Epistemic status: Spent a few hours reading and thinking about the report. I buy a weak variant of its thesis (if you want to have influence, you better have a good model of the institutions you’re working within / with), and am sympathetic to but not fully convinced of the strong variant (‘in order to reliably affect decision-making, you must yourself be the decision-maker’).
TLDR
This is a review of a report by Samo Burja and Zachary Lerangis on how Leo Szilard and other scientists involved in the Manhattan project failed to influence decision-making around the nuclear bomb, including the choice to use it in Japan and nuclear proliferation after the war. There are strong parallels to X-risk governance, in particular around AI. To quote the report,
The implication of these historical outcomes is that in order to reliably affect decision-making, you must yourself be the decision-maker. Prestige, access to decision-makers, relevant expertise, and cogent reasoning are not sufficient; even with all these you are liable to be ignored.
The Report
Some of the scientists involved in the building of the atomic bomb (primarily Leo Szilard, author of the famous Einstein-Szilard letter, and Niels Bohr) tried to influence decision-making around nuclear governance, including
Increasing the influence of scientists on how the Manhattan project was led
Convincing Truman not to use the bomb in Japan
Convincing Truman that the Soviets would soon develop their own nuclear weapons (which US gov leadership at the time did not believe)
Establishing international governance of nuclear weapons / arms control regime.
Both Szilard and Bohr had access to powerful people within the US and UK governments, including FDR and Churchill. They nevertheless failed at achieving their goals.
I’ll briefly excerpt two of the four examples from the report:
Bohr talks to Churchill and FDR to convince them to coordinate with the Soviets
In 1944, Niels Bohr, one of the great, Nobel Prize-winning physicists of the 20th century and whose reputation was very well established at the time, was very concerned about the threat of nuclear war and a nuclear arms race after the inevitable end of the war. He wanted the Allies to cooperate with the Soviets on the control of nuclear technology so as to avoid a nuclear arms race or all out war. Towards this end, he wanted the Allies to tell the Soviets about the existence of the atomic bomb project, in the hopes that this would build goodwill and foster future coordination.
[...] Off the bat, Churchill was aggressively opposed to any cooperation with the Russians, or even to telling them of the existence of the bomb project. The meeting went nowhere.
[...] FDR told Bohr that he read Bohr’s memo, that he agrees that the Soviets ought to be approached, and that he is optimistic about the prospects of cooperation based on his assessment of Soviet leader Joseph Stalin. He also said that he was optimistic that Churchill would come around [...]
The next month, FDR and Churchill convened in Hyde Park. They had a short, secret agreement written up on the nuclear question that included the following resolutions: the nuclear program will continue to be kept absolutely secret; the US and Britain will continue to develop nuclear weapons after the end of the war; Professor Bohr should be put under surveillance.
Leo Szilard attempts to prevent the use of the bomb against Japan
[Szilard] went to Washington in May 1945, while the Interim Committee—the committee commissioned to make recommendations about the use of the atomic bomb—was meeting. There he met with Oppenheimer, one of four scientists on the scientific advisory panel to the Interim Committee, and tried to convince him that the bomb shouldn’t be used against Japanese cities. Oppenheimer disagreed. He may have contacted others involved in the committee, but we have not found a record of this. The scientific panel eventually recommended immediate use of the bomb against Japanese cities.
Soon afterwards, he drafted a petition to the President [...] There is a good chance that the petition never reached Truman.
Despite Szilard’s prominence within the scientific hierarchy of the Manhattan Project from day one, he was never able to gain any meaningful standing in the state hierarchy
My Takeaway
The opinions of leading researchers in a field do not (by themselves) determine how the technology they develop will be deployed.
There are many AI researchers who take AI X-risk seriously, including researchers in leading university and industry labs. By itself, this is not enough, if X-risk conscious people are not in position to actually make the crucial decisions (e.g. whether or not to pour another 100x of compute into scaling up a seemingly safe model, or to decide what publication norms should be like).
More generally, in order to have influence it is important to have a good understanding of the institutions one is working within and a concrete model of how exactly one will affect decisionmaking.
Thanks to Ben Landau-Taylor for a good discussion and for pointing me to this report, and to Sam Clarke for feedback and encouraging me to write this.
Linkpost: The Scientists, the Statesmen, and the Bomb
Link post
Epistemic status: Spent a few hours reading and thinking about the report. I buy a weak variant of its thesis (if you want to have influence, you better have a good model of the institutions you’re working within / with), and am sympathetic to but not fully convinced of the strong variant (‘in order to reliably affect decision-making, you must yourself be the decision-maker’).
TLDR
This is a review of a report by Samo Burja and Zachary Lerangis on how Leo Szilard and other scientists involved in the Manhattan project failed to influence decision-making around the nuclear bomb, including the choice to use it in Japan and nuclear proliferation after the war. There are strong parallels to X-risk governance, in particular around AI. To quote the report,
The Report
Some of the scientists involved in the building of the atomic bomb (primarily Leo Szilard, author of the famous Einstein-Szilard letter, and Niels Bohr) tried to influence decision-making around nuclear governance, including
Increasing the influence of scientists on how the Manhattan project was led
Convincing Truman not to use the bomb in Japan
Convincing Truman that the Soviets would soon develop their own nuclear weapons (which US gov leadership at the time did not believe)
Establishing international governance of nuclear weapons / arms control regime.
Both Szilard and Bohr had access to powerful people within the US and UK governments, including FDR and Churchill. They nevertheless failed at achieving their goals.
I’ll briefly excerpt two of the four examples from the report:
Bohr talks to Churchill and FDR to convince them to coordinate with the Soviets
Leo Szilard attempts to prevent the use of the bomb against Japan
My Takeaway
The opinions of leading researchers in a field do not (by themselves) determine how the technology they develop will be deployed.
There are many AI researchers who take AI X-risk seriously, including researchers in leading university and industry labs. By itself, this is not enough, if X-risk conscious people are not in position to actually make the crucial decisions (e.g. whether or not to pour another 100x of compute into scaling up a seemingly safe model, or to decide what publication norms should be like).
More generally, in order to have influence it is important to have a good understanding of the institutions one is working within and a concrete model of how exactly one will affect decisionmaking.
Thanks to Ben Landau-Taylor for a good discussion and for pointing me to this report, and to Sam Clarke for feedback and encouraging me to write this.