Seng (2021)
Contents
Source Details
| Seng (2021) | |
| Title: | Copyrighting copywrongs: an empirical analysis of errors with automated DMCA takedown notices |
| Author(s): | Seng, D. |
| Year: | 2021 |
| Citation: | Seng, D. (2021) Copyrighting copywrongs: an empirical analysis of errors with automated DMCA takedown notices. Santa Clara High Technology LJ, 37(2) |
| Link(s): | Open Access |
| Key Related Studies: | |
| Discipline: | |
| Linked by: | |
| About the Data | |
| Data Description: | The study draws on takedown requests catalogued in two datasets: • The Chilling effects repository, filtering for notices submitted to Google |
| Data Type: | Primary and Secondary data |
| Secondary Data Sources: | |
| Data Collection Methods: | |
| Data Analysis Methods: | |
| Industry(ies): | |
| Country(ies): | |
| Cross Country Study?: | No |
| Comparative Study?: | No |
| Literature review?: | No |
| Government or policy study?: | No |
| Time Period(s) of Collection: |
|
| Funder(s): |
|
Abstract
“Under the Digital Millennium Copyright Act (DMCA), reporters issuing takedown notices are required to identify the infringed work and the infringing material and provide their contact information (functional formalities), attest to the accuracy of such information and their authority to act on behalf of the copyright owner, and sign the notices (non-functional formalities). Online service providers will evaluate such notices for compliance with these DMCA formalities before acting on them. This paper seeks to answer questions about the quality of takedown notices, especially those generated by automated systems, which are increasingly being used by copyright owners to detect instances of online infringement and issue takedown notices on their behalf. After parsing three million takedown notices and more than eighty million takedown complaints served on Google between 2011 and 2015, this paper analyzes each notice for errors. This paper finds that almost all notices comply with the non-functional formalities. However, at least 5.5% of all takedown notices between 2011 and 2015 fail to comply with the functional formalities in that they are missing copyright work descriptions. In addition, at least 9.8% of the takedown notices exhibit have empty takedown requests, misidentify the infringing site or provide inactive URIs as takedown requests. To ensure that the takedown system remains fast, efficient and error-free, this paper proposes to strengthen the attestation requirements of notices, to require reporters to validate all submitted takedown complaints and requests, and to subject recalcitrant reporters to the “slow lane” of a two-tier system for processing takedown notices. This methodology reflects the use of accountability metrics in the design of automated systems and suggests a verifiable response to address concerns pertaining to the use of systems that supplant human decision making”
Main Results of the Study
• Overall, the volume of takedown requests for online content is increasing (est. almost double every year, approx. 249%). Based on self-reporting from Microsoft, Google and Twitter, these requests have a high degree of success (latest 99.77%, 98% and 74% respectively). The number of complaints per notice has also been steadily increasing over time (from 1.71 complaint per notice in 2011 to 26.14 per notice in 2015).
• Errors in non-functional formalities of a notice (e.g. electronic signatures, completion of statements of accuracy) are low (0.02% error rate) despite the increase in form-based takedown notices. This may be in part because automated forms oblige the completion of certain parts of the notice, minimising the prevalence of such errors.
• Errors in functional formalities of a notice are higher, with an error rate of up to 5.5%. Errors include e.g. no copyright work description, or no valid URI description. Such errors exist even where the notices are submitted by trusted users, with the most prevalent sender of ‘empty’ requests being BPI (28.6% of all their notices).
• Whilst the scale of substantive errors (e.g. misidentified rightsholder, no allegation of wrongdoing) is relatively small, they are rarely rejected. For example, an ex-Microsoft employee erroneously claimed ownership, and sent notices in respect of, 259,731 pieces of online content, of which only 2,950 (1.14%) were rejected.
Policy Implications as Stated By Author
The study finds that automated notice and takedown systems increase the risk of substantive errors. The author proposes three changes to the DMCA to minimise this risk:
• Create a new condition that reporters will act under penalty of perjury to attest to accuracy and good faith of the notice;
• Require the reporter to verify their submitted requests, and;
• Penalise erroneous takedown requests by using an accountability metric (e.g. by ‘slowing down’ the takedown process for repeated offenders).
Coverage of Study
Datasets
| Sample size: | 56,991,045 |
| Level of aggregation: | Takedown requests |
| Period of material under study: | 31 December 2012 |
| Sample size: | 88.0 million"million" can not be assigned to a declared number type with value 88. |
| Level of aggregation: | Takedown complaints |
| Period of material under study: | 1 January 2011 – 31 December 2015 |
| Sample size: | 1.17 billion"billion" can not be assigned to a declared number type with value 1.17. |
| Level of aggregation: | Takedown requests |
| Period of material under study: | 1 January 2011 – 31 December 2015 |
