Skip to main content
SpringerLink
Log in

Australian Copyright Law Impedes the Development of Artificial Intelligence: What Are the Options?

  • Article
  • Published:
IIC - International Review of Intellectual Property and Competition Law Aims and scope Submit manuscript

Abstract

Artificial Intelligence (AI) is an emerging technology that has a huge potential in contributing to the Australian economy and addressing economic and social problems in society. However, Australian copyright laws are likely to impede the development of AI, and machine learning in particular, by requiring authorisation every time copyrighted content is used in machine learning processes. This puts Australian AI industries at a competitive disadvantage, since other AI-focused jurisdictions, such as the US, EU, UK and Japan, allow such use under copyright exceptions. After analysing the scope of copyright in relation to machine learning and licensing options, this paper examines different copyright exceptions (fair use, fair dealing and an EU-style specific TDM exception) as potential solutions for Australia.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Similar content being viewed by others

Notes

  1. Barbaschow (2020).

  2. Osman (2019).

  3. Tashea (2017).

  4. Heilweil (2019).

  5. Tufekci (2019).

  6. White and Matulionyte (2020).

  7. PWC, The macroeconomic impact of artificial intelligence, February 2018. https://www.pwc.co.uk/economic-services/assets/macroeconomic-impact-of-ai-technical-report-feb-18.pdf.

  8. E.g. Australian AI Ethics Framework, https://www.industry.gov.au/data-and-publications/building-australias-artificial-intelligence-capability/ai-ethics-framework; European Ethics Guidelines for Trustworthy AI, https://ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai.

  9. For more about different licensing regimes that currently exist in Australia see section “Could licensing be a solution” below.

  10. English Oxford Living Dictionary (online at 25 May 2019) “artificial intelligence”.

  11. Other subsets are Reasoning, Natural Language Processing (NLP) and Planning, see Hurwitz and Kirsch (2018).

  12. Ibid, p. 4.

  13. Ibid, p. 15.

  14. Ibid, p. 33.

  15. For instance, Clearview AI developed AI-based face recognition model by scraping pictures from Facebook, Vimeo, YouTube and thousands of other websites, see Kashmir Hill (2020).

  16. Hurwitz and Kirsch (2018).

  17. The need for a new text and data mining exception has been discussed in Australia, EU and elsewhere for almost a decade. The recently introduced EU TDM exception is discussed below.

  18. European Commission (2014), p. 3.

  19. See Global Expert Network on Copyright User Rights, “Joint Comment to WIPO on Copyright and Artificial Intelligence”, (17 February 2020), http://infojustice.org/archives/42009.

  20. I.e. while text and data mining identifies patterns, machine learning makes decisions based on the patterns identified during text and data mining.

  21. European Alliance for Research Excellence et al. (2018)

  22. E.g., see Microsoft and Bjork project discussed above; as another example see Megan Friedman, (2019).

  23. E.g. Facebook offers very broad Terms of Service that could potentially include use of users’ content for machine learning (e.g. with the purpose to improve service), see https://www.facebook.com/about/privacy/update.

  24. E.g. ClearView did not have authorisations to use photos for the development of facial recognition AI; in response, Facebook and Google send them cease and desist letters, see Alfred Ng (2020).

  25. See also Australian Law Reform Commission (ALRC), Copyright and the Digital Economy (Report 122, 2014), para. 11.64 (“Where the data or text mining processes involve the copying, digitisation, or reformatting of copyright material without permission, it may give rise to copyright infringement”).

  26. Sec. 31(1)(a)(i) Copyright Act 1968.

  27. While the term “reproduction” is not defined in the Copyright Act 1968 (Cth) (“Copyright Act”), Sec. 21(1A) clarifies that the conversion of the work from analogue to digital and vice versa constitutes reproduction of the work.

  28. Sobel (2017).

  29. See Sec. 10 of Copyright Act 1968 (“literary work” includes “table” and “compilation”). For comment on database protection in Australia see e.g. Fitzgerald et al., (2011) paras. 4.130–4.140. Notably, while creating datasets for the purpose of machine learning is still quite a common practice, in the future the use of such centralised datasets might decrease as new technology allows training AI based on content hosted on other decentralised devices (e.g. mobile phones), see Sobel, supra note 48, fn 95.

  30. Sobel, supra note 28, pp. 62–63.

  31. E.g. Clearview face recognition technology.

  32. Romano (2016).

  33. See Sec. 31(1)(a)(iv) and (ii) Copyright Act 1968.

  34. Burrell et al. (2013).

  35. A related but less relevant moral right is a right not to be falsely attributed, see Sec. 195AC, 195AHA Copyright Act 1968.

  36. See Sec. 193 (authors), Sec. 195ABA (performers) Copyright Act 1968.

  37. Sec. 195(1) Copyright Act 1986.

  38. Sec. 195AA Copyright Act 1968.

  39. Sec. 195AB; similar requirements apply to performers Sec. 195ABA–195ABE Copyright Act 1968.

  40. E.g. if texts are scraped from public online materials, some of texts might have no identified author.

  41. E.g. UK provision on text and data mining requires identifying authors “unless it is this would be impossible for reasons of practicality or otherwise”, see Sec. 29A Copyright, Design and Patent Act (UK); See also Corby v. Allen & Unwin Pty Limited [2013] FCA 370 (court rejects defendant’s argument that the requested attribution was not reasonable since it is not an accepted “industry practice”).

  42. Secs. 195AI, 195ALA Copyright Act 1968.

  43. Secs. 195AJ (all works except artistic works); 195AK (artistic works), 195AL (cinematographic work), 195ALB (performers’ integrity right) Copyright Act 1968.

  44. See Australian Society of Authors submission to the Australian Government Copyright Modernisation Consultation, https://www.communications.gov.au/have-your-say/copyright-modernisation-consultation.

  45. See e.g. Perez v. Fernandez (2012) 260 FLR 1.

  46. Sec. 40 Copyright Act 1968. The analogous fair dealing provision for Part IV subject matter is found in Sec. 103C.

  47. See Aja Romano (2016), cited from Sobel, supra note 28, fn 104 (AI model developed by a student and trained with elements from the film Blade Runner, was used to reconstruct the full film).

  48. Sec. 40(2) Copyright Act 1968.

  49. E.g. see Yukun Zhu et al (2015).

  50. Most previously it was applied in Authors Guild v. Google, Inc. 804 F.3d 202 (2d Cir. 2015).

  51. See R Burrell (2013), it has also been discussed in ALRC Report 122, para. 11.62.

  52. See similar conclusion in a US TDM case Authors Guild v. Google, Inc. 804 F.3d 202 (2d Cir. 2015).

  53. Works primarily based on facts and other unprotected elements are likely to show low levels of originality and therefore establishing fair use is also likely to be easier.

  54. See ALRC Report 122, para. 11.65.

  55. Sec. 40(3) and (4) Copyright Act 1968.

  56. Sec. 40(5) Copyright Act 1968.

  57. E.g. when an entire drawing is used as an illustration for research and study purposes.

  58. De Garis v. Neville Jeffress Pidler Pty Ltd, (1990) 37 FCR 99, 105−106.

  59. E.g. the BookCorpus dataset was subsequently used by Google to train Google Assistant and Google Search Engine, see Richard Lea (2016).

  60. “A Trump Speech Written by Artificial Intelligence”, The New Yorker (27 July 2017), https://www.youtube.com/watch?v=EFHyzuqjaok.

  61. See e.g. article “Political donations plunge to $16.7m – down from average $25m a year” generated by AI ReporterMate, https://www.theguardian.com/australia-news/2019/feb/01/political-donations-plunge-to-167m-down-from-average-25m-a-year.

  62. Marr (2018).

  63. Sourdin (2018), p. 1114.

  64. An illustrative case where fair dealing for news reporting was applied in Australia is Network Ten Pty Limited v. TCN Channel Nine. Pty Limited [2004] HCA 14.

  65. E.g. in National Rugby League Investments Pty Limited v. Singtel Optus Pty Ltd [2012] FCAFC 59 (27 April 2012) (confirmed that the time-shifting exception could be relied upon by private user and not by an entity that facilitates the private use covered by the exception).

  66. See e.g. National Rugby League Investments Pty Limited v. Singtel Optus Pty Ltd [2012] FCAFC 59 (27 April 2012).

  67. See e.g. National Rugby League Investments Pty Limited v. Singtel Optus Pty Ltd [2012] FCAFC 59 (27 April 2012) (while users were able to rely on a time shifting exception, a service provider that enabled time shifting could not rely on the same exception).

  68. Sec. 42 Copyright Act 1968.

  69. Sec. 41 Copyright Act 1968.

  70. Fitzgerald et al. (2011) para. [4.720].

  71. Ibid.

  72. See section “How machine learning works” above.

  73. E.g. a similar temporary copying exception available under the EU law covered screen copies that end users of press clipping service received on their screens; see EU (CJEU) Case C-360/13 Public Relations Consultants Association v. Newspaper Licensing Agency Ltd.

  74. Sec. 43(B)(2)(a)(i) Copyright Act 1968.

  75. Sec. 43(B)(2)(b) Copyright Act 1968.

  76. Australian Publishers Association submission for Copyright modernisation consultation, both available at https://www.communications.gov.au/have-your-say/copyright-modernisation-consultation.

  77. Australian Copyright Council submission to Copyright Modernization Consultation, available at https://www.communications.gov.au/have-your-say/copyright-modernisation-consultation, para. 32.3.

  78. For instance, in order to train its AI DeepFace technology, Facebook used four million photos of faces from their user photo database, see “Facebook Creates Software That Matches Faces Almost as Well as You Do”, https://www.technologyreview.com/s/525586/facebook-creates-software-that-matches-faces-almost-as-well-as-you-do/.

  79. E.g. Willey, a scientific publisher, has allowed TDM for non-commercial research purposes without additional fees, see Wiley Submission to Copyright Modernization Consultation, p. 2.

  80. See Asha Barbaschow (2020)

  81. See similar arguments in European Commission, "Standardisation in the area of innovation and technological development, notably in the field of Text and Data Mining: Report from the Expert Group" (April 2014), p. 52.

  82. Such as anti-competitive behaviour, lack of transparency or irrelevance in light of new technologies such as Blockchain, see e.g. Tresise, (2018).

  83. As of 2018, Copyright Agency represented 2907 visual artists, see Copyright Agency Annual Report 2018, p. 40, https://www.copyright.com.au/about-us/governance/annual-reports/copyright-agency-annual-report-2018/.

  84. E.g. this content is especially useful in developing recommender algorithms, see Zeinep Tufekci (2019).

  85. Their use is normally defined by Terms of Use provisions for a particular service (Mail Box, Website, etc.).

  86. See e.g. Axel Metzger (2015) 11 para. 1, para. 2.1.1.

  87. For more about the definition of orphan works and legal issues surrounding them in Australia, see ACLR Report 211, Chapter 13.

  88. See e.g. Directive 2012/28/EU of the European Parliament and of the Council of 25 October 2012 on certain permitted uses of orphan works, OJ L 299, 27.10.2012, pp. 5–12 (EU Orphan Works Directive); UK Copyright, Patent and Design Act, Sec. 44B.

  89. Several recommendations in relation to orphan works have been made by different reviews but government has taken no action to date, see e.g. ALRC Report 122, Chapter 13; Productivity Commission Report on IP Arrangements (2016).

  90. See section “Individual licensing” above.

  91. Court (1987).

  92. Ibid.

  93. E.g. in case of EU ECL scheme for out-of-print books, it will allow libraries and other institutions to digitize their vast archives and make them accessible to public.

  94. See e.g. Benjamin Cheatham (2019).

  95. Guibault (2015).

  96. Art. 8(4) EU Digital Single Market Directive.

  97. This option might be used by publishers who already license their content for TDM for researchers, e.g. Wiley.

  98. See “Introduction” above.

  99. ALRC Report 122, paras. 3.51–3.68.

  100. ALRC Report 122, Chapter 5.

  101. Productivity Commission Report on IP Arrangements (2016), p. 2.

  102. Sec. 107 Copyright Act (US).

  103. See Sobel, supra note 28, p. 12 (agrees that in most cases machine learning is covered by fair use).

  104. Authors Guild, Inc. v. Google Inc. 954 F. Supp. 2d 282 (S.D.N.Y. 2013).

  105. Transformativeness criteria was initially formulated in Campbell, 510 U.S. 569, 578–579.

  106. Authors Guild v. Google, Inc. 804 F.3d 202 (2d Cir. 2015), see also Rita Matulionyte (2016) pp. 44–71.

  107. In case of Google Books case, both the District Court judge Chin and Second Circuit court highlighted the positive impacts of Google Books service on authors, such as increased awareness of books among readers and potential increase of sales, see Authors Guild, Inc. v. Google Inc. 954 F. Supp. 2d 282 (S.D.N.Y. 2013), 22.

  108. 17 U.S.C. § 107(2) – if the work used is informative rather than expressive, this normally weights in favour of fair use.

  109. 17 U.S.C. § 107(3)– the smaller part is used, the more this criteria weights in favour of use.

  110. See e.g. Authors Guild, Inc. v. Google Inc. 954 F. Supp. 2d 282 (S.D.N.Y. 2013), 22.

  111. See Next Rembrant project, available at www.nextrembrandt.com.

  112. Similarly, Sobel argues that certain expressive machine learning outputs would not meet transformativeness test, see Sobel, supra note 28, pp. 12–13.

  113. See also European Commission, "Standardisation in the area of innovation and technological development, notably in the field of Text and Data Mining: Report from the Expert Group" (April 2014), p. 3

  114. See similarly, Lyombe S. Eko (2013), p. 221.

  115. See e.g. Matthew Sag (2019) p. 291.

  116. PWC, The macroeconomic impact of artificial intelligence, February 2018. https://www.pwc.co.uk/economic-services/assets/macroeconomic-impact-of-ai-technical-report-feb-18.pdf.

  117. See McKinsey, Digital Australia: Seizing the opportunity from the Fourth Industrial Revolution (Sydney 2017), https://www.mckinsey.com/featured-insights/asia-pacific/digital-australia-seizing-opportunity-from-the-fourth-industrial-revolution.

  118. Creative industries are generally defined as including music and performing arts, film, television and radio, advertising and marketing, software development and interactive content, writing, publishing and print media, architecture and design and visual arts sectors, see SGS Economics and Planning, Valuing Australia’s Creative Industries. Final Report (2013), available at https://www.sgsep.com.au/assets/main/Valuing-Australias-Creative-Industries-Final-Report-December-2013_Email.pdf.

  119. See ALRC Report 122, paras. 3.51–3.68.

  120. National Endowment for the Arts, How the US Funds the Arts (2012).

  121. In 2013 Australian creative industries were valued at $90 billion, of which 37% is the value of software development and interactive content, see SGS Economics and Planning, Valuing Australia’s Creative Industries. Final Report (2013), available at https://www.sgsep.com.au/assets/main/Valuing-Australias-Creative-Industries-Final-Report-December-2013_Email.pdf. Consulting firm AlphaBeta and CSIRO’s Data61 estimates that digital technologies, including AI, will be potentially worth $315 billion to the Australian economy by 2028, see AlphaBeta, Digital innovation: Australia’s $315B opportunity (2018, AlphaBeta Sydney).

  122. E.g. ALRC Report 122, Chapter 5; Copyright Law Review Committee, Simplification of the Copyright Act 1968, Part 1: Exceptions to the Exclusive Rights of Copyright Owners (1998), para. 6.08; Intellectual Property and Competition Review Committee, Review of Intellectual Property Legislation under the Competition Principles Agreement (2000), p. 15.

  123. See e.g. ASA Submission, https://www.communications.gov.au/have-your-say/copyright-modernisation-consultation.

  124. See section above.

  125. ALRC Report 122, 11.5–11.9.

  126. Report 122, para. 11.74.

  127. See Roundtable on incidental and technical uses of copyright, 30 May 2018, https://www.communications.gov.au/have-your-say/copyright-modernisation-consultation.

  128. See discussion above.

  129. Cf Sec. 40 Copyright Act 1968 (fair dealing for research and education contains additional presumptions).

  130. These issues are e.g. clarified in the EU TDM Directive, see discussion below.

  131. E.g. certain presumptions and clarifications could be implemented in Copyright Regulations which could be amended by the Government.

  132. Arts. 3 and 4 of the Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC, L 130, 17 May 2019 (“EU Digital Single Market Directive”).

  133. Arts. 3(2) and 4(2) of the EU Digital Single Market Directive.

  134. Art. 4(3) EU Digital Single Market Directive.

  135. Art. 4(3) EU Digital Single Market Directive. Recital 18 provides further guidelines as to how the reservation of rights can be expressed in online and offline environment.

  136. See e.g. European Alliance for Research Excellence, “Europe’s ability to lead in AI will be helped by the new TDM exception” (3 April 2019), http://eare.eu/europes-ability-to-lead-in-ai-will-be-helped-by-the-new-tdm-exception-2/.

  137. Bernt Hugenholtz (2019).

  138. See discussion above.

  139. See discussion above.

  140. Art. 4 DSM Directive.

  141. Bernt Hugenholtz, supra note 137.

  142. E.g. science publishers Wiley that has already given free licenses to use its content for non-commercial TDM purposes, might want to charge licensing fees if their content is used in commercial TDM projects, see Wiley Submission to Copyright Modernization Consultation, p. 2.

  143. See e.g. Arts. 3(2) and 4(2) EU Digital Single Market Directive.

  144. European Alliance of Research Excellence (2018).

  145. This problem has been identified and discussed in Copyright Law Review Committee, Simplification of the Copyright Act 1968, Report (1998).

  146. In contrast, they tend to give restrictive meaning to exceptions, see e.g. National Rugby League Investments Pty Limited v. Singtel Optus Pty Ltd [2012] FCAFC 59 (27 April 2012).

  147. See e.g. Copyright Regulations 2017.

References

Download references

Author information

Authors and Affiliations

  1. Ph.D., LL.M. (Munich); Senior Lecturer, Macquarie Law School, Macquarie University, Australia; Senior Associate Researcher, Lithuanian Social Research Centre, Vilnius, Lithuania

    Rita Matulionyte

Authors
  1. Rita Matulionyte
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Rita Matulionyte.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Matulionyte, R. Australian Copyright Law Impedes the Development of Artificial Intelligence: What Are the Options?. IIC 52, 417–443 (2021). https://doi.org/10.1007/s40319-021-01039-9

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s40319-021-01039-9

Keywords

Search

Navigation