Discussion: (0 comments)
There are no comments available.
A public policy blog from AEI
The latest on technology policy from AEI, published daily.
Since Adam and Eve first exchanged their innocence for an apple (or whatever creation myth performs the same role in other traditions), the distribution of the gains of trade have hinged on the distribution of information (or lack of it) relevant to the transaction. The serpent knew things about the apple that Adam and Eve didn’t and things about them that made them better targets (delivered him higher surplus) than the alternatives. So Adam and Eve paid too high a price for it (gained less than the price paid), and consumer welfare in paradise was reduced — with the long-term consequences being borne by all their heirs and successors.
An essential part of the act of trading itself reveals information about the participants that can be stored and recalled to inform subsequent transactions — both with the same transactors and the same categories of transactors. Thus, honest traders can develop good reputations, and welfare-destroying trades can be avoided. But equally, individual and group foibles can be remembered and exploited to enable more of the gains to be directed to the exploiter than would be the case if this information was not known.
While such behavior will leave some parties worse off than if the exploitation had not occurred, it may lead to higher total welfare if it enables trade by those who could not participate if the exploitation had not occurred (e.g. third-degree price discrimination, such as versions of software sold to professional customers at high prices and students at low prices). Furthermore, actual goods and services need not be exchanged for the information to be generated or collected. Simply interacting is sufficient.
Information has always challenged regulators
For as long as they can be cost-effectively monitored and enforced, various forms of regulations requiring disclosure of both the known characteristics of the products and services exchanged and the terms of trade (including compensation and restitution if terms are violated) have been imposed to enable informed transacting decisions by all participants. Restrictions have also been placed on the collection and harmful use of information (e.g., privacy laws). However, such regulations have little effect if either the existence or the nature of the relevant information and its use can be concealed from both the uninformed party and the regulator.
In this context, the capture and use of information revealed from internet-enabled interactions has come under increasing regulatory scrutiny in the past two decades. Paper records are costly to create and store, placing limits on the amount of data collected. However, it is relatively easy to observe what they contain, thereby facilitating lower-cost regulatory enforcement. By contrast, electronic records are comparatively cheap to create and store but much harder to cost-effectively observe and verify, due to both their volume and the need for specialized software, which is typically proprietary to the party doing the collecting. These difficulties have been exacerbated by massive reductions in the cost of data transport, storage and processing, and technological changes such as cloud computing that have enabled data to be stored and processed in locations outside the control of (typically) state-specific regulators. But the fundamental fact is that electronic records of electronic transactions impose further layers of opacity between the transacting parties. It becomes even harder for the parties (or their regulatory protectors) to observe the data collected or its use in subsequent interactions. This necessarily increases the risks that information asymmetries will be exploited in ways that may disproportionately harm some transactors. At the same time, new opportunities to confer benefits on others that might not otherwise have been realized are made possible.
Big data and artificial intelligence: New scale and scope
On this trajectory, the development of big data and its doppelganger artificial intelligence (AI) does not necessarily pose any different types of challenges to regulators and policymakers over and above those posed by replacing paper with electronic records. Arguably, their effect, like that of decreasing costs of data transport, storage, and processing, is one of scale rather than type.
Big data is differentiated from other electronic data by its volume, variety, velocity (speed of collection and processing) and veracity (accuracy and fitness for analytical purpose) — albeit that it may require new approaches to computer networking to deal with large volumes and algorithms capable of interpreting and processing different data (e.g., images in addition to traditional alphanumeric characters). AI, in contrast, focuses primarily on the algorithms processing the data and, specifically, the ways in which they can use the outputs of historic analysis to improve the models used to conduct further analysis. Arguably, this automates the processes by which humans have always analyzed the output of their models and adjusted either the model or the data input into subsequent analyses to incrementally improve their value for their specific purposes.
But so far, the body of regulatory knowledge is not well stocked with tools to deal with even simple issues of increasing data opacity, such as levying and collecting sales taxes in internet transactions across state borders.
Conclusions for policy and regulatory processes
While some policymakers and regulators seek to minimize the potential harms these new technologies may pose, the current store of knowledge itself requires enhancement. Information asymmetries have always existed and have always enabled both good and harmful outcomes to arise. But the state of formal modeling of the circumstances in which they apply — which will inform future regulation — continues to evolve. Almost always, it lags behind the emergence of the patterns of interaction that cause concern, because rational humans cannot foresee all outcomes. It also bears remembering that those creating and implementing new technologies also may not be fully aware of what they are unleashing.
So if there is a principle to guide the future regulation of big data and AI, it is to focus on first understanding information asymmetries and how they affect the distribution of the gains, rather than the technologies that they are associated with. There is plenty in the history of economic modeling to guide this thinking, but there is also still much to be learned.
There are no comments available.
1789 Massachusetts Avenue, NW, Washington, DC 20036
© 2019 American Enterprise Institute