Acting Comptroller Hsu Calls for "Responsible" Approach to Tokenization/AI Innovation
Commentary by Steven Lofchie
Acting Comptroller of the Currency Michael J. Hsu called for a "responsible" approach to innovation in tokenization and artificial intelligence ("AI") that progresses in "tightly controlled stages" to manage identified risks.
In his remarks before the American Bankers Association Risk and Compliance Conference, Mr. Hsu emphasized the need for a principled approach that (i) innovates in stages, (ii) "build[s] the brakes while building the engine" and (iii) engages with regulators throughout each stage. With regard to developments of tokenization and AI, Mr. Hsu examined the following ways to address potential risks and benefits.
Tokenization
Mr. Hsu called a central premise behind public blockchains - "trustlessness" - a "key design flaw." He warned against the "trilemma" created by (i) decentralization, (ii) security and (iii) scale. He said the "mechanisms" used in a "trustless" or "trust-minimized" blockchain are inefficient (requiring "proof of work" or "proof of stake"), attractive to those engaged in illicit finance, and make full compliance with AML regulations very difficult to achieve. By contrast, he touted the security and scale offered by centralized, "trusted" blockchains. which make settlements more efficient by minimizing lag times and associated costs and risks.
Mr. Hsu called for further developments in legal frameworks with respect to the tokenization of real-world assets and liabilities, including addressing ownership and property rights. Mr. Hsu said that better frameworks will inform related risk management practices and the controls needed to conduct transactions and to support innovation. He added that "trust" blockchains, as opposed to public blockchains, are "better positioned" to foster such innovation and create real values for businesses through settlement efficiencies.
Artificial Intelligence
Mr. Hsu raised concern that AI systems, generally based on neural networks, are unlike most software programs, and as a result, their outputs are often unpredictable. He said that many banks rely on third parties to develop their AI capabilities, which raises significant concern about where the responsibility lies regarding an AI system’s performance and results. Mr. Hsu also raised concerns over (i) bias and discrimination in training data, (ii) fraud and the spread of misinformation, ranging from check fraud to synthetic identity and media fraud, and (iii) the creation and dissemination of false information via social media.
Commentary
Ultimately, Mr. Hsu's approach would kill innovation. One can say "innovate responsibly," and it sounds good, but it is not meaningful unless there are real actions that one can take to achieve that end.
It is true that AI currently leads to unpredictable results and the creators do not fully understand why the programs reach certain conclusions. That said, if their use is prohibited until those issues are solved (assuming they can ever be solved), the problems won't be solved because no one will find it worthwhile to work on the problem.
As to tokenization, it is very possible that permissioned blockchains will be more successful than unpermissioned. But in the practical world, there would never have been such interest in permissioned blockchains had not innovators developed unpermissioned blockchains.
The attempt to eliminate risk means stifling change that enables progress. Stagnation is the risk that over-regulation imposes.