There is no clear definition of software. However, one of the more helpful definitions is the definition in the Model Provisions on the Protection of Computer Software, where software is defined as "a set of instructions" to "perform or achieve a particular function, task or result" [1]. The lack of clear definitions allows greater judicial discretion to shape the law, allowing this area of law to develop as common law instead of through statutes. However, it also creates greater uncertainty on legal matters and may discourage businesses from innovating or shift to a different jurisdiction with more certainty.
AI systems comprise instructions to feed data into machine learning models, the models themselves, and instructions to query the models to obtain an output [2]. The instructions are clearly software, however the model itself is unintelligible to human comprehension and will more likely lean more towards data rather than software. However, Karpathy argues that these models can be considered code written by machine learning methods [3]. In other words, humans write code that can write further code that humans no longer understand. On one hand, machines can definitely execute these set of valid instructions, however when humans can no longer understand them, can we still consider them instructions? Paragraph 12 of the Artificial Intelligence Act states that AI systems should be distinguished from "simpler traditional software systems" [4]. Thus, AI systems are more than just software, they belong in a sui generis class of their own, requiring lex specialis legislation to regulate their use.
Developers writing code can be thought of as legislators drafting legislation or lawyers drafting contracts. Legislation can be thought of as open source software, while contracts can be thought of as proprietary software. Software bugs can be thought of as lacunas in existing legislation and loopholes in contracts. Software version control and bug fixes can be thought of as amendments to legislation or contract variations. The judiciary is similar to the machine running the software while data is similar to each case that comes before the judiciary. When you use Word to open a different file, the outcome is different each time depending on the contents of the file. Similarly, when a different case comes before a judge, the outcome differs based on the facts of the case.
[1] Practical Law IP and IT, Legal Protection of Software (Practical Law (Westlaw) 2024), pp. 4
[2] Otero, BegoÑa Gonzalez (2021) 'Machine Learning Models Under the Copyright Microscope: Is EU Copyright Fit for Purpose?' GRUR International (Print), 70(11), 1043–1055. https://doi.org/10.1093/grurint/ikab077, pp. 14
[3] ibid, pp. 15
[4] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act), para. 12