INSIGHTS

Automation, artificial intelligence and legal issues

By Newyorka Musabelliu and Chris Chang – January 2020

It is not surprising that the law cannot keep up with the growth in artificial intelligence.  Patent applications have already been filed on behalf of and in the name of a machine called Dabus that used AI to design two products[1].  Nevertheless, progress is being made regarding law reform in this space. 

In December 2019 the Australian Law Reform Commission released a report for law reform in connection with artificial intelligence over the next five years[2] and the World Intellectual Property Organization (WIPO) announced a public consultation process on artificial intelligence and intellectual property policy[3].  

Below are some of the associated legal issues[4].

Key takeaway points

The implementation of different types of permanent controls, checks and balances on the AI which cannot be overriden is important given the different types of risks. The ALRC report notes recent commentary and states

“appropriate design choices are crucial for automated systems to achieve consistency with concepts such as transparency and accountability; predictability and consistency; and equality before the law”

and calls for effective regulation of technological developments. There are doubts as to whether mere principle based regulation will suffice and therefore the task of drafting such legislation will be both challenging and interesting.

IP Law

  1. There are a multitude of issues to consider in view of the current landscape but, in summary, it is important to ensure the retention of personal control over the creative process and provide the input to at least a substantial part of the end result. Otherwise the risk is that you will lose rights in works authored by AI or jointly authored with AI unless reform addresses this.
  2. Reform debate primarily focuses on ownership where the work has been created by AI. Who should own the intellectual property rights (1) no-one (2) the AI, or (3) an interested party such as the producer.  The latter is in our opinion the preferred option.  In any other scenario this would be unsatisfactory to many businesses who have invested in AI and the fruits of the AI’s labour, including adaptations and compilations etc of those works.  If left unaddressed, questions as to whether inventions presently created by AI can be registered as patents by persons other than the AI[5], or whether future applications based on such inventions are in fact already part of the prior art base and do not constitute inventive steps may also arise, and so forth.
  3. Who should be responsible for infringement by AI? We suggest the principles of agency law should apply. This question also opens the doors to liability for other positive acts of AI and questions concerning torts to the person for example which, given current advances in AI, is a possibility. 

Admin law

Even when automated systems are deployed carefully and effectively by government agencies, questions remain as to their compatibility with core administrative law principles and the rule of law.  In particular, it is not yet clear whether (or how) such systems can be said to act with procedural fairness.  The legality of actions and decisions by public bodies is called into question as is transparency and accountability of government decisions (in particular, whether automated decision-making software can provide adequate reasons that would in turn facilitate access to judicial and merits review where appropriate)[6].  The independent scrutiny of automated systems prior to implementation is already envisaged by the ALRC e.g. by way of algorithmic impact assessment[7] and therefore creators of AI are no doubt already focusing upon how to demonstrate compliance such tests if they havnt already done so. 

Privacy law

Information handled about individuals may not satisfy the definition of personal information under the Privacy Act 1988 (Cth) if the relevant person is for example allocated a code and is therefore not identifiable. Collection by AI may also occur without knowledge or consent (both of the public and the owner of the AI).  Permanent checks and balances are likely to already be in place.

Directors duties and defences

It is important for directors to understand the technology upon which they place any reliance, and any limitations if they intend to rely on the safe harbor provisions contained in s.189 of the Corporations Act 2001 (Cth).

Discrimination

Studies have shown that inadvertent discrimination may occur in the AI’s implementation of its algorithms.

Misuse of big data

Other issues raised include the propensity for improper use of big data in the financial sector, and consumer law. The WIPO statement of issues goes into greater detail[8].


Contact us for assistance with any intellectual property rights issues. 

Discussion

BACKGROUND AND STATUS OF AI

  1. As the phrase suggests, “artificial intelligence” (AI) is designed with the objective of imitating human intelligence artificially (and as has been proven, exceeding human intelligence in many ways, in the same way traditional computers have served us for so many years). Human intelligence includes accurate perception, analysis, logic / rationale, the ability to make calculated decisions based on objectives and, ultimately, solve problems.  It also includes creativity.  Machine learning has been available for years (e.g. for the purposes of electronic discovery in litigation).  Algorithms are programmed in machines designed to learn both from large amounts of historic data and through their own, “personal”, experience and evolution, to make accurate predictions. 
  2. Aside from significant advances in the use of artificial intelligence in the legal sector, we note some other examples of developments in recent times:
  3. The advanced use of AI as applied to gastrointestinal diagnostics. In particular, AI has been used in the “automated detection of disease and differentiation of pathology subtypes and disease severity[9]”;
  4. In October 2011, Takahiro Yamaguchi and So Kanno explored the link between the machine and art. The “Senseless Drawing Bot” is an autonomous device that sits on a skateboard which draws abstract lines using many colors. The robot moves from side to side, riding the wave of today’s modern graffiti. The lines drawn are complex and create abstract modern graffiti and one would have a difficult time arguing that the same work, if originally created by a human, does not constitute an artistic work for the purposes of the Copyright Act 1959 (Cth) (Act)[10].
  5. AI now appears to be controlling other robots with reference to the controller loaded with Olis Robotics’ AI-driven operating system / software that is capable of autonomously operating and maneuvering other robots on the ocean floor, as well as in space, without the need for specialist operators[11].
  6. Machine learning algorithms are capable of achieving a 92.93% accuracy score at reading human emotions[12].
  7. Scientists at the University of Cincinnati have managed to get 5 robots to work independently but collaboratively with each other on common tasks, in this case to solve a puzzle using strings to move an attached token to a designated spot on a table[13]. The list goes on.
  8. In the above examples it is more than likely that there is often little, if any, human input in the decision making that produces the fruits of the AI’s actions. The AI is making decisions, adapting, taking initiative, and being creative.  It is arguably self aware.  In many circumstances, the only difference is the fact it is not human.  

IP LAW

  1. Originality, authorship and ownership of works created by AI independently or jointly, as well as liability for infringement by AI, needs to be addressed by reform.
  2. Authorship and ownership: The fundamental issue is the question of ownership and this depends largely on whether or not it is the AI that can be said to have independently created the relevant work. It is clear that it is possible for AI to be deemed to have done so and therefore that the work is not protected by copyright[14].    
  3. Infringement: Another issue is liability for any infringement carried out by AI (which the AI might either do because its algorithms did not prevent it from doing so, or perhaps even because it chooses to do so). Inevitably, these machines must frequently  “reproduce” information including literary works (as computers always do). 
  4. Copyright Act 1959 (Cth): Under s.35 of the Copyright Act 1959 (Cth) (Act) the “author” of a literary, dramatic, musical or artistic work is the owner of any copyright subsisting in the work. The word “author” is only defined in relation to a cinematograph film as meaning the “maker” of the film.  The “maker” is defined as the “director of the film, the producer of the film and the screenwriter of the film”.  There are exceptions under s.35 which include circumstances where a person makes a work in pursuance of the terms of their employment (and in this case the work is owned by the employer).  The author is therefore the owner, and this would in many scenarios be the AI even before the work is published.
  5. However the AI cannot be a qualified person for the purposes of s.32 of the Act. Furthermore, under s.190 of the Act only individuals have moral rights.  Copyright does not subsist otherwise than under the Act and s.8 of the Act purports to cover the field.  Copyright is an exclusive right and it is personal property.  It is only transmissible by assignment in writing signed by or on behalf of the assignor under s.196 of the Act. 
  6. Therefore there is no owner of the rights in a work independently created or authored by AI[15]. A recent Australian case[16] is authority for this, where it was held that a work generated with the intervention of a computer could not be protected by copyright because it was not produced by a human.
  7. Ownership of copyright would also likely not vest in the owners of original algorithms for machine learning artificial intelligence whose algorithms evolve over time through its own learning and experiences[17].
  8. Furthermore, recent decisions indicate that copyright subsisting in original works created jointly with humans and AI (independently but collaboratively) would be held to be works of joint authorship[18] (therefore resulting in only fractional ownership by the relevant author(s) who are “person(s)”). It would depend on the circumstances and evidence as to whether or not the person(s) fractional share constituted a “substantial part” of the work in question so as to enable infringement proceedings to be successful against third person(s).  The evidence required would need to focus on the functionality of that part and demonstrate qualitative rather than quantitative similarities between the original and infringing work.  This might be an impossible task in the case of constantly evolving machine learning algorithms[19]
  9. Designs, Patents, Trade Marks: The position is similar in respect of designs, patents, and trademarks.  13 of the Designs Act 2003 (Cth) provides that a “person” is entitled to apply for registration.  Under s.2C of the Acts Interpretation Act 1901 (Cth), “expressions used to denote persons generally (such as “person”, “party”, “someone”, “anyone”, “no‑one”, “one”, “another” and “whoever”), include a body politic or corporate as well as an individual…”.  S.15 of the Patents Act 1990 provides that patents may only be granted to a “person”.  The definition of “registered owner” in s.6 of the Trade Marks Act 1995 (Cth) achieves the same result.
  10. Ownership: If works are currently not owned by anyone, consequently, issues may occur regarding ownership and providing chain of title in works which for example are, in truth and substance, reproductions or adaptations of works that were originally authored by AI in the current legal landscape. A decision on ownership must occur.  Questions still arise as to whether inventions presently created by AI can be registered as patents by persons other than the AI[20], or whether future applications based on such inventions are in fact already part of the prior art base and do not constitute inventive steps, and so forth.  The current situation could be unsettling for those invested in the techonology.  This could be dealt with by way of a simple exclusion to the meaning of inventive step etc covering inventions created by AI operated by person(s) during this time.  The issue of ownership pending reform might be partially remedied by carefully worded conditions in agreements having regard to provisions such as s.47H of the Copyright Act 1959 (Cth) which renders any agreement that excludes or limits the operation of certain other provisions in the Act concerning computer programs as void and of no effect.   
  11. AI is the owner: One option being debated is that the AI could be the owner of the rights. After all, it is the AI that has created the work and that has taken the inventive step.  It is assumed in this scenario that the AI would logically then also be responsible for its own infringements.  This view is fraught with difficulty in Australia because:
  12. Whilst legislation could override the common law, the effect of current precedent case law is that copyright that subsists in original works, which have been independently created or authored by AI, is not owned by any person in Australia[21]. The position is similar in respect of designs and patents.  It follows that, in the case of algorithms in machine learning artificial intelligence that evolve over time[22], ownership would also not vest in the investors or owners of that artificial intelligence;
  13. In our opinion, it also follows from recent decisions that copyright subsisting in original works created jointly with humans and AI (independently but collaboratively) are works of joint authorship[23] (therefore resulting in only fractional ownership by the relevant author(s) who are “person(s)”). It would depend on the circumstances and evidence as to whether or not the person(s) fractional share constituted a “substantial part” of the work in question so as to enable infringement proceedings to be successful against third person(s).  The evidence required would need to focus on the functionality of that part and demonstrate qualitative rather than quantitative similarities between the original and infringing work.  This might be an impossible task in the case of constantly evolving machine learning algorithms[24].
  14. ownership would presumably mean the AI is entitled to any fruits of commercialization. If AI can own copyright, why not shares, or land? It is also difficult to imagine how AI could legally assign or deal with ownership (and perhaps one day it might choose not to).  It is important to also consider what AI might look like in 10 or 20 years’ time.  A special purpose vehicle might be used, but then the owners would be the human owners anyway.
  15. The Federal Court of Australia has already held, in the particular circumstances of that case and on the available evidence presented, that source code in AI was not authored by any single human author, but rather it was created by the AI. Accordingly, the relevant humans had no rights in those works even if they contributed to the AI’s algorithmic programming which enabled the resulting work in question to be created[25].   
  16. University of Sydney v ObjectiVision Pty Limited [2019] FCA 1625 is also an important recent decision regarding joint authorship and provides helpful guidance on the level and quality of expert evidence required in these cases.
  17. On the other hand, an interested individual such as the producer of the AI can be the owner. In our opinion the position is less clear as to whether or not the owner should be liable for the actions of the AI.  A possible analogy might be the relationship between employee and employer, or internet service provider and the infringing end-user.  It is suggested that liability should be similarly determined by reference to causation, and having regard to the fact that infringement can occur by “authorizing” any act comprised in the copyright[26] if the requisite level of control is present. 
  18. Overseas jurisdictions: Different jurisdictions across the globe have taken different approaches or have not yet specifically addressed the issue. The US doesn’t grant rights to works which don’t have a human author.  The UK however does and grants rights to the producer.
  19. A common theme globally is that the acts speak of the owner as being the “person” who has “created” or used “intelligence” to produce fruits of intellectual and original work. Some do not yet address the issue of non-human ownership.  Others such as the United States and Spain specifically refer to the rights as only applying to a human being (however this in itself doesn’t address the problem that, if only a human can enjoy such rights, then no one owns the rights if the fruits of the intellectual labour were created by AI). 
  20. Infringement: There does not appear to have been much debate concerning infringement by AI. It is noted that the United Nations Convention on the Use of Electronic Communications in International Contracts (Authentic text)[27] article 12 canvassed this issue in the context of contracts formed by AI:

Article 12. Use of automated message systems for contract formation

A contract formed by the interaction of an automated message system and a natural person, or by the interaction of automated message systems, shall not be denied validity or enforceability on the sole ground that no natural person reviewed or intervened in each of the individual actions carried out by the automated message systems or the resulting contract.

“…At present, the attribution of actions of automated message systems to a person or legal entity is based on the paradigm that an automated message system is capable of performing only within the technical structures of its preset programming. However, at least in theory it is conceivable that future generations of automated information systems may be created with the ability to act autonomously and not just automatically. That is, through developments in artificial intelligence, a computer may be able to learn through experience, modify the instructions in its own programs and even devise new instructions. 212. Already during the preparation of the Model Law on Electronic Commerce, UNCITRAL had taken the view that that, while the expression “electronic agent” had been used for purposes of convenience, the analogy between an automated 70 message system and a sales agent was not appropriate. General principles of agency law (for example, principles involving limitation of liability as a result of the faulty behaviour of the agent) could not be used in connection with the operation of such systems. UNCITRAL also considered that, as a general principle, the person (whether a natural person or a legal entity) on whose behalf a computer was programmed should ultimately be responsible for any message generated by the machine (see A/CN.9/484, paras. 106 and 107). 213. Article 12 of the Electronic Communications Convention is an enabling provision and should not be misinterpreted as allowing for an automated message system or a computer to be made the subject of rights and obligations. Electronic communications that are generated automatically by message systems or computers without direct human intervention should be regarded as “originating” from the legal entity on behalf of which the message system or computer is operated. Questions relevant to agency that might arise in that context are to be settled under rules outside the Convention.

There should also be a causal connection or sufficient objective similarity and it is the quality of that which has been copied rather than the quantity which is of importance[28].


ADMINISTRATIVE LAW

  • Even when automated systems are deployed carefully and effectively by government agencies, questions remain as to their compatibility with core administrative law principles and the rule of law. In particular, it is not yet clear whether (or how) such systems can ensure procedural fairness; the legality of actions and decisions by public bodies; and transparency and accountability of government decisions (in particular, whether automated decision-making software can provide adequate reasons that would in turn facilitate access to judicial and merits review where appropriate). [29].
  • By way of example, s.135A of the Designs Act 2003 (Cth) provides for computerized decision-making of the Registrar of Designs. It acknowledges the potential incorrectness of such decisions at s.135A (3).  Similar provisions exist in the case of trade marks and patents.

DIRECTORS DUTIES

Directors who wish to seek protection under s.189 of the Corporations Act should ensure that they are ready to explain why it was reasonable for them to rely on AI in the circumstances. In many cases this is likely to be the case and perhaps AI is more reliable than many humans for certain tasks but if there is anything to suggest the algorithms or data is compromised, then this would suggest that reliance should not have been placed.


DISCRIMINATION

Furthermore there is already concern that they may inherently and inevitably discriminate based on correlations of predictive patters and segmentations founded on prescriptive actions[30].


[1] https://www.dezeen.com/2019/08/01/patent-ai-machine-products-technology/

[2] https://www.alrc.gov.au/publication/the-future-of-law-reform-2020-25/

[3] https://www.wipo.int/pressroom/en/articles/2019/article_0017.html

[4] See here for full WIPO issues paper  https://www.wipo.int/edocs/mdocs/mdocs/en/wipo_ip_ai_ge_20/wipo_ip_ai_2_ge_20_1.pdf

[5] Noting that registration in the name of the AI would be fraught with numerous difficulties discussed below

[6] https://www.alrc.gov.au/wp-content/uploads/2019/11/Future-of-Law-Reform-Final-Report_v3web.pdf, para 2.15

[7] https://www.alrc.gov.au/wp-content/uploads/2019/11/Future-of-Law-Reform-Final-Report_v3web.pdf, para 2.5

[8] https://www.wipo.int/edocs/mdocs/mdocs/en/wipo_ip_ai_ge_20/wipo_ip_ai_2_ge_20_1.pdf

[9] Artificial Intelligence Applied to Gastrointestinal Diagnostics A Review, by Patel, Vatsal*; Khan, Marium N.†; Shrivastava, Aman‡; Sadiq, Kamran§; Ali, S. Asad§; Moore, Sean R.†; Brown, Donald E.‡,||; Syed, Sana†, Journal of Pediatric Gastroenterology and Nutrition: January 2020 – Volume 70 – Issue 1 – p 4–11

[10] https://www.designboom.com/art/senseless-drawing-bot/

[11] https://www.olisrobotics.com/products-rev

[12] https://www.pnas.org/content/115/14/3581

[13] University of Cincinnati. “Satellite broken? Smart satellites to the rescue: Aerospace engineers are developing technology to repair and refuel stranded satellites in space..” ScienceDaily. ScienceDaily, 26 November 2019. <www.sciencedaily.com/releases/2019/11/191126155502.htm>.

[14] Acohs Pty Ltd v Ucorp Pty Ltd [2012] FCAFC 16

[15] See IceTV Pty Limited v Nine Network Australia Pty Limited [2009] HCA 14 at [33]; originality lies with the author

[16] Acohs Pty Ltd v Ucorp Pty Ltd [2012] FCAFC 16

[17] See Data Access v Powerflex Services [1999] HCA 49 at [85] as to what constitutes a “computer program”

[18] University of Sydney v ObjectiVision Pty Ltd |[2019] FCA 1625 distinguishing Acohs

[19] It is assumed the definition of “computer program” will be varied and expanded in any event

[20] Noting that registration in the name of the AI would be fraught with numerous difficulties discussed below

[21] See IceTV Pty Limited v Nine Network Australia Pty Limited [2009] HCA 14 at [33]; originality lies with the author

[22] See Data Access v Powerflex Services [1999] HCA 49 at [85] as to what constitutes a “computer program”

[23] University of Sydney v ObjectiVision Pty Ltd |[2019] FCA 1625 distinguishing Acohs

[24] It is assumed the definition of “computer program” will be varied and expanded in any event

[25] Acohs Pty Ltd v Ucorp Pty Ltd [2012] FCAFC 16

[26] S.36(1) Copyright Act

[27] https://wipolex.wipo.int/en/text/201667

[28] University of Sydney v ObjectiVision Pty Limited [2019] FCA 1625

[29] https://www.alrc.gov.au/wp-content/uploads/2019/11/Future-of-Law-Reform-Final-Report_v3web.pdf, para 2.15

[30] Burdon, Mark; Harpur, Paul — “Re-conceptualising Privacy and Discrimination in an Age of Talent Analytics” [2014] UNSWLawJl 26; (2014) 37(2) UNSW Law Journal 67

Contact us for a free fixed fee quote

Call us on +61 2 8005 7388, email us at contact@hglaw.com.au, or book in an appointment online in 1 minute by clicking here