Engstrom and Feamster (2017)
|Engstrom and Feamster (2017)|
|Title:||The Limits of Filtering: A Look at the Functionality & Shortcomings of Content Detection Tools|
|Author(s):||Engstrom, E., Feamster, N.|
|Citation:||Engstrom, E. And Feamster, N. (2017) The Limits of Filtering: A Look at the Functionality & Shortcomings of Content Detection Tools. Available: https://www.engine.is/the-limits-of-filtering (last accessed 23 May 2019)|
|Key Related Studies:|
|About the Data|
|Data Description:||The report includes a technical analysis of the basic functionalities of the most commonly-employed filtering technologies, including: content metadata search; hash-based identification, and; audio and video fingerprinting. The report also includes a case study of Echoprint, an open-source audio fingerprinting tool employed by Spotify.|
|Data Type:||Secondary data|
|Secondary Data Sources:|
|Data Collection Methods:|
|Data Analysis Methods:|
|Cross Country Study?:||No|
|Government or policy study?:||No|
|Time Period(s) of Collection:|
“Nearly twenty years ago, Congress passed the Digital Millennium Copyright Act (DMCA), establishing copyright rules specifically tailored for the online world. Although the DMCA has succeeded admirably in fostering the growth of the internet, some copyright industry representatives have advocated for a rewrite of the law to force OSPs to implement content filtering technologies as a prerequisite to obtain the protections of the DMCA's safe harbors. In evaluating these calls for mandatory filtering, policymakers should consider how such filtering tools actually work in practice and the impact they are likely to have on infringement, startup activity, and creative production.
This paper, co-authored by Engine's Evan Engstrom and Princeton University's Nick Feamster, examines the functionality and inherent limitations of the most common filtering technologies to demonstrate why a mandatory filtering regime would pose grave dangers to the viability of the internet ecosystem in exchange for a minimal effects on online infringement.”
Main Results of the Study
The study concludes that filtering technologies are limited in their capabilities, and whilst they are able to identify the contents of a file, they are unable to determine whether this use constitutes an infringement. Several common filtering technologies were surveyed:
• Metadata-based filters are often imprecise, identifying many false positives and false negatives. This is due to the fact that metadata is often inaccurate and mislabeled, or may be altered entirely during a format change. Similarly, hash-based filters suffer from similar problems, with format changes invariably altering the unique identifier of the hash.
• Audio and video fingerprinting remains imperfect, often being proprietary in nature (making it difficult to evaluate their technical functionality) and tend to be limited to specific types of media (e.g. entirely audio, or entirely image-based). Furthermore, any minimal distortion or alteration to a media file may impact the technology’s ability to accurately identify copies, easily undermining its purpose.
Summarising, filtering technologies remain fundamentally flawed for the purpose of identifying infringements due to their lack of context-specific determinations.
Policy Implications as Stated By Author
Given the limited capabilities of filtering technologies, the authors caution against the adoption of such mechanisms at the expense of softening or removing safe harbour legislation. Any mandatory filtering regime is likely to have negative implications on three grounds:
• The expense of filtering technologies are likely to be prohibitively costly for small OSPs.
• The certainty of the safe harbour would be undermined, as any “reasonableness” or “sufficiency” standard for filtering would require clarification from courts.
• Filters are likely to have a minimal effect on infringement, particularly as larger OSPs already employ such technologies (e.g. YouTube, SoundCloud, Spotify) and many large-scale infringement pirate sites are based outwith the jurisdiction of the DMCA.