Joshua S. Gans

Joshua S. Gans

Machine learning systems embed preferences either in training losses or through post-processing of calibrated predictions. Applying information design methods from Strack and Yang (2024), this paper provides decision problem agnostic conditions under which separation training preference free and applying preferences ex post is optimal. Unlike prior work that requires specifying downstream objectives, the welfare results here apply uniformly across decision problems. The key primitive is a diminishing-value-of-information condition: relative to a fixed (normalised) preference-free loss, preference embedding makes informativeness less valuable at the margin, inducing a mean-preserving contraction of learned posteriors. Because the value of information is convex in beliefs, preference-free training weakly dominates for any expected utility decision problem. This provides theoretical foundations for modular AI pipelines that learn calibrated probabilities and implement asymmetric costs through downstream decision rules. However, separation requires users to implement optimal decision rules. When cognitive constraints bind, as documented in human AI decision-making, preference embedding can dominate by automating threshold computation. These results provide design guidance: preserve optionality through post-processing when objectives may shift; embed preferences when decision-stage frictions dominate.

Neil Gandal, Joshua S. Gans

This paper extends the blockchain sustainability framework of Budish (2018) to consider proof of stake (in addition to proof of work) consensus mechanisms and permissioned (where the number of nodes are fixed) networks. It is demonstrated that an economically sustainable network will involve the same cost regardless of whether it is proof of work or proof of stake although in the later the cost will take the form of illiquid financial resources. In addition, it is shown that regulating the number of nodes (as in a permissioned network) does not lead to additional cost savings that cannot otherwise be achieved via a setting of block rewards in a permissionless (i.e., free entry) network. This suggests that permissioned networks will not be able to economize on costs relative to permissionless networks.

Joshua S. Gans

Here we examine the paperclip apocalypse concern for artificial general intelligence (or AGI) whereby a superintelligent AI with a simple goal (ie., producing paperclips) accumulates power so that all resources are devoted towards that simple goal and are unavailable for any other use. We provide conditions under which a paper apocalypse can arise but also show that, under certain architectures for recursive self-improvement of AIs, that a paperclip AI may refrain from allowing power capabilities to be developed. The reason is that such developments pose the same control problem for the AI as they do for humans (over AIs) and hence, threaten to deprive it of resources for its primary goal.