|Rodarte S/S 2019|
Yesterday The New York Times published a very interesting article by fashion expert Vanessa Friedman, which praises the spring/summer 2019 collections presented by Telfar and Rodarte during New York Fashion Week, as being about fashion that consumers do not really know they want yet.
Apparently it has become commonplace for fashion houses to produce garments that consumers already want. As explained by Friedman,
There’s a lot of pressure these days to design by algorithm. We know too much about buying habits and likes, and the result is an insidious bias toward giving people what they have already indicated they want. It may be safe, and easier to sell, but it’s antithetical to the whole point of fashion, which should be about giving people what they never knew they wanted — what they couldn’t imagine they wanted — until they saw it.
But how is all this 'profiling' done?
It will not come as a surprise that identification of what we will want, and what we will be eventually given and shop, owes a great deal to our own social media activity, that is the 'likes' we give, whom we follow, and the content that we share ourselves. Is this data difficult to collect and elaborate, and are online platforms the (gate)keepers of all this information that can be then translated into revenues and personal success of fashion designers? The answer is: not necessarily.
Mining social media content
A few months ago, as I was looking for applications of text and data mining (TDM) techniques in various fields, I read an intriguing study conducted at Cornell University. With the aid of TDM techniques researchers were able to mine 100 million photographs made available on Instagram and devise patterns on how clothing styles vary around the world, and tackle the frequency of use of certain garments and colours. By training a machine-learning algorithm, the researchers were able to identify a set of visual themes and study how these would vary by time and place, and also identify the preference for certain colours.
A study of this kind is interesting from an anthropological standpoint, but may be even more interesting for fashion businesses in order to understand and even anticipate the next big trend ... or the next cerulean sweater. As explained by the authors of the study themselves,
Individuals make fashion choices based on many factors, including geography, weather, culture, and personal preference. The ability to analyze and predict trends in fashion is valuable for many applications, including analytics for fashion designers, retailers, advertisers, and manufacturers. This kind of analysis is currently done manually by analysts by, for instance, collecting and inspecting photographs from relevant locations and times. We aim to extract meaningful insights about the geo-spatial and temporal distributions of clothing, fashion, and style around the world through social media analysis at scale.
The question that arises from an IP standpoint is whether and to what extent unauthorized mining activities of this kind may be considered lawful.
|Matzen et al, 'StreetStyle: Exploring world-wide clothing styles from millions of photos' (2017) arXiv:1706.01869 [cs.CV]|
According to Instagram's Platform Policy, by making content available there, users grant Instagram and its affiliates a non-exclusive, transferable, sub-licensable, royalty-free, worldwide licence to use any data, content, and other information made available by users or on their behalf in connection with their use of Instagram. This licence survives even if one stops using the platform feature. Users are responsible for obtaining the necessary rights from all applicable rightholders to grant such licence.
It would thus appear that, unless one is an affiliate of Instagram, the licence granted to the platform may not cover the making of annotated datasets containing thousands of images to be made publicly available.
From a copyright standpoint, the extraction and reproduction of Instagram images may pose copyright issues (among other things). One may thus wonder whether and to what extent liability might arise for the making of such restricted acts without permission from relevant rightholders or whether, instead, no permission is needed because of applicable copyright exceptions.
Under US law, the question would be one of fair use under §107 of the Copyright Act. The fair use assessment requires to consider - among other things - whether the use made of a work 'adds value to the original - if the quoted matter is used as raw material, transformed in the creation of new information, new aesthetics, new insights and understandings - this is the very type of activity that the fair use doctrine intends to protect for the enrichment of society' (PN Leval (1990), 'Toward a Fair Use Standard' 103 Harv L Rev 1105, 1111).
In the longstanding litigation over the Google Books Library Project, the 2nd Circuit considered relevant for a finding of fair use also the fact that the search engine 'makes possible new forms of research, known as “text mining” and “data mining.”' by using the Google Library Project corpus 'to furnish statistical information to Internet users about the frequency of word and phrase usage over centuries'.
There is case law in the US (see Sag and Schultz (2013), 'Brief of Digital Humanities and Law Scholars as Amici Curiae in Authors Guild v Hathitrust', 4) that suggests that acts of incidental or intermediate copying which do not ultimately result in the external re-use of protectable (expressive) parts of a copyright work (this might not be the case of the whole of the Cornell project) should not be considered infringing, ie such as to supersede the objects or purposes of the original creation.
It may thus be the case that, under US law, the mining of Instagram content - insofar as protected parts of the content mined are not re-used or made available as such - might be considered fair use. That would be so because the goal of the mining activity is not creating a replacement for the original content, but rather extracting information (ideas and facts are not protectable as such under copyright) in order to obtain new information.
Things, however, may be different in Europe, especially considering the limitations that might be envisaged for an EU-wide TDM exception.
The approach in the EU?
While a limited number of Member States in the EU has already introduced (UK, France, Germany and Estonia: see here) or is planning to introduce (eg Ireland, on which see Katposts here and here) a specific text and data mining exception, in the context of the current copyright reform debate, a provision (Article 3) has been included in the draft Directive on copyright in the Digital Single Market.
In its original formulation, the mandatory EU TDM exception would allow any type of TDM (commercial and non-commercial). However, the catalogue of beneficiaries and the purpose of permitted mining activities would be limited: Article 3 would only apply to 'research organisations in order to carry out text and data mining of works or other subject matter to which they have lawful access for the purposes of scientific research'.
As proposed in the JURI Committee Report of the European Parliament, Member States should remain free - despite the introduction of a mandatory EU-wide TDM exception - to retain or introduce TDM exceptions in accordance with point (a) of Article 5(3) of the InfoSoc Directive, this being the legislative basis used to introduce non-commercial TDM exceptions that, as is the case of the UK (section 29A CDPA), do not envisage any limitations as regards the beneficiaries of the exception.
Algorithmic fashion is a thing: fashion houses have been increasingly giving consumers something that consumers have already suggested - more or less implicitly - to be wanting. This knowledge may be acquired manually but also through TDM techniques. These allow to learn much more and much more reliably, not just about things as they are, but also about things as they will be.
From an IP standpoint, all this raises a number of legal questions at the interplay between copyright subsistence, exceptions and licensing. From a fashion perspective, it instead raises the question whether fashion should be about what we want and everything consumer profiling and purchasing habits or, instead, what we didn't know we wanted ... yet.
[Originally published on The IPKat on 11 September 2018]