What criticisms have been made of Bostrom's Open Global Investment model?
The Open Global Investment (OGI) model has received three main types of criticism: it won’t work; even if it does, it won’t do enough; and it will have bad side-effects. Bostrom has given responses to each of these criticisms, emphasizing that OGI is meant as a politically feasible, near-term proposal for AGI governance. In Bostrom’s view, OGI should be compared with other AGI governance proposals within the Overton window, such as a Manhattan Project for AGI, or the current default: nothing.
“It won’t work”
One of the assumptions of the Open Global Investment model — that companies will not be nationalized and that investors and corporations will enjoy strong property rights — does not hold across the globe. E.g., the Chinese government limits foreign financial investment and exerts significant control over domestic corporations. Even the US government may not be able to credibly make such commitments.
To this point, Bostrom responds that the lack of suitable laws and norms for open global investment is not a criticism of the desirability of the model, but of its probability of being implemented. He also notes that high investor demand for US financial products, such as bonds, implies that investors are still confident that the US will continue to respect financial commitments.
“It won’t do enough”
Other critiques note that the OGI model does not deal with rogue AI or misuse, change the status quo, or build toward a pause.
Bostrom replies that the OGI model can be extended to handle most of these issues. Extending it is part of the plan, as OGI is a governance model for the near to medium term. The exception is pausing AI, which would be a better fit for a different proposal, one which is not intended as an alternate path to governance of AGI.
“It will have bad side-effects”
Wei Dai argues that the OGI model rewards lab leaders and early investors in AGI companies for increasing existential risk relative to the counterfactual, which is not something we want to incentivize.
Bostrom responds that we currently don’t tax people for increasing existential risk, and it might be bad to retroactively punish people financially for acting on the incentives that exist in our current system. He suggests moral praise and condemnation are more appropriate tools for this. He does say increased taxes would be compatible with the OGI model, though not to the extreme of taxes nearing 100%.