Social and Governance Implications of Improved Data Efficiency

Abstract

Many researchers work on improving the data efficiency of machine learning. What would happen if they succeed? This paper explores the social-economic impact of increased data efficiency. Specifically, we examine the intuition that data efficiency will erode the barriers to entry protecting incumbent data-rich AI firms, exposing them to more competition from data-poor firms. We find that this intuition is only partially correct: data efficiency makes it easier to create ML applications, but large AI firms may have more to gain from higher performing AI systems. Further, we find that the effect on privacy, data markets, robustness, and misuse are complex. For example, while it seems intuitive that misuse risk would increase along with data efficiency -as more actors gain access to any level of capability -the net effect crucially depends on how much defensive measures are improved. More investigation into data efficiency, as well as research into the "AI production function", will be key to understanding the development of the AI industry and its societal impacts.

A Causal Framework for AI Regulation and Auditing

“…Vast leaps in the absolute capabilities of AI systems in recent years have been, to a large extent, driven by similarly vast increases in the size of systems, the amount of data, and the amount of computation used to train them [60][61][62]. There has also been significant algorithmic progress, which reduces the amount of compute needed to obtain the same performance [63,64]. This means that even if we held compute constant, algorithmic progress would mean that effective compute would continue to increase, where effective compute is the product of the amount of compute used and the efficiency of how that compute is used.…”

The Societal Implications of Deep Reinforcement Learning

“…Advances in data efficiency and simulation quality would make it easier for those without access to large amounts of computing power and data to make use of DRL (Tucker et al, 2020). This could make it easier for small groups to misuse DRL capabilities for malicious purposes.…”

Artificial Intelligence in Local Government: Enabling Artificial Intelligence for Good Governance in UK Local Authorities

“…[19] The available data on IT contracting in local government suggests that between 2012 and the present, there has been a concentration in the supplier market towards large suppliers. This, combined with the benefits of technological advances, such as data efficiency gains, which disproportionately accrue to large AI firms, [20] may increase supplier strength. An example of this strength is Nesta's 2 finding that "where services or IT are outsourced, a public sector body may even find that it cannot access the data relating to its own service or must pay an additional fee".…”