Why the 2024 election matters, the AI risk case for Harris, & what you can do to help

Alex Lintz24 Sep 2024 19:32 UTC
23 points
2 comments19 min readLW link
Jason Vines, 2015. Creative Commons

Epistemic status: I’ve attempted to be objective in my comparison of candidates but the result is surely far from perfect. Also, given the election is so soon, I haven’t had time to solicit nearly as much feedback as I would have liked. Please take our conclusions with a grain of salt.

[Note: This is similar to a post I wrote on the EA Forum but contains important revisions and more information on cost effectiveness & AI risk concerns.[2]]

Executive Summary

I have spent recent months investigating and working on the election with a group of volunteers[1], many of whom come from AI safety circles. This piece makes the case that electing Harris is both incredibly important and that it’s still possible to meaningfully affect her chances of victory. Taking into account key considerations (e.g. AI risk concerns and the potential for short timelines) and using some rough modeling, we conclude that contributing to the 2024 US election may be among the most impactful uses of time and money currently available.

This election matters more than most give it credit for

  • We believe electing Harris rather than Trump would be highly beneficial in expectation. Judging the candidates on characteristics which would likely matter under short timelines scenarios, we think that while Trump has some advantages (e.g. project execution, willingness to work outside the Overton Window), his disadvantages (e.g. disrespect for democratic institutions, extreme power-seeking behavior, and lack of competence) far outweigh them.

  • While relatively less is known about Harris, we conclude she’s much more likely to govern responsibly and that electing her is the right call from an AI risk perspective.

This election is more tractable than most believe

  • US presidential elections are often decided by slim margins in a few swing states (by 537 votes in 2000, 118,601 in 2004, 77,744 in 2016, and 42,918 in 2020).

    • The best polling data we have as of September 2024 suggests that this election’s margins are likely to be in similar range.

  • Given how small margins of victory have been, many organizations have arguably been responsible for flipping past presidential elections (and many more could have flipped them with modest investments).

  • While a lot of good work is being done, poor incentives among organizations attempting to elect Democrats result in many inefficiencies. Some examples include: incentives to get credit and funding often conflict with impact and low wages and career instability reduce access to talent.

Donating & volunteering are likely among the best ways to spend money and time this year

  • Our model suggests that $10 million in effective donations could increase the chance of a Harris victory by something like 0.16%.

  • Based on this model and some educated guesses, we tentatively conclude the reduction in existential risk gained from donations to the election are competitive with or better than top donation opportunities available now.

  • Research by a team with expertise evaluating election interventions and non-profits suggests the following are the most effective donation opportunities right now:

  • We’re also collecting useful volunteer opportunities and would be happy to share them with you. You can sign up to hear more here.

  • We find it motivating to reflect on how we would feel if Trump becomes the next president, especially if he wins by a small margin (the most likely outcome at the moment), knowing we didn’t do what we could to stop him.

Introduction

The US is the world’s most powerful state and the only superpower that is also a liberal democracy. If one of the presidential candidates is expected to be much worse on important issues (e.g., protecting liberal democracy, AI safety, biosecurity, climate change, global health, or animal welfare), and if the election is expected to be very close, then contributing to helping the better candidate win has the potential for extraordinary value.

We know of no lever, other than the US presidential election, that allows as many individuals to clearly influence such important trajectory changes. Several community members have dropped other AI work to stop Trump from winning this year.

The 2024 election is much more important than other elections

This is an especially critical period for AI

Many leaders and employees at frontier AI labs have discussed having TAI timelines before 2030 (Sam Altman: 4-5 years, Dario Amodei: Human level in 2 years, Shane Legg: 50% AGI by 2028). This means the next President has a good chance of presiding over a decisive period for addressing AI risk concerns.

If we reach AGI in the next 4 years there will be many key decisions to be made around governing AI, deployment, and international cooperation. The ability of the president to think critically, learn from advisors, and work with competent people may be crucial for humanity’s future. Even if timelines are longer than 4 years, the next few years will be important years for getting in place protections, safety funding, and legislation which can mitigate risk down the line.

Assessing the candidates on their ability & willingness to mitigate catastrophic risks from AI

While other factors matter a lot when it comes to the value of a presidency, let’s focus for now on how we expect the candidates to fare on preventing existential risk from AI. To add a degree of objectivity and specificity, we’ll focus on individual characteristics we’d like to see in a world leader governing the transition to potentially dangerous AI systems.

Appreciation of risks from AI: Unclear

  • Harris was personally behind securing safety commitments from top companies in July of 2023, though not many of these focused on catastrophic risks. Harris has generally put more emphasis on current harms, highlighting that local/​personal harms feel existential to individuals in a November 2023 speech. That said, in the same speech, she acknowledged that AI might “endanger the very existence of humanity”, citing “AI-formulated bioweapons” and “AI-enabled cyberattacks” as particular concerns.

  • Trump has voiced some concerns about dangers from AI, especially around deepfakes and nuclear concerns. Trump seems likely to adopt the views of his backers on AI safety. This could go either way as he has some strong anti-regulation donors but Elon could also be a key influence. His platform does say he plans to repeal Biden’s executive order on AI safety on day one, with no sign that he would keep safety provisions.

Willingness to listen to AI-concerned advisors: Moderate Harris win

  • Harris’ administration would likely inherit much of the Biden administration’s relatively good relationship with the safety community. To the extent we’re making progress on AI safety legislation now, that will likely continue apace under a Harris administration.

  • Trump may be influenced by Elon Musk to take AI safety seriously (e.g. Elon supported SB 1047). That said, he has some top donors who are strongly against AI safety concerns (e.g. Marc Andreesen, Ted Cruz) and, on priors, we should expect Elon will eventually fall out of favor.

General competence: Moderate Harris win

  • Harris has a far larger pool of talent to draw from and will likely retain many Biden staff who have performed at least adequately. Harris appears to be unusually difficult to work with and has a high staff turnover rate. This could reduce her ability to recruit top talent.

  • Trump’s top people have typically been relatively incompetent—with the most competent often leaving. The former head of Trump’s National Economic Council, Gary Cohn, purportedly said, “It’s worse than you can imagine. An idiot surrounded by clowns. Trump won’t read anything… He gets up halfway through meetings with world leaders because he is bored. And his staff is no better.” Trump’s tentative interest in putting RFK Jr. in power as head of HHS is quite damning. That said, Trump could bring in some unusual minds like Musk or Ramaswamy who could make radical changes to US bureaucracy.

Sense of caution and taking general risks seriously (e.g. about deployment): Moderate Harris win

  • Harris’ views toward risk aren’t very clear. She hasn’t occupied positions where we could learn much about her decision-making on this issue. That said, she seems likely to fall in a similar camp to someone like Biden in being reasonably concerned about risk.

  • Trump has had a tendency to take risky actions with little consideration. Whether or not it was the right move, his decision to assassinate Iranian General Soleimani was apparently made with almost no deliberation. He reportedly seriously discussed nuking North Korea.[3]

Willingness to do things outside the Overton Window (which might be necessary to ensure AI goes well): Moderate Trump win

  • Harris appears to be fairly conservative when it comes to pushing the Overton Window. We’re not aware of actions she’s taken which deviate significantly from likely status quo actions. To be fair, she hasn’t faced many situations where we would have heard of her doing so even if she did.

  • Trump is clearly willing to do unusual things without worrying too much about upsetting the status quo. Assassinating Soleimani is one clear example. Project Warp Speed is another where perhaps another administration could have gotten bogged down in red tape or been less willing to cooperate with the private sector. He was also willing to attempt to overthrow the election, which is very far outside the Overton Window.

Smart and able to discern good arguments: Strong Harris win

  • Harris doesn’t have a stellar record with former employees of hers but none question her competence. By all accounts she seems to be a relatively strong critical thinker, if not necessarily a standout.

  • Trump’s prior advisors and people he’s worked with nearly universally come away with a poor view of Trump’s reasoning ability. Some comments from advisors include: ‘moron’ (Tillerson), ‘He’s like a … defiant 9-year-old kid, who’s always pushing the glass toward the edge of the table defying his parents to stop him from doing it.’ (Barr), and he doesn’t have ‘the competence to carry out the job’ (Bolton).

Respect for democratic norms: Strong Harris win

  • Harris has fairly normal views on democratic norms. She would respect the outcome of the election, has not made statements questioning the validity of election results, and is generally a standard US politician in this regard (that is to say, reasonably good).

  • Trump has repeatedly taken actions which indicate little respect for democratic norms. Among other things, these include expressing admiration for dictators, saying he would be ‘dictator on day one’, and vowing retribution against his political opponents. He made blatant and undeniable attempts to overturn the 2020 election. More on this in the next section.

Not overly motivated by power[4], someone who is unlikely to use powerful AI for personal or political gain: Strong Harris win

  • Harris appears highly ambitious and, on priors, given she’s a powerful career politician, seems likely to be motivated by power to a strong extent. That said, articles about her management style don’t seem to reveal any unusual degree of power-seeking when compared to most career politicians. Overall, we don’t know much about how she thinks behind the scenes but what we’ve seen publicly suggests she’s not unusual on this metric.

  • Trump, by all accounts, appears incredibly motivated by attaining and keeping power. His former National Security Adviser, Bolton, said he was, “hard-pressed to identify any significant Trump decision during my tenure that wasn’t driven by re-election calculations.” Trump’s former attorney general, Bill Barr, remarked “he is a consummate narcissist. And he constantly engages in reckless conduct. … He will always put his own interests, and gratifying his own ego, ahead of everything else, including the country’s interests.” Trump’s own ghostwriter disclosed that Trump has an “absolute lack of interest in anything beyond power and money”.

Competence in executing projects, not prone to bureaucratic red tape: Moderate Trump win

  • Harris doesn’t have much of a record to look at here. She doesn’t seem to have started large initiatives as VP. She made some contributions to stemming illegal immigration by addressing root causes but it’s unclear whether she had much impact. Her senate career wasn’t marked with any particular successes on getting bills passed, though she was in the minority while in office. Overall, she doesn’t appear to have led major projects as VP, Senator, or AG but it’s unclear how this would change if she had more ability to delegate as she would in the Presidency.

  • Trump managed to push through at least one impressive public-private partnership during his term, Operation Warp Speed. He also decreased red tape in the pharma sector to positive effect, made an energy trade deal with Europe, and initiated the withdrawal from Afghanistan. These actions indicate a willingness to make moves quickly without getting bogged down in red tape. That said, Trump had several initiatives which failed due to legal challenges or opposition from within government.

Other characteristics relevant to navigating risks from AI:

  • Forward-looking, able to adapt to a new paradigm: Weak Harris win

  • Desire not to unnecessarily hobble companies:[5] Weak Trump win

  • Ability & interest in international negotiations: Weak Harris win

Ultimately, people will rank the candidates on each of these metrics differently and the relative importance of each metric is unclear.

In our eyes some of the more important factors in Harris’ favor are her relative lack of power-seeking, the general competence of her & her team, her respect for democratic institutions, and the likelihood that she will listen to reasonable advisors about AI safety. On the other hand, Trump seems able and willing to execute on unusual policies and actions to a greater degree than Harris. It’s also possible he could swing hard toward being concerned about AI safety, though his position ultimately seems higher variance and it’s unclear whether he’d take reasonable actions were he convinced.

Though Trump has some advantages, we think his strong power-seeking drive, lack of respect for democratic institutions, and general lack of competence are essentially disqualifying all on their own. We think these characteristics represent too great a risk when considering that the next president could be responsible for guiding the country through a transition to AGI. Overall, we believe Harris is considerably more likely to safely manage the transition to powerful AI systems compared to Trump.[6]

A second Trump term would likely be far more damaging for liberal democracy than the last

Another key consideration for the value of a presidency is the likelihood that they would weaken Democratic institutions. Democracy welcomes free and open debate of ideas. It challenges leaders to benefit constituents — or be quickly, peacefully booted. Authoritarianism, meanwhile, has more often let an insulated group of elites sit back as famines, genocides, and other catastrophes unfolded around them — sometimes at their command. We expect a second Trump presidency to be much worse for liberal democracy, both domestically and globally (see the section on US-China and international relations below), than a Harris presidency.

In his first term, Trump and allies were held back by sane civil servants who rejected their dangerous ideas, such as launching nuclear weapons or using the military to overturn the election. Trump’s failure to overturn the 2020 election has shaped his second term agenda, which is now aimed at destroying checks and balances through unprecedented power over the military, courts, and key agencies. His plans reflect these new priorities:

  1. Execute a massive purge of independent civil servants: Trump has mentioned several times his desire to pass the “Schedule F” executive order. This would give him the power to fire up to 50,000 civil servants who have traditionally checked the president’s power, including in legal, regulatory, and military contexts.

  2. Assemble vetted loyalists: Trump’s allies have spent tens of millions on “Project 2025”, one goal of which is to screen Trump loyalists to replace independent civil servants. In their own words, “Our goal is to assemble an army of aligned, vetted, trained, and prepared conservatives to go to work on Day One to deconstruct the Administrative State.”

  3. Expand presidential control: As reported, “Project 2025 proposes that the entire federal bureaucracy, including independent agencies such as the Department of Justice, be placed under direct presidential control.”

  4. Appoint more loyal judges: While this happened in his first term as well, it’s become apparent just how impactful the appointment of judges loyal to Trump has been, at the Supreme Court, appellate, and district levels. Both Trump’s immunity ruling & Cannon’s ruling to throw out Trump’s classified documents case have been highly unusual & appear politically motivated.

Another key change from Trump’s last term is the presence of an extremely Trump-friendly Supreme Court which has, among other things, granted the president unprecedented immunity from the law. It remains to be seen just how much immunity Trump would have, but in her dissenting opinion, Judge Sotomayor said, “[When the president] uses his official powers in any way, under the majority’s reasoning, he now will be insulated from criminal prosecution… [if he] orders the Navy’s Seal Team 6 to assassinate a political rival? Immune.” It seems likely that Trump would be legally able to take or offer bribes and it’s possible he couldn’t be prosecuted even for organizing a military coup. While the full implications remain unclear, experts are near-universally concerned about this ruling and its implications for how Trump could behave in a second term.

It’s worth spelling out just how concerning and forceful Trump’s attempts to overturn the 2020 election were. He pressured state officials not to certify the election, considered using the military to seize ballots in an effort to ‘prove fraud’, tried to use the DoJ to legitimize his fraud claims, pressured his VP to dispute the election results, and ultimately was responsible for a violent march on the capital.

Perhaps most damning, Trump tried to implement a fake electors plot in which he asked Mike Pence to certify electoral college votes which falsely claimed that Trump had won the election in some states where he had lost. Dozens of the false electors have since been indicted criminally. Had Pence been willing to go along with the plot, it’s unclear what would have happened. Unfortunately, no such check is likely to exist in 2028 should an attempt be made to circumvent the process for the Republican nominee—Vance has said he would have gone along with the plot had he been VP at the time.

While each has been deniable, Trump has made allusions to attempting to stay in power beyond his second term, saying things like, “We are going to win four more years. And then after that, we’ll go for another four years because they spied on my campaign. We should get a redo of four years.” Characteristically, he’s also said he wouldn’t attempt a third term. It appears highly likely that Trump will take actions to further degrade democracy in the United States. The possibility that he attempts to illegally tamper with the 2028 election should be taken seriously.

Influencing the US election is tractable

US presidential elections are often close

US presidential elections are often surprisingly close.

  • Biden won the last election by 42,918 combined votes in three swing states. Trump won the election before that by 77,744 votes. In 2000, just 537 votes (and likely some Republican meddling) in Florida decided the election for Bush.

There’s a good chance the 2024 election will be very close too (i.e. likely decided by <300,000 votes)[7]

  • While Biden dropping out has improved our chance of victory massively, the race is still tight. In August 2020 Biden was up about 8 points nationally while today Harris is up just 2.4 points. Prediction markets & analysts differ on who is favored to win, but all expect a tight race.

  • Tipping point states are highly likely to be Pennsylvania, Wisconsin, or Michigan where the candidates are neck-and-neck. Public polls show a tight race in these states. The Economist claims, ‘this presidential race is tighter today than any since polling began.’

Given how close US elections tend to be, many efforts can be counterfactually responsible for flipping the election.

  • For example, most major donors or moderately-resourced Republican political organizations which operated in Florida in 2000 can reasonably claim counterfactual responsibility for Bush’s victory. We’ll never know for sure, but there are likely thousands of individuals who were counterfactually responsible for the election results. Tens of thousands of individuals likely could have done enough to flip the election to Gore had they put in the effort.

Having an impact on the election is relatively straightforward

The election is more tractable than a lot of other work.

  • The general theory of change is fairly straightforward: Increase the share of votes to the candidate who is more likely to effectively govern the US & deal with major societal risks.

  • The closeness of the election allows many individuals with non-extraordinary skills or power to have a shot at making a decisive difference in the outcome of the election.

  • Feedback loops are highly measurable and concrete. Whether someone voted in an election is publicly available information, and how they voted is highly predictable from demographic and other data. This has made US electoral politics much more RCT-able than most EA cause areas.

There is still low-hanging fruit

Estimates for how effectively top RCT-tested interventions generate net swing-state votes this election range from about several hundred to several thousand dollars per vote.[8] Top non-RCT-able interventions could be even better.

Most political donations don’t go to effective organizations.

  • Because most political donors aren’t particularly evidence-focused or attempting to optimize their giving, most donations go to relatively ineffective groups. This means that many effective groups are often not funded to a point of saturation despite the money pouring into the presidential election. That said, many top organizations are doing good work that isn’t dramatically less effective than our recommendations (e.g. Harris’ campaign, Future Forward)

Many promising strategies have not been sufficiently explored or are underfunded.

  • Many cost-effective strategies can’t scale as much as they’d like because they’re underfunded. See here for more info on those organizations.

  • Anecdotally, people with little political background have been able to generate many ideas that haven’t been tried and were received positively by experts.

    • One group we’re in touch with, started just a few years ago by people with almost no political background, has rocketed to being among the top organizations in Democratic politics. This organization advises donors with sophisticated research on effectiveness & helps coordinate giving. We believe their work may have had crucial impacts on prior elections.

    • Though new to this field, the team I’ve been working with has managed to generate some ideas which highly experienced advisors think are promising and which are not being addressed by existing groups.

Even many top organizations aren’t operating very efficiently.

  • Biden’s post-debate campaign spent tens of millions on ads long after it became clear Biden should drop out. The campaign team consistently rationalized away evidence against Biden, ignored polling, and transparently lied to the public.

  • Many of the most prominent Democratic organizations use messaging which testing indicates are ineffective. For example, when attempting to mobilize people to vote, most organizations use issue-focused language rather than best practice methods despite repeated evidence that using these methods would make their efforts far more effective.[9]

  • According to an expert we’re in touch with, as much as ~$400M is expected to be spent on paid canvassing. While paid canvassing can work, most organizations in this sector don’t use best practices. Research shows these efforts often result in null or even negative results.[10]

  • We’ve been frustrated with the unwillingness of some top fundraising organizations to scale their outreach and operations. They are often using outdated technology and failing to find ways to automate simple tasks (e.g., managing email lists).

Electoral politics has poor incentives and talent retention

From an outside view it’s reasonable to believe that, given the amount of money spent each cycle, some kind of efficient market should exist within electoral politics. However, there are a number of characteristics of the sector which make it less efficient than one might think.

Top talent is not well-incentivized to work in electoral politics

  • Jobs in political campaigns are cyclical/​temporary, demanding, poorly compensated, and offer uncertain career capital (i.e., low rewards for working on losing campaigns).

  • In part because of these poor incentives, politics often attracts politically motivated people who often have a hard time empathizing with undecided voters and moderates.

Efficacy is often poorly incentivized

  • Because elections are necessarily adversarial, much of the research about effective election interventions is considered risky to share. If the other side finds and uses the research, the potential upside is nullified. This means that information sharing is limited and the best information about effective interventions is not as widely known as one might hope.

  • Those working on campaigns are often seeking jobs in the administration after winning. Because of this, they’re often more interested in their reputation with the candidate than actually winning. This competition can lead to internal tension which is not conducive to effective decision-making.[11]

  • Similar to the international aid sector, election nonprofits are accountable to donors first and foremost. This means they’re incentivized to overvalue the effectiveness of their work and to operate in a way which appeals to donors rather than aiming to win the most votes (e.g., messaging which appeals to donors rather than undecided voters).

Should you donate to the election?

What’s the chance that donations flip the election?

To get a better sense of the tractability of influencing the election, we made a model to estimate a lower bound for the effectiveness of donations. The model make the following key assumptions:

  • It costs $2000 to attain one net democratic vote in swing states

    • This is a fairly conservative estimate, some experts claim it’s closer to $1000 on the margin but others say it’s around $2500. This takes into account funding gaps and counterfactuals.

  • Election outcomes follow a normal distribution centered on the vote shares predicted by current polling.

  • Only about 25% of donations go to the most effective state, Pennsylvania. The other 75% will go to less effective swing states.

Our model predicts that $10 million in donations to the most effective organizations would have between 0.11% and 0.46% chance of flipping the election. Our mainline estimate is that $10 million would have a 0.16% chance of changing the outcome in Harris’ favor.

For more details on the model, see the Appendix.

Overall effects on existential risk: Trump vs Harris

The following is a rough attempt to pull together the effect of a Harris victory on mitigating existential risk from an AGI catastrophe within the next 8 years. This is based on extremely rough and made up numbers but we think it’s a useful exercise.

Here’s some numbers (feel free to make a copy to run your own numbers)

  • Probability of AGI before 2032 = 25%

  • Probability of existential catastrophe before 2032 assuming AGI arrives in that period and Harris wins[12] = 30%

  • Probability of existential catastrophe before 2032 assuming AGI arrives in that period and Trump wins[13] = 35%.

This quick guesswork implies a ~15% (or 1.25 percentage point) reduction in the likelihood of an existential catastrophe due to AGI within the next 8 years if Harris wins.

Other large contributing impacts are the effect of a Trump administration on our ability to survive AGI after 2032, probability of nuclear war, ability to reduce the chance of catastrophic biorisks, influence on future election cycles, and stable totalitarianism. We can also consider the potential decreased value of the future should a Trump administration guide us into the post-AGI future.

If we conservatively assume these other risk factors are equivalent in value to the risk from AGI over the next 8 years, we estimate a Harris victory would reduce existential risk by 2.5%.

Assessing the relative value of donations

Combining these estimates with the earlier model of the effect of donations on flipping the election, the implied overall decrease in existential risk is 0.003 − 0.01% for each $10 million donation to improve Harris’ chance of victory. To get an intuition check on what these numbers mean, we can calculate the theoretical impact of spending all the money committed to EA causes (let’s assume $26 billion) at this level of impact. Doing so would be similar in value to a 10-30% reduction in existential risk. It’s not clear how that compares with other causes, but it seems likely worthwhile to use all EA money in exchange for a guaranteed 10 percentage point x-risk reduction . Of course, the election can’t productively absorb $26 billion (we know of $4 million in funding gaps we’re especially excited about and expect at least another $20 million could be absorbed productively).

It’s challenging to compare donation opportunities against significantly different cause areas. However, considering marginal donation opportunities for the election and other top causes, we feel reasonably confident that election donations are among the best opportunities for donations currently available.

Downside risks & clarity of impact

The primary potential downside of donations to the election is that the EA & Rationalist communities could become more politicized and be less impactful under a potential Trump administration given more donations. We think this is a real concern and are wary of politicizing discussions around AI safety. We expect posts like this won’t have large effects on the margin and are likely worthwhile in expectation, especially given other posts arguing for Trump. We’re doing this in part because there are so few posts about the election on Rationalist or EA sites.

On the other hand, perhaps we’re fundamentally wrong about which candidate is best. For instance, we could be wrong about Trump’s impact on the probability of existential risk. Perhaps, for instance, it will be crucially important that our next president carries out some highly unusual and risky actions quickly. In a case like that, a Trump administration may be a better fit. Perhaps Elon Musk will influence the next Trump administration’s AI policy to prioritize x-risk mitigation, as evidenced by his support of SB 1047 (though his other actions and beliefs are cause for concern on this front).

Conclusion & recommendations

There is of course a high degree of uncertainty but we lean toward more donations from the rationalist and EA communities being worthwhile. We feel confident that at least $20 million would be worth donating but haven’t yet analyzed the extent to which cost effectiveness decreases after that point.

It’s worth noting as well that the election is an unusual opportunity in that it can absorb a lot of money in ways that AI governance or safety cannot.

Donation, volunteer, & fundraising opportunities

Top recommendations

We believe these recommendations, put together over the last few months by a team with considerable expertise in election impact evaluations, are the best publicly shareable resource for donors available today. Their top recommendations include the following:

  • The Center for US Voters Abroad (CUVA) contacts Americans living abroad in an effort to mobilize them to vote. Because US citizens living abroad overwhelmingly lean Democrat and are a historically neglected demographic, mobilizing them is expected to be a highly cost-effective intervention. Donate here (US citizens only) or here.

  • Working America (WA) pairs a year-round membership-based program with evidence-based electoral contact during campaign seasons in order to maximize its impact. They have several high-quality RCTs demonstrating effectiveness and their interventions are highly scalable. Donate here (US citizens only).

  • American Independent Radio (AIR) runs radio ads which aim to increase voter turnout. These interventions have a solid evidence base supporting their ability to turn out voters. The evaluation team expects them to be among the most cost-effective ways to generate net Democratic votes. Donate here.

The recommendations we can make publicly are, by necessity, sparse on details due to the dual-use nature of the research, among other things. If you’re interested in more information on which organizations to donate to and why, please reach out to ee.interventions@gmail.com. The authors of these recommendations would be happy to send more information about alternative donation opportunities and the methodology used to determine the best organizations.

The election as an intervention to improve AI governance is unique in that it’s relatively easy to fundraise among donors not concerned about existential risk. If you know high net-worth individuals who might be interested in a similar (but less AI-oriented) analysis, consider sending them our Substack post on this topic (sample email draft here). Please do reach out if you’re interested in doing this and we’d be happy to help!

Get involved

If you’re interested in getting involved (even for an hour or two a week), check out this collection of opportunities and fill out this form to let us know your background, interests, and availability. Once you fill out this form we can share opportunities with you which we believe are likely to be particularly impactful. These will include opportunities with established organizations, but also options to get involved with projects run by our team (depending on funding there may be paid options).

Appendix

How the model works: Estimating the probability of $10 million flipping the election

  • The model starts with a focus on Pennsylvania and the probability that $10 million could flip the state away from Trump and in favor of Kamala.

  • We conservatively assume that the cost per net Democratic vote is $2000 (many experienced people estimate $1000)[14]. This means that $10 million should increase the number of Harris votes by about 5,000.

  • We assume that election outcomes are a normal distribution with a mean outcome centered on the differential in vote share predicted by the best available polling.

    • The polling we use predicts that Trump will get 46.9% of the vote & Harris 45.2%[15]. Given this, we assume the mean outcome is that Trump wins by a margin of 1.7 points.

  • To calculate the probability that $10 million will shift the election in favor of Harris, we look at the area under the curve between an even split of votes for Harris-Trump and a +4,999 lead for Trump. This returns the probability that an increase in Harris votes by 5,000 flips the PA election.

  • Given that there’s a 35% chance that PA is the tipping point state we need to discount our p(Flip) accordingly.

  • Further, most donation opportunities split their spending between the top 7 swing states.

    • We use Nate Silver’s Voter Power Index numbers to get a sense of the relative effectiveness of money spent in PA vs other swing states. Nate Silver’s VPI numbers imply money spent in non-PA swing states is roughly 2.3 times less effective.

    • We assume 25% of all spending goes to PA and take a weighted average of likelihood that $10 million flips the election in each state.

  • Given the above, our mainline estimate for the probability that $10 million would flip the election is 0.16%.

    • This implies the average individual vote being influenced would have a 1 in 3 million chance of flipping the election[16].

The opportunity cost of a Trump presidency

As we detail below, Trump is likely to actively cause great harm (relative to the regulatory status quo) in key EA cause areas. But even if Trump did not actively cause harm – e.g., by repealing Biden’s executive order on AI, which he vowed to do on day one –, his presidency would still be extremely bad in opportunity cost terms. He would be the center of societal attention for at least four years and the entire political conversation would revolve around his agenda and rhetoric (e.g., attempts to grab power across the branches of government, “revenge” against his political opponents, the border/​immigration, culture war topics, etc.). An enormous amount of progressive resources would be bound up resisting Trump’s agenda instead of improving public policy.

The candidates on some EA cause areas

The following are quick assessments, often based on limited information. Given how little we know of Harris’ current policy views and intentions, much of our assessment assumes she’ll behave similarly to Biden on key issues (which seems likely).

Artificial Intelligence

Candidates’ Strengths and AI

With how quickly AI is advancing, arguably what matters most is not the candidates’ current stances on AI, but how the candidates will respond to new evidence of risks:

  • Who will have the humility to, if needed, change their mind about AI’s risks and pursue international cooperation?

  • Which administration would have the expertise and culture to effectively track and govern AI?

  • Who would govern with societal well-being most in mind, if given power over extremely powerful AI?

For the reasons in our sections on the candidates’ personal character and international relations below, the answers strongly favor Harris.

Positions of the candidates on AI

There are similarities in how both candidates have discussed AI so far, with both candidates discussing global leadership and staying ahead of China as a priority when discussing AI, and pointing out its potential for causing harm. There are also important differences

Trump has expressed some concerns about AI but some of his top supporters seem to be vocal AI accelerationists with a strong interest in shaping AI policy in a second Trump administration.

  • Trump has voiced some concerns about dangers from AI, especially around deepfakes and nuclear concerns.

  • Trump plans to repeal Biden’s executive order on AI safety on day one.

  • Prominent AI catastrophic risk deniers and accelerationists are backing Trump, including a16z leadership and Marc Andreessen, who are supporting the Trump campaign and angling to influence his administration’s AI policy.

  • Elon Musk has given large, and very public, donations in an effort to help Trump win. He has voiced concern about AI risk in the past and may have some ability to influence Trump. While this does mitigate some fears that Trump will throw in his hat entirely with the accelerationist crowd, I have little trust in Musk’s judgment. I’ve been alarmed by what appear to be poorly thought-out statements on how to address AI safety[8] from Musk and have found other recent statements of his concerning.

  • One proposal from a group attached to Trump, the America First Policy Institute, has promoted messaging around limited regulation and reliance on corporate self-governance for AI companies. This proposal also brought up the idea of a ‘Manhattan Project for AI’. It’s unclear the extent to which Trump endorses these views.

Harris tends to focus on present harms, but has expressed some concern about existential risk.

  • Harris has generally put more emphasis on current harms, highlighting that local/​personal harms feel existential to individuals in a November 2023 speech. That said, in the same speech, she acknowledged that AI might “endanger the very existence of humanity”, citing “AI-formulated bioweapons” and “AI-enabled cyberattacks” as particular concerns.

  • At a minimum, Harris seems highly unlikely to reverse the Biden-Harris administration’s previous actions on AI safety. The Biden administration has made impressive progress on AI safety policy, including the establishment of the US AI Safety Institute, securing voluntary commitments on AI safety from many companies, and the 2023 AI Executive Order.

Expected impact of the candidates on AI policy

Competence and integrity of the administration, relationships with other countries and labs, respecting science/​strong arguments, not acting out of self-interest, corruption, and other intangibles strongly point in favor of transformative AI going better under a Harris administration than Trump.

Pandemic response and biosecurity

While we don’t know much about Harris’ opinions about biosecurity, the Biden administration has taken reasonable steps to address the risks from pandemics. Our best guess is that Harris would continue in a similar vein.

The Biden Administration has taken action on biosecurity which Trump has expressed plans to reverse if reelected.

  • In 2023, the Biden White House launched the Office of Pandemic Preparedness and Response Policy. Trump said he would disband this office if reelected, arguing that “we can mobilize,” while with a preparedness office, “[a] lot of the things that you do and a lot of the equipment that you buy is obsolete when you get hit with something.”

  • The Biden Administration is also creating standards and strong financial incentives for many DNA and RNA sequencing companies to screen requests, so they avoid handing hazardous sequences to bioterrorists. This is part of Biden’s Executive Order on AI safety, which Trump has said he would cancel “on day one.”

Trump presided over the beginning of the Covid pandemic and had, at best, a mixed record in terms of handling it. He may appoint known anti-vaccine activist RFK Jr. to a leading role in his administration.

  • There’s a reasonable chance that RFK Jr., an anti-vaccine extremist, will be put into a leading role in a Trump administration. It’s hard to imagine a worse person to have in charge of managing a potential future pandemic or to give responsibility for ensuring countermeasures are in place for future pandemics.

  • According to Wikipedia, “Trump was initially described as optimistic about the country’s response to the pandemic and the threat level the coronavirus disease 2019 presented the public. As the pandemic’s severity escalated in the U.S., Trump repeatedly made false or misleading statements. In contrast, officials within the Trump administration made numerous statements in support of physical distancing measures and business closures.”

  • That said, under Trump, “The federal government managed the development of several vaccines for the virus through Operation Warp Speed in 2020. Distribution of the vaccines was overseen by the Biden administration during 2021, during which time many pandemic measures were ended.”

    • According to a report from the non-partisan Federation of American Scientists, “Operation Warp Speed (OWS) [under Trump] was a public-private partnership that produced COVID-19 vaccines in the unprecedented timeline of less than one year. This unique success among typical government research and development (R&D) programs is attributed to OWS’s strong public-private partnerships, effective coordination, and command leadership structure.”

Global health

A proxy for the candidates’ track records on global health is how much money their administrations asked Congress to approve for global health programs. (How much money Congress actually approved probably depended less on the candidates and more on Congress.) The most recent budget request explains that these programs work “to combat infectious diseases, prevent child and maternal deaths, bolster nutrition, control the HIV/​AIDS epidemic, and build the capacity of partner countries to prevent, detect, and respond to future infectious disease outbreaks to prevent them from becoming national or global emergencies.”

On average, the Biden Administration requested $4.4 billion more per year for global health than the Trump Administration[9]. For reference, GiveWell moved about $1 billion (including funding from Open Philanthropy) in 2022. While it’s unclear how Harris will compare to Biden on international aid spending, it seems highly likely she’ll allocate billions more. If she spends at the same level as Biden (and Trump reverts to his prior spending), getting her into office would lead to ~$16 billion going to international aid that otherwise wouldn’t have.

Given the above model and the assumptions above, each dollar spent on the election likely nets $2.6 dollars of USG global health spending. Assuming a 0.16% chance of flipping the election by spending $10 million, this implies a more than $26 million expected return in terms of foreign aid spending alone. If we assume US government global health spending is 1/​10th as effective as GiveWell top charities, that would imply election giving is about 25% as effective as GiveWell on global health alone. We think the fact that election spending is in the same ballpark as GiveWell in terms of effectiveness in this domain alone is evidence election donations are likely worthwhile.

Climate change

We don’t yet know the specifics of Harris’ climate platform, but she has long prioritized climate change.

  • Her record includes co-sponsoring the Green New Deal, creating a unit to target polluters in poor neighborhoods as the San Francisco district attorney, and bringing charges against major California polluters as attorney general.

  • Her campaign spokesperson said Harris will follow Biden’s lead. A key example of Biden’s climate efforts was the 2022 Inflation Reduction Act, which allocated hundreds of billions in funding and tax incentives to help both the public and private sector transition to renewable energy and otherwise reduce emissions.

Trump is clearly not going to take positive action on climate change and is likely to roll back much of the progress made under Biden

  • Analysis from Carbon Brief shows that a second Trump administration could lead to an estimated extra 4 billion tonnes of CO2 emissions by 2030 than a second Biden term. This is comparable to the annual CO2 output of the European Union and Japan

    • Put another way, the extra 4 gigatons of CO2 from a second Trump term would negate – twice over – all of the savings from deploying wind, solar and other clean technologies around the world over the past five years.

  • Looking beyond 2030, and assuming no further policy changes, a Trump win could yield an extra 27 billion tonnes of CO2 emissions by 2050, compared with policies that would be implemented under a second Biden administration, the analysis shows.

  • Trump has actively solicited money from oil executives, apparently in exchange for major actions to cut pollution regulations, reduce subsidies for electric cars, and expand offshore drilling permits.

Nuclear Risk

Trump probably exacerbated nuclear risk in his first term and would likely do so again in a second term.

  • Kingston Reif, a missile expert at the Arms Control Association in Washington, D.C., summarized Trump’s nuclear initiatives: “The Trump administration’s nuclear legacy is one of failure,” Reif said. “The administration inherited several nuclear challenges, to be sure, but it has made nearly all of them worse. Trump became the first president since the 1960s not to negotiate any new nuclear arms-control agreement. Instead, he did the very opposite—loosened controls, encouraged proliferation and, as a result, is “the first post-Cold War president not to reduce the size of the nuclear warhead stockpile.”

  • Two years after Trump withdrew from the Iran nuclear deal, Tehran has cut in half the time it would need to produce enough weapons-grade fuel for a nuclear bomb.

  • Trump is generally unstable and prone to making risky bets. He has not demonstrated a solid understanding of nuclear policy nor the serious risks of using nuclear weapons. In fact, his former Secretary of Homeland Security claims that Trump, “cavalierly discussed the idea of using a nuclear weapon against North Korea, saying that if he took such an action, the administration could blame someone else for it to absolve itself of responsibility”

Harris seems likely to continue with the status quo on US nuclear policy, perhaps making marginal progress on increasing safety.

  • As for Harris, we again know fairly little about how she would differ from the Biden administration on nuclear policy. It seems likely she would essentially continue with the status quo but be far more likely to handle a crisis gracefully.

Farm animal welfare

Harris appears to be a strong advocate of animal welfare, even addressing some issues for farmed animal welfare.

  • As California’s Attorney General Harris, “defended a series of pioneering animal protection laws approved in the state. She continuously defended California’s ban on foie gras sales after producers launched a series of challenges in the courts. She also took the right side—and the winning side—of multiple separate challenges to Prop 2 and AB 1437, California’s improved farm animal welfare laws.” She also, “defended a law which banned killing pigs for food if they were too sick or injured to move”.

  • In her 2020 primary campaign she also mentioned she would support changing dietary guidelines to incentivize a reduction in the consumption of red meat, though she cites climate change rather than welfare concerns.

  • On the other hand Walz, who has done some work on agriculture in the past, is less clearly favorable for animal welfare concerns.

We know little about Trump’s views on animal welfare but it seems unlikely we’ll see any positive developments on farm animal welfare from his administration.

  • As president, Trump signed a bill making various forms of animal cruelty federal crimes, but the bill exempted customary agricultural practices. In 2024, Florida governor Ron DeSantis signed Senate Bill 1084 into law, making Florida the first U.S. state to ban the sale of cell-based meat (also known as “cultivated” or “lab-grown” meat). This is part of a broader MAGA anti-lab grown meat polarization, with efforts to reduce factory farmed production

US-China & international relations

The next administration will inherit a tense international situation which could easily get worse. AI has the potential to add further challenges.

  • Back in 2021, a panel of superforecasters estimated the likelihood of a China-Taiwan conflict over the next five years at 14% (8-23% CI). Metaculus (currently) assigns a 25% probability to a full-scale Chinese invasion of Taiwan by 2030.

    • Over the last few years, purges have become far more common in China, and the political situation more volatile.

  • In short timeline worlds, international coordination and diplomacy might be extremely important, especially with China.

Harris, assuming she acts similarly to Biden, is likely to perform well when it comes to international relations.

  • A Harris administration seems much more likely to successfully navigate important relationships with China, the EU, Taiwan, South Korea, Japan, the Netherlands, and top labs than a Trump administration. For instance, the Biden administration seems to have successfully navigated the complex issue of instituting semiconductor export controls on China.

  • “Harris, like Biden, has emphasized the need to reduce Western economic reliance on China without strangling trade between the two nations or potentially hurting the U.S. economy through steep tariffs of the sort former President Donald Trump espouses. Harris also echoes Biden’s stress on engaging in competition with China but not conflict.”

  • Biden and Harris have credibly committed to help Taiwan. Trump appears much more isolationist and less likely to intervene, which might make China more likely to invade.

  • Biden prepared military-military communication between China and the US that Republicans have criticized, as well as an APEC summit.

While Trump had some foreign policy successes, his second term would likely degrade US alliances and could open up new security problems for the US and its allies around the world.

  • The Trump Administration showed near-universal disdain for international coordination, as evidenced by pulling out of the Iran Nuclear Deal, the Paris Climate Accords, and the TPP. He appears to have a poor understanding of the importance of stable liberal international order, and the role the US plays in maintaining it.

    • On the other hand, Trump did manage to successfully pressure some NATO countries to raise their military spending closer to the 2% threshold suggested in the treaty. He has, however, discussed withdrawing the US from NATO if re-elected (though he seems unlikely to do so).

    • Both Trump and Vance have threatened to stop all US military aid for Ukraine. (Vance said: “I gotta be honest with you, I don’t really care what happens to Ukraine one way or another.”) This would likely embolden the Russian dictator as well as undemocratic regimes all around the world.

  • Trump seems to have little regard for sending a consistent message around which countries the US is or isn’t prepared to defend. Misunderstandings on this topic have caused wars in the past (e.g., Persian Gulf War). This seems especially important in the case of Taiwan, where he has seemingly backed away from traditional US positions.

  • Trump has proposed a 10% universal import tariff (including allies), and a 60% tariff on all Chinese goods. This would likely hurt international diplomacy and the American economy.

  • Trump deserves some credit for pivoting the US toward a more aggressive stance on China and paving the way for the first round of export controls on its semiconductor industry (though these were devised and implemented by the Biden administration, whose stance on China would likely have been similarly aggressive without Trump’s preparatory moves).

  • Overall, we expect a Trump presidency to destabilize the global security environment. If the foremost military superpower withdraws from its global policing role or sends mixed signals about its willingness to protect allies, conflicts are likely to break out at significantly greater rate. Many undemocratic regimes are likely to pursue territorial and other ambitions if the US check on their power weakens. Along with direct harms, a single war relevant to US interests could absorb much of the nation’s political attention and vast material resources for months or years. This is particularly dangerous during times as technologically critical as ours (see the section on AI above).

  1. ^

    We’re a group of researchers and activists who dropped our other work (mostly related to AI governance) or are volunteering to help beat Trump this year. We’ve built a strong network and contributed to a few high-leverage projects we believe are impacting the election. Some team members would rather not be named publicly, mostly due to concerns about a public partisan record, especially under a Trump administration.

  2. ^

    If you already read that piece the most important additions are found in the sections: ‘Should you donate to the election?’ & ‘Assessing the candidates...’

  3. ^

    “What scared Kelly even more than the tweets was the fact that behind closed doors in the Oval Office, Trump continued to talk as if he wanted to go to war. He cavalierly discussed the idea of using a nuclear weapon against North Korea, saying that if he took such an action, the administration could blame someone else for it to absolve itself of responsibility”. NBC News

  4. ^

    The concern here is similar to that raised in Reducing long-term risks from malevolent actors

  5. ^

    I expect people will disagree on whether this is good or bad.

  6. ^

    In a survey run in March of this year, the mean prediction of 26 AI safety and governance experts was that the expected value of the future given a Biden victory would be 25% greater than given a Trump victory (the median answer was 9%). We suspect respondents would give similar answers for a Harris administration today.

  7. ^

    The reason US elections are generally so close is because the electoral college system means that only a few states with evenly split political demographics actually matter for the election. That means the entire presidential race affecting 333 million people comes down to just a few states with a total population of around 30 million voters.

  8. ^

    Experts we’ve talked to estimate that the cost per net vote was somewhere between $400 & $10,000 under Biden (post-debate, getting votes for Biden was rough). Under Harris cost per net vote was likely between $700-2000 a month ago but that has gone up to perhaps $1000-2500 now.

  9. ^

    This was relayed to us by someone from one of the largest progressive political organizations in the country. You’ll need to have access to the Analyst Institute in order to see it but, if you do, you can find some relevant research here.

  10. ^

    You can find the relevant information here, though again only if you have access through the Analyst Institute.

  11. ^

    This was very clearly the case in Hillary’s 2016 campaign. Obama managed this well in 2008 but his team struggled with it in 2012.

  12. ^

    P(doom from AGI before 2032 | Harris victory & AGI in 2024-32)

  13. ^

    P(doom from AGI before 2032 | Trump victory & AGI in 2024-32)

  14. ^

    This is a fairly conservative estimate that attempts to take into account that many funding gaps are likely to be filled. A reasonable first-guess estimate would be more like $1000 but that doesn’t account for the fact that large funders may jump in to fill the gap if not enough donations are received.

  15. ^

    These are from a top pollster but are now at least 3 weeks old (e.g. before the debate). Using updated numbers slightly increases likely effect size, but not by much.

  16. ^

    This estimate is similar, if somewhat more optimistic, than a recent EA Forum post on the topic. They estimated generic swing state votes as having a 1 in 6 million chance of flipping the election and Pennsylvania votes as 1 in 3 million.

Alex Lintz24 Sep 2024 19:32 UTC
23 points
2 comments19 min readLW link