The EU Artificial Intelligence Act: Regulating Subliminal Al Systems (Routledge Research in the Law of Emerging Technologies) by Rostam J. Neuwirth*

Book Review by Yeliz Figen Döker and Habibe Deniz Seval

1. Overview

Rostam J. Neuwirth, in his latest eclectic book on the subliminal effects of Artificial Intelligence (“AI”) systems[1], conducts a root cause analysis: Following the author’s own words, this book is an attempt to contribute to the argument by taking a closer look at the function of the mind in general and the influence of subliminal techniques on the mind in particular. Through close reading of Article 5 of the Artificial Intelligence Act (”AIA”)[2], Neuwirth discovers the root causes of problems in order to identify the ambiguities surrounding the so-called prohibition of subliminal techniques, as well as to offer plausible solutions for overcoming those ambiguities. 

The book has three chapters. In the first chapter, Neuwirth gives a short background for the EU Artificial Intelligence Act. After that, he goes into detail regarding the employment of AI systems with subliminal effects. Finally, he lays the groundwork for his argument by introducing each problematic notion surrounding “the prohibition” with scientific foundations regarding how to comprehend the term “subliminal.”

It is worth mentioning that, to understand this book on the level it deserves, one should have an interest in AI systems, cognitive science, neuroscience, and their intersection with the law—for example, how the human brain functions in terms of its perceptual and cognitive abilities. So, this book is not your typical textbook in light of how the author plays with language, as well as the thorough analysis he presents of both law and the intense conduct of subliminal manipulation requires dedication and cogitation. Still, readers should not be intimidated from approaching this in-depth, original and timely book, especially those interested in the subjects of law, psychology, neuroscience, or anyone trying to connect the dots between interdisciplinary areas in order to tackle the complex problems that AI systems (may) bring.

In terms of flow, the author starts by approaching the AIA,  examining it from legal and scientific perspectives with a focus on the proposed prohibition of subliminal techniques employed by AI as well as various related technologies. He describes the extent to which lawmakers need to be more aware of regulating emerging technologies. Moreover, he offers a scientifically established foundation that could be used by the EU regarding Article 5 of AIA. So, with respect to the authors’ outline, this review will also examine the aforementioned legal and scientific challenges in the order presented by the book.

2. Legal Aspect

First, to highlight what the author wants to address with this book, one must mention “future-proofing.”[3] When emerging technologies are introduced, society is more committed to embedding them into its infrastructures; hence, society will likely face disputes stemming from these technologies. Therefore, emerging technologies and a need for future-proofing laws are like a chain of events that reinforce each other. Thus,  the author calls for a re-evaluation of the cognitive underpinnings and laws in order to better address cases of the manipulation of thoughts and behavior by  AI systems.

The author underscores that it is unclear how the specific term “subliminal techniques” was added to the text of the AIA. So the AIA, in particular Article 5, lacks a scientific foundation on that the term “subliminal” may have emerged from a combination of technological, psychological, and legal concerns. However, the critical word is “may”; the author assumptions due to his broad knowledge of the specific area. So, we can’t be sure what the term “subliminal techniques” means explicitly or why it is important; but why it is? On paper, Article 5 may not cause any ambiguities, however, it will be crucial in enforcing the prohibition once it enters into force, the author says.

The book takes significant time to analyze and point out further ambiguities and shortcomings regarding Article 5. The author starts by explaining the vagueness around the concepts of “subliminal technique.” When a term or a provision is vague, it causes interpretational dilemmas and question marks.  To quote the author, “…this word derives from the Latin “sub” and “limen, -inis,” and etymologically means” below the threshold of consciousness.” So, one of the first ambiguities about “subliminal” will be determining the level of interference of a person’s conscious or consciousness and how these terms relate to one another. 

It is clear that the author aims to expose in-depth meanings and interpretations of the AIA, particularly Article 5. The author gives the impression that he benefits from law and literature methodology, as Dworkin states that the understanding of the law can be improved with literature. Thus, the author’s interpretative approach to Article 5 aims to reveal the “true meaning” behind it. Hence, he presents the ambiguity of the provision in a literary plate to us, since he is interpreting the AIA, a piece of law, as a literature piece. Moreover, the author uses a very common literary technique called “the close reading method,” paying close attention to particular words and concepts in Article 5, such as subliminal or subliminal AI systems. Given his emphasis, it’s clear that the provision calls for a firm grasp of what qualifies as subliminal AI systems. Moreover, the provision seeks to embrace and predict the nature of future technologies, which compounds the difficulty of this technical question. Therefore, it’s important to speculate soon on the types of AI systems that will employ covert methods.

One of the compelling points the author underlines is that addressing the ambiguities and difficulties around AIA in general,  and specifically as related to Article 5, a broader perspective is important. The author is trying to highlight a rooted tradition of the EU; he offers a “contextual” point of view. One of the foundational interpretation techniques of the European Court of Justice contextually evaluating parts of the EU’s legal system requires looking at the functional relationship between a specific EU provision and the normative system it belongs to. However, one shall have a legal system that is more unified or, in other words, that has more “integrity.”  In earlier parts of the book, he mentions the Digital Markets Act, the Digital Services Act, and the AIA and how they comprise parts of the bigger puzzle. He later suggests that the EU’s legal system shall be seen as a whole, in parallel with the development of other related legal areas. In essence, the author emphasizes the application of a contextual interpretation, a technique employed by the European Court of Justice.

Finally, throughout the book, the author mentions very interesting and sometimes overlooked concepts regarding privacy, such as “mental privacy,” “decisional privacy,” and “cognitive liberty,” or “the right to privacy of the mind.” The author also sees multi-dimensional privacy, an understanding of the right to privacy conceptually reaches through different layers stemming from fundamental rights and freedoms.

3. Scientific Aspect

The necessity for the adoption of multi-dimensional privacy is due to the manipulative nature of subliminal techniques and their proven impact on freedom of thought. These techniques touch the fundamental right to freedom of thought as the primary, but not the only, fundamental right in matters of the mind. Considering recent scientific studies and the revised concept of the unconscious, where only 5% of our cognitive activity is conscious, and the remaining 95% is unconscious, the author stresses that if AI meets subliminal techniques, it has the potential to manipulate humans to an unpredictable degree. In other words, the author argues that the search for constructive responses of the law should start in the mind because many of these new technologies not only determine our lives in the so-called “black box society,” but also have the potential to reach the hidden depths of the “black box” of our mind. The author, therefore, speaks of the privacy of our conscious mind as well as our unconscious, reminding us again of the need for privacy in all respects and the necessity to look at freedom of thought from a multi-dimensional perspective. 

The author later posits that whoever is responsible for these developments will also be responsible for the human mind (a significant part of it) and the future itself when paired with big data and other technologies, tools, or applications. The author, therefore, frequently reminds us that there needs to be effective regulation in order to prevent such an occurrence, yet for this to be the case, open-ended terms that could create deadlocks must be avoided. The author recapitulates the jeopardies caused by these deadlocks holistically: “A major source of the persistent controversy revolves around the proliferation of competing terminology and lack of clarification concerning what the terms precisely mean, which leads to competing claims and apparent contradictions within the literature.”

Subsequently, the author reminds us that subliminal techniques are nothing new under the EU’s Sun. They have been on the EU regulatory agenda since 1989, but case law still remains rather weak. This contrast makes the author question why the situation abruptly differs when it comes to AI. Following this contrast, the author adds another paradoxical dimension: Despite the fact that there are a plethora of regulatory bodies set up to monitor such techniques, after extensive research and email correspondence with these bodies, documented examples of subliminal advertising are yet to be found. However, he underlines that the reason behind this axiom is not the absence of such a problem in advertising, but the lack of oversight in this sector. He also notes that regulators do not monitor subliminal techniques on an ongoing or permanent basis, but rather carry out random checks. Considering the current age of digitalization and 24-hour stories, the author poses a fair caveat: Well, then, aren’t these random checks ineffective and unnecessary? Especially in the face of constantly updated and changing new media, do such ineffective monitoring actually serve the law?

Here we would like to refer to one set of correspondence as it is quite thought-provoking in the context of possible/plausible future manipulations through these techniques. The author came across a case of alleged subliminal advertising for dog food and learned that tones that are not perceptible to humans but can be perceived by dogs were used to promote the product. The case was unfortunately not investigated further as the addressee of the prohibition should be humans, not animals; therefore, no further details can be found in the book. This raises another problem that the author emphasizes numerous times in the book: The fewer the number of cases and jurisprudence where such techniques are applied, the weaker the possibility to analyze the implementation and enforcement of the law. Without an enforcement mechanism and case law to monitor the use of subliminal techniques for AI, the AIA will remain a dead letter law. Referring back to our dog case, if the dog’s reaction to the advertisement persuades its owner to buy the advertised product, could this be considered as an indirect manipulation of humans? In other words, can’t we say that the advertisement is still clandestinely manipulating the owner/human?

The author also emphasizes the importance of senses for a human brain to process subliminal techniques no matter whether they target the conscious or subconscious mind. The brain relies on its five senses as its supplementary way of gathering information about its surroundings. If the senses aren’t functioning correctly, the brain cannot take in information, the imagination and emotions cannot process it, reasoning cannot make a call, and memory has nothing to hold on to or discard. Stated another way, each of the senses has its own unique characteristics, but the senses also interact with one another and therefore have the potential to provide a unified or multi-sensory experience that is greater than the simple addition of the perceptions provided by each individual sense. Yet again, there is a tendency to prioritize vision over other senses due to its high information transmission rate. This brings to mind a quote by the proverbial mind bender Leonardo Da Vinci: “An average human looks without seeing, listens without hearing, touches without feeling, eats without tasting, moves without physical awareness, inhales without awareness of odor or fragrance, and talks without thinking.” To have a multi-sensory experience, one should therefore avoid the hierarchy of senses; the same applies to the law since it is highly criticized for being ocular-centric. 

Considering that AI (also its subliminal techniques) can be used both on its own and through or in conjunction with other technologies, when Extended Reality technologies enter the picture, weighing ocular-centrism in the context of the law should be reconsidered. This is especially so due to developments in recent years (odor machine, synthetic tongue, so on so forth) and the fact that the ultimate goal of Extended Reality technologies is to provide a multi-sensory/synesthetic experience in a virtual environment. Therefore, the relationship between the sensory organs and the brain should be viewed with more importance.

Given the unprecedented capabilities of subliminal techniques and AI and their impact on unexplored regions of the human mind, it is easy to agree with the author’s call for representatives from interdisciplinary fields to come together to work on effective enforcement. Yet because of the relevance of subliminal techniques to the human mind, its intricate relationship with the senses, and the attempts to mimic these senses by emerging technologies, it is obvious that more experimental, dynamic, and unifying laws should be seized.

To quote the author, “It is first important to consider the wider context of the AIA and its relationship to other existing (as well as planned future) regulatory acts such as, notably, the GDPR, the PLD, the DMA, and the DSA. At the same time, it is important to rethink, in a creative way, the interaction between different legal fields such as competition law, intellectual property law, and international trade law or, to put it simply, the coherence and consistency of the legal system as a whole.”

In summary, Rostam J. Neuwirth’s latest book rigorously illustrates the need for synesthesia, or unity among the senses. Harmonization between the law and cognitive/computer/neuroscience is, according to Neuwirth,  a must and not a might. 

* Rostam J. Neuwirth is a Professor of Law / Head of the Department of Global Legal Studies at the University of Macau — Specializing in Generalization, Applied Philosophy, and Oxymora

[1] Rostam J. Neuwirth, “The EU Artificial Intelligence Act Regulating Subliminal Al Systems” (2022, Routhledge).

[2] European Commission, Proposal for a Regulation Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act), COM (2021) 206 final (21 April 2021) [AIA].

[3] Explanatory Memorandum of the AIA, supra note 1, 15.