In his new article, Hacohen discusses algorithmic auditing as an important tool for responsible AI promotion. The article introduces “user-based algorithmic auditing” as a novel approach, analyzing obstacles that prevent such auditing from accelerating and proposing a few suggestions to promote and incentivize user-based auditing.
In his article, Hacohen discusses algorithmic auditing as a key tool for policymakers to hold digital platforms like Meta, Google, and Amazon accountable for irresponsible algorithm practices. As these platforms use AI and cloud computing to personalize services, they can benefit users but also pose significant risks of algorithmic bias and manipulation. All major jurisdictions – the United States, the EU, and Britain – are currently devising new algorithmic auditing regulations and targeting them at big digital platforms. Audits can be executed by different actors – the platforms themselves (first-party audits), external auditors solicited by the audited company (second-party audits), or external audits conducted by independent auditors who have no formal relationship with the audited company (third-party audits). Traditional first-party audits, self-assessed by platforms, are flawed by design due to the misalignment between the platforms’ commercial interests and the social interests. Third-party audits, while more independent, still rely on information controlled by the platforms and thus are difficult to accomplish. In his article, Hacohen suggests a focus on algorithmic user-based auditing, where users lead or assist in the auditing process, ensuring more impartial assessments. This method can corroborate platform-provided information and offer a balanced view of potential harms. Hacohen distinguishes between user-assisted and user-driven audits and supervised and unsupervised methods. For each, Hacohen elaborates on the current obstacles that stand in the way of effective large-scale user-based audits. These obstacles include legal liability risks, high costs, lack of motivation among users, and collective action problems. Lastly, Hacohen set forth a few policy interventions to address these challenges and promote effective user-based auditing. Hacohen’s proposals include safe harbor legislation protecting third-party auditors from liability, governmental funding for auditors, financial incentives for users participating in data collection and reporting, and government-supported data trusts. The conclusion emphasizes the importance of reliable audits for digital platforms to ensure societal well-being