would distribute a basic income to the population.
A fund paying out dividends equal to current median incomes to the unemployed would be unfeasibly large. With small compute taxes or 5-10% taxes on market capitalization, the fund would not be able to sustain payments meeting the requirements of a displaced workforce, especially at the limit.
Such policies may be useful in the interim when capabilities timelines are uncertain.
A robust social-democratic strategy for managing the transition to an automated economy is an under-developed crucial area of planning. There is a vast space of policy interventions in 2029 that may be useful, such as using all available procurement and national security state leverage against the labs and establishing sectoral bargaining or board co-determination to expand employee leverage in negotiating the terms of automation.
As none of these policy proposals are likely to be sufficient to the challenge of truly transformative artificial intelligence, progressive candidates should pursue a vote-share maximizing platform that responds to Americans’ values, including the virtue and meaning of work, unwillingness to consider transformative outcomes, and child safety. Such proposals should be good on the merits, even if they are actually woefully insufficient to address an intelligence explosion.
However, in truly transformative scenarios, where a vast majority of human labor is automated, and ‘new jobs’ are automated in turn, nationalization under a progressive President and Congress seems to be the only viable strategy for ensuring egalitarian distributions, especially once necessary payments to the Global South are taken into account.
This should radically accelerate the political timelines of most progressives and encourage primaries in deep blue districts against incumbent leadership that would be unwilling to carry out such a programme.
Should the left wake up to TAI, there are many areas of AI safety research that egalitarians seem well-suited to approach.
This includes the work by Forethought on extreme power concentration, or on international governance.
As well as gradual disempowerment. As the authors state, “Much like cattle in an industrial farm — fed and housed by systems they neither comprehend nor influence — humans might become mere subjects of economic forces optimized for purposes beyond their understanding.” Political economists and theorists on the left have thought about this domain for more than a hundred years and certainly have insights in how to structure worlds in which human agency is maintained.
The left could make substantial contributions to several other research areas, such as well-designed national AGI projects, democratic backsliding and regulatory frameworks.
Such research would in turn strengthen the policy approach of progressive candidates and plausibly generate a healthy ecosystem.
Should the left continue in this epistemically flawed equilibrium and fail to take advantage of the situation, the space of bad worlds combining political outcomes with rapid capabilities is vast:
A Vance administration utilizing the national security state to select a national champion under executive control, resulting in oligarchic concentration of power with no precedent, enforcing a set of norms and values antithetical to human flourishing.
A Newsom administration, unwilling to take a hard line against the lab CEOs, gradually ceding power to the labs as more and more jobs and more and more government functions are automated by the models.
Or either, unable to enforce a safety-first regulatory regime or embark on the scale of a national project required to solve the technical problem of alignment, resulting in catastrophic loss of control.
Or a left-wing administration that simply bans new data center construction or AI deployment in the workforce, permitting China to advance ahead and develop potentially unsafe and unaligned AI systems.
However, if the left simply recognizes the world around them and locks in, the window for wildly good, flourishing futures crashes open. We have an opportunity to embark upon the central and noble task of our politics: freeing the human hand from the plow. A world of abundance and freedom is there for the taking.
The labs and VCs have become quite used to their enemies being unable to effectively politically mobilize or communicate, content to post about reducing the rest of the population to a permanent underclass, lobbying against any and all regulations, and ridiculing the left for not understanding that they are transforming the world. I think that if the left wakes up, rises to the occasion, and mobilizes the American population against their displacement and immiseration, they will have wished they had embarked on a more cooperative course.
This was a timed post, with the draft to be deleted if not completed by the end of a 1-hour timer. Please excuse all derelictions of sourcing and quality, but I felt I needed to get these thoughts out without obsessing over details. I have pretty low confidence in many of the claims above and seek to discuss them extensively in the future.
