Enhancing or Jeopardizing Human Creativity? Will Humans Be Able to Defend Themselves Against AI Superpowers in an Age of Ethics Washing and Law Washing?

Abstract

I recently introduced the concept of eco-cognitive openness and situatedness to explain how cognitive systems—human or artificial—dynamically interact with their environments to generate information and creative outputs through abductive cognition. Humans display high eco-cognitive openness, integrating tools and cultural contexts through “unlocked strategies” that also enable exceptional creativity. By contrast, generative AI like LLMs operates via “locked strategies” based on pre-existing datasets with limited real-time interaction, which constrains higher creativity. Although LLMs surpass humans in many cognitive tasks, they lack the openness required for truly advanced abductive performance. Notably, most human cognition is repetitive and imitative—humans themselves often resemble “stochastic parrots.” In this sense, LLMs reveal human intellectual poverty more than they expose flaws in artificial intelligence. I will illustrate how LLMs can act as powerful enhancers of human performance while simultaneously threatening our most distinctive prerogative: creativity. Future human–AI collaboration could expand our eco-cognitive openness, but demands vigilant oversight to counter bias and so-called overcomputationalization. GenAI can serve as an epistemic mediator toward unlocked creativity only if humans maintain agency and embed its outputs in broader socio-cultural frameworks. My greatest concern is that ethical and legal safeguards will remain ineffective in practice, resulting in mere “ethics washing” and “law washing” without genuine enforcement.

1. Introduction: Human Creativity in Jeopardy? Generative AI as Enhancer and Threat

A fundamental philosophical and anthropological question arises in an era where generative artificial intelligence has quickly acquired remarkable cognitive powers: will humans still be able to defend their most distinctive prerogative—creativity—against the superpowers of AI systems, or are we witnessing the subtle erosion of that very capacity? Large Language Models and other generative tools undoubtedly improve human performance on a variety of tasks, but they also reveal an unsettling mirror: the majority of everyday human cognition is repetitive, imitative, and pattern-driven, resembling what has been aptly called “stochastic parroting”. According to this viewpoint, the remarkable accomplishments of generative AI highlight the pervasive atrophy of true abductive and creative intelligence in modern human agents rather than solely exposing technology flaws. This conflict is examined in this essay using the conceptual frameworks of situatedness and eco-cognitive openness. It contends that comprehending both the possible improvement and the potential threat to human creativity depends on contrasting the “locked strategies” of current generative AI with the “unlocked strategies” of high-level human cognition. We can only expect to retain genuine epistemic agency in the face of more potent artificial systems by maintaining and fostering our eco-cognitive openness while avoiding the dangers of overcomputationalization, ethics washing, and law washing.

2. “Eco-Cognitive Openness and Situatedness” and Generative AI

“Eco-cognitive openness and situatedness”, a central concept in my philosophy of cognition, which I introduced in [1,2], helps explain how cognitive systems, whether artificial or human, can dynamically interact with their surroundings to produce information, theories, and favor creative outputs. Through abductive cognition (the process of formulating hypotheses to explain observations)¹, “Eco-cognitive openness and situatedness” highlights how external resources—such as tools, social interactions, cultural artifacts, or physical environments—shape human cognition. The concept of eco-cognitive openness and situatedness also offers a framework for assessing how generative AI interacts with its environment and whether it is capable of producing flexible, adaptive “meaning making” in the context of artificial intelligence. In the sections that follow, I will examine eco-cognitive openness in human cognition, its consequences for artificial intelligence (particularly generative AI), and how it relates to the concepts of locked and unlocked strategies I recently introduced. Moreover, as we will see, when applied to generative AI and Large Language Models (LLMs), eco-cognitive openness also highlights the importance of considering the complex interplay between these AI systems, their human users, and the broader ecological context in which they operate.

2.1. Defining Eco-Cognitive Openness

The ability of a cognitive system to exploit external resources in addition to internal processes to solve problems, develop ideas, or produce innovative results is related to what I called eco-cognitive openness and situatedness, as already outlined. This idea is based on my eco-cognitive approach to abduction, which holds that cognition permeates the environment rather than being limited to the mind through: (1) external mediators, resources that support reasoning, such as notes, diagrams, artifacts, and language itself; (2) environmental interaction as active involvement with the social or physical world to test theories or obtain new information; (3) contextual adaptability as the capacity to adjust cognitive processes in response to dynamic, real-world inputs, favoring the production of original or creative results.

For instance, a scientist using a telescope and other diverse cognitive resources to study some aspects of the stars exhibits an ideal example of eco-cognitive openness’ capability to enable new discoveries by integrating external resources and environmental information into the internal thinking process. On the other hand, a system with modest or even absent eco-cognitive openness operates inside a constrained framework that limits its adaptation and flexibility; this means that eco-cognitive openness refers to the idea that social and material environmental contexts have a strong impact on human cognitive systems, in addition to the role played by their internal structures. It highlights the dynamic, interacting nature of cognition and how it is integrated into a broader system. This comprises the following components: (1) engagement with the environment—people “constantly” engage with their surroundings, and these interactions influence how they learn, think, and process information; (2) because cognitive functions are closely linked to the body and sensory experience, our sensory encounters with the outside world influence our cognitive processes, that are, as it is said, “embodied”; (3) just as our thinking is impacted by our interactions with others and the common knowledge we share, human cognition is also influenced by the multiplicity of our social settings, cultural contexts, and language; (4) people are able to adjust to their environments in a variety of ways, incorporating new contexts and social dynamics into their cognitive processes, thus learning from novel experiences.

In essence, eco-cognitive openness and situatedness highlights how the human mind interacts with the environment, suggesting that cognition is flexible and responsive to new information, resources, and interactions with the outside world. I would also add that my concept of eco-cognitive openness and situatedness seeks to integrate and broaden the well-known 4E framework (embodied, embedded, extended, and enactive cognition) rather than replace it. It fully incorporates those perspectives: cognition is embodied (rooted in sensory-motor experience), enactive (constituted through active sensorimotor coupling), embedded (very dependent on environmental scaffolding), and extended (off-loaded onto external objects and mediators). In particular, my eco-cognitive approach adds an explicit focus on abductive cognition as the dynamic engine that takes advantage of this openness and situatedness. It highlights how agents, whether human or artificial, manage ignorance and produce innovative hypotheses by optimizing their situatedness in the cognitive niche through both internal processes and external manipulative actions. To put it briefly, eco-cognitive openness defines the epistemic approach that enables high-level creativity, whereas the 4E explanations address the structure of cognition.

2.2. Eco-Cognitive Openness in Human Cognition

Higher forms of human creative abductive intelligence are characterized by good “eco-cognitive openness and situatedness”². This refers to the fundamental nature of the human brain as an “open” or dissipative system, continuously coupled and attempting to achieve equilibrium with its environment. This constant interaction and openness to new, unconstrained information and experiences is the physical basis for human creativity and “unlocked” cognitive strategies. It allows for a rich, adaptive, and situated understanding of the world, going beyond pre-defined rules or data. Human cognitive agents have at their disposal a high level of eco-cognitive openness because they can: (1) integrate multimodal inputs, integrate cultural knowledge, social interactions, and sensory information (visual, auditory, and tactile) to generate abductive hypotheses. An artist may get inspiration, for example, in nature, in discussions, and in equipment such as software or brushes. (2) Actively manipulate the environment, whereby humans deliberately alter their surroundings (for example, by building models or taking notes) to externalize and refine cognitive processes. In this perspective, I have called manipulative abduction the case in which the environment is actively modified and cognitively exploited to generate hypotheses (see above; Note 1). (3) Generate powerful hypotheses: humans develop contextually rich, frequently cross-domain solutions through abduction, including developing new technology, cultural icons, and new knowledge.

In this framework, the openness enables people to develop what I call “unlocked strategies” (see below), in which cognition is adaptive, versatile, and able to overcome predetermined limitations. A poet writing a metaphor, for instance, creates meaning that is universal by combining environmental cues, cultural symbols, and personal experiences. The readers might want to know whether there is any established criterion for abductive consequence to distinguish between strong and weak abductive reasoning: since abduction is intrinsically defeasible and ignorance-preserving, my eco-cognitive framework lacks a single, strict formal criterion. However, a helpful pragmatic-epistemological distinction arises from the nature of the generated hypothesis and its relationship to the available evidence and cognitive resources. Strong abduction frequently results in hypotheses with greater “plausibility” or “loveliness” in context and can swiftly lead to explanatory or predictive power. It also tends to produce hypotheses that are more robust, consilient (explaining a wider or more significant set of data), and open to quick additional testing or manipulative elaboration. In contrast, weak abduction produces more tentative or “hard-to-negate” hypotheses, which are sometimes barely accommodating of the unexpected truth. They also have lower initial consilience or are more susceptible to withdrawal due to additional knowledge or negation-as-failure. These are still reasonable places to start when conducting more research, particularly in creative or manipulating contexts when knowledge is scarce. Therefore, the primary “criterion” is eco-cognitive optimization rather than just logic: how effectively the abduced hypothesis improves the agent’s situatedness in his cognitive niche, provides novel actions or external representations, and reduces ignorance. To put it briefly, strength is not just determined by a specific probability or consistency threshold, but rather by how productive it is in the specific cognitive niche. This perspective keeps abduction dynamic and open to both human and artificial cognition.

3. Locked and Unlocked Strategies

The relationship between generative AI and my concept of locked strategies is an appealing intersection of computational creativity and cognitive philosophy. I have introduced locked and unlocked strategies to describe cognitive processes in humans and machines, particularly in deep learning systems like AlphaGo/AlphaZero. In this article, I will also analyze how generative AI aligns with or diverges from my concept of locked and unlocked strategies, drawing on my ideas and the broader context of AI cognitive roles.

3.1. Locked and Unlocked Strategies

I distinguished [3] between locked and unlocked cognitive strategies according to their degrees of eco-cognitive openness, which is the capacity to interact with and adjust to the environment in a flexible manner while coming up with ideas or innovative outputs:

Locked strategies describe cognitive systems that function inside constrained, pre-established data domains, such as computational AI programs like AlphaGo/AlphaZero. They excel in specific tasks but lack broader adaptability due to limited interaction with the external environment. Because locked strategies, in the case of Go games, both human and computational, rely on rigid, finite, pre-structured data and rules, they restrict their capacity to provide innovative or contextually diversified solutions, leading to low-level kinds of hypothetical creative cognition. AlphaGo’s impressive performance in Go, for instance, is limited to the finite entities and rules of the game, meaning that its abductive reasoning (the creation of hypotheses) is “locked” to that domain, due to the fixed rules, board, and pieces; this restricts the AI to a predefined set of possibilities. Strategies that generate moves are of course infinite and rich, but locked.

In contrast, my studies on abductive cognition [1] led me to characterize human cognition as always potentially possessing various degrees of “eco-cognitive openness”. Abductive reasoning can become richer and more creative because humans can exploit “unlocked strategies” that involve adapting to new environments, creating novel rules, or redefining contexts. AlphaGo/AlphaZero’s inventive moves, such as Move 37 against the champion Lee Sedol, for instance, are inventive within the confines of the game but fall short of the greater flexibility that humans display in situations with more freedom. I contended that AlphaGo/AlphaZero’s, but also in the case of Go game played by humans, dependence on a locked system restricts its cognitive outputs to a more constrained (low-level) range of creativity than in the case of human “unlocked” methods, despite its ability to explore vast move combinations through deep learning and reinforcement learning.

Unlocked strategies: these are typical of human cognition, where the creation of richer, more flexible, and high-level creative hypotheses is made possible by the full exploitation of eco-cognitive openness. Humans develop powerful creative outcomes, such as new artistic expressions or scientific discoveries, by utilizing cultural contexts, external settings, and multimodal interactions (such as visual, manipulative, or social clues). Unlocked strategies can adjust to a variety of open-ended situations and are less restricted by preset restrictions. In contrast to high-level examples of creative human cognition, I argue that AI locked methods are limited by their lack of full eco-cognitive openness, despite their great accomplishments. When it comes to humans, we encounter:

• Interaction with the world in terms of interaction through their senses (hearing, touch, vision, etc.), which have a direct impact on cognition.
• Cultural transmission related to the inheritance and adaptation of what we inherit and adapt cultural knowledge, ideas, language, and social practices that are not always explicitly stated but are frequently unconscious and have a significant impact on how we view and engage with the world; our cognition is dynamic and context-sensitive.
• Embodied cognition regards the facts that our thoughts are entwined with the physical world because our bodies, thanks to movements, emotions, and sensory experiences, are the foundation of our cognition. For instance, there is a close relationship between how we think and how we utilize tools or gestures.
• High adaptability of humans when faced with novel environments and challenges—they may quickly alter their perspectives or methods of problem-solving.

3.2. Eco-Cognitive Openness in Generative AI—GenAI Is Unable to Interact with the External World in Real-Time or Adjust to Completely New Domains Without Retraining

Examples of generative artificial intelligence (GenAI) with restricted eco-cognitive openness include large language models (LLMs), picture generators, and systems like AlphaGo/AlphaZero, which are consistent with my idea of closed strategies. This is how:

Limited environmental interaction:

• Limited environmental interaction compels GenAI to make use of pre-existing information (such as text corpora and image libraries) in place of real-time environmental inputs. A LLM, for example, does not actively interact with the environment like a human might; rather, it creates text based on patterns it has learnt during training. As a result, its “mental processes”—so to speak, through anthropomorphization—are “locked” to a static data environment.
• Limited manipulative abduction: GenAI is unable to independently engage with or alter its surroundings, in contrast to humans who use symbolic or physical instruments to develop ideas (such as sketching to brainstorm). Instead of using the iterative input from outside mediators that humans do, its outputs are produced algorithmically.
• Weak abductive reasoning: GenAI generates believable outputs (such as realistic image or a coherent story) in a manner similar to low-level human creative abduction. Without “retraining” or fine-tuning, it cannot naturally integrate new environmental elements, such a sudden change in culture or a physical event. Without “retraining” or fine-tuning, it is unable to spontaneously incorporate novel, environmental factors, such as an abrupt cultural shift or physical occurrence.

For example, DALL-E may generate images in response to commands, but rather than directly interacting with the dynamic complexity of the external environment, its “creativity” is limited by the patterns in its training data. This is, in contrast, for example, to a human artist who might alter her works in response to public comments or observations or other sensorial inputs arising in real time.

Using sophisticated neural architecture—such as transformers (the core mechanism behind models like ChatGPT), GANs (Generative Adversarial Networks, in which two neural networks compete to generate highly realistic synthetic data), or VAEs (Variational Autoencoders, which learn compact latent representations of data and can then decode them into new, plausible outputs)—generative AI systems, which include models like ChatGPT (OpenAI’s conversational large language model), Grok (xAI’s generative AI system), DALL-E (OpenAI’s text-to-image generator), or Midjourney (a widely used AI-powered image creation tool), learn patterns from large datasets to produce more or less novel material, such as text, images, audio, or code.

This perspective claims that there are several significant facets of GenAI that are relevant to my eco-cognitive and abductive epistemological framework. GenAI creates outputs that often seem unique but are based on learned patterns by interpolating and extrapolating from training data (e.g., emulating Dickens’ writing style or generating realistic images). Contextual limitations are crucial: within the confines of its model architecture and training data, GenAI functions. Even if it can generate a wide range of outputs, its “creativity” is shaped by the data it was trained on and the prompts it receives, which limits its ability to engage with the outside world in real-time or adapt to entirely new domains without retraining. Of course, the lack of “agency” is patent because GenAI automates content creation but lacks the intentionality or environmental interaction that characterizes human cognition. Its outputs are often evaluated for realism, coherence, or diversity rather than true novelty or contextual understanding.

3.3. Generative AI and Its Locked Strategies

Generative AI aligns closely with my concept of locked strategies due to its constrained eco-cognitive openness and reliance on predefined data environments. As already said, we face a situation of limited eco-cognitive openness because GenAI systems, like AlphaGo/Alpha Zero, operate within a “rigid scenario of data” (see above). They are trained on large but specialized datasets (such as picture datasets or text corpora for LLMs). We have to clearly understand that this is also a limit of their ability to generate hypotheses or creative outputs (abductive cognition, as I contend) to the patterns and structures seen in their training material. A language model such as ChatGPT, for example, can surely produce coherent text, but its responses are not based on new conceptual leaps or real-time environmental interaction, but rather on statistical patterns in its training data. This mirrors my description of locked strategies, where AI creativity is “poor” compared to various extreme high-level kinds of human cognition due to its lack of “dynamic” engagement with the external world. The constraints are basically domain-specific because GenAI is often specialized (e.g., text generation, image creation) and struggles to generalize beyond its training scope without fine-tuning. For instance, unlike humans, who are able to combine multimodal experiences (visual, tactile, and social) to produce unique solutions, a text-to-image model such as DALL-E cannot spontaneously adapt to a new modality (such as audio generation) without extensive “reprogramming”. During my studies on cognition, I have always made that abduction is essential to creativity; that is, completely new hypotheses are generated, then later on are eventually also confirmed and accepted as true. While GenAI does engage in some abduction cognition by producing believable results based on data patterns (like finishing a story or creating a product), this is a locked form of abduction because it lacks the human ability to generate open-ended, full context-sensitive hypotheses. Let me reiterate: unlike human high-level creative inventions (like in the case of scientific discovery) that transcend various domains, AlphaGo/AlphaZero’s surprising and successful play versus Lee Sedol (see again above, play 37) was undoubtedly inventive, but only inside the Go framework, characterized by a finite, rigid number of rules and stones, and a standard board—infinite strategies, but locked in a finite environment.

This distinction between automated and human cognition affects how individuals solve problems, come up with concepts, and engage with the external world. Because of their eco-cognitive openness (and appropriate situatedness), humans are more suited for what I have called strong creative “knowledge-enhancing abductions”, which involve navigating unexpected information and developing understanding in a changing environment. My usage of the concept of eco-cognitive openness essentially highlights the fact that human intelligent creativity is “inherently” anchored in our robust, ongoing, multifaced, and extremely open engagement with our surroundings—a feature that current generative AI, despite its remarkable powers, basically lacks.

3.4. Eco-Cognitive Openness in LLMs (Generative AI Systems)

In summary, generative AI models like LLMs, for example GPT or Grok, are made to mimic human understanding and produce results depending on the training data. However, when compared to humans, the interactions of these models with the environment differ in a number of significant ways. First of all, LLMs do not “experience” the world through their senses (hearing, touch, sight, etc.). They do not engage with the physical world in the same manner as people do. Large datasets and statistical patterns, rather than sensory or physical interaction, are the foundation of their “learning”. They lack embodiment because LLMs do not absorb information in an embodied manner, since, in contrast to humans, as already said, they lack bodies and sensory organs. LLMs substantially rely on context, but this context is simply informational (e.g., prior text, data patterns); their outputs are solely dependent on mathematical computations, with no foundation in physical experience. Beyond the data they have been exposed to, they do not interact with real-world environments. Their “learning”, entirely data-driven, lacks the adaptive, real-time open environmental interaction that defines human cognition and is solely data-driven. Even if LLMs are capable of processing huge volumes of data generated by humans, they are unable to “truly” comprehend culture or social situations in the same way that people do; they can provide outputs that are culturally relevant and are trained to identify patterns in text, but they are not based on social or cultural experiences. Due to their lack of lived experience, their “openness” is restricted to the textual information they come across, which excludes both lived cultural experience and active social participation and, consequently, unlike humans, LLMs are not able to adapt in real time, such that to adjust to new knowledge, they need to be retrained or fine-tuned on fresh data. They can mimic learning from previous interactions, but, unlike humans, they do not change their behavior on their own in response to novel environmental situations.

The main distinctions between LLMs and humans in terms of eco-cognitive openness can be schematically indicated as follows (Table 1):

Table 1. The main distinctions between LLMs and humans in terms of eco-cognitive openness.

Aspect	Humans	LLMs
Environmental Interaction	Are strongly interactive with the external world via sensory input and body movements.	Interaction is purely text-based, with no sensory engagement.
Embodiment	Cognition is conditioned by bodily experiences and movements.	LLMs do not possess a body and sensory apparatuses: cognition is purely abstract.
Cultural Context	Social and cultural context shapes cognition through lived experience.	Can simulate cultural context from data, but are not open to it like humans.
Adaptability	Can learn and adapt to new environments and contexts in real time.	Adaptability is limited and requires retraining to adjust to new contexts.
Learning	Learn from experiences gained through senses, social interactions, and new challenges.	Learn from huge datasets but lacks real-world sensory interaction.

4. Potential for Enhancing Human Eco-Cognitive Openness Through AI: Divergences and Potential for Unlocked Strategies

Although GenAI is now based on, locked strategies there are ways to make it more eco-cognitively open and get closer to unlocked strategies: thanks to human–AI collaboration and co-creation, AI can act as what I called a cognitive (or “epistemic”, in the case of scientific cognition) mediator, for example fostering human creativity by combining with human feedback loops like co-design procedures or iterative prompting. To partially offset AI limitations, architects might use AI tools to produce design possibilities and then enhance them depending on cultural or environmental aspects. Eco-cognitive openness emphasizes the interplay between the human mind and the surrounding environment. This can be extended to how humans interact with generative AIs. In collaborative tasks, humans and AIs “co-create” in ways that expand and shape cognitive processes. LLMs, when used in partnership with human users, become tools that extend human cognitive abilities. For example, an AI might suggest new ideas, assist in writing, or help with problem-solving. Human cognition is thus also “open” to AI suggestions, leading to a richer, more dynamic creative process than either the human or the machine could produce alone.

In order to interact more dynamically with the environment and become more open, future AI systems may integrate sensors or real-time data streams (from social media or IoT—Internet of Things—devices, for example). More eco-cognitive flexibility would be exhibited, for example, by an AI that modifies its outputs in response to current cultural trends on platforms such as X. Generative AI and LLMs are designed to collaborate with humans, generating text, images, or other outputs that can be used, modified, or extended by humans. This cooperative approach creates a shared cognitive environment by obfuscating the distinctions between artificial and human cognition, and generative AI and LLMs can improve human eco-cognitive openness by automating specific cognitive processes or giving people new tools for creativity and problem-solving, creating new opportunities for eco-cognitive interaction. Finally, multimodal AI developments (such as merging text, image, and audio processing) may make it possible to integrate various inputs in a way that is more human-like; for example, mimicking the manipulative abduction that people carry out. To avoid being locked by algorithmic limitations, deliberate design would still be necessary. As already anticipated, according to my eco-cognitive framework, manipulative abduction is a type of abductive reasoning where the agent actively manipulates external objects, artifacts, or representations—diagrams, models, instruments, or even the physical environment itself—instead of just “thinking” internally in order to produce or choose tenable hypotheses. It is “thinking through doing”, wherein the activity in the outside world generates additional knowledge that would not otherwise be available, transferring the cognitive burden and creating new chances for abduction. Manipulative abduction is inherently contextual and action-oriented, in contrast to simply theoretical (sentential or model-based) abduction.

I see the following type of development when I propose that sophisticated multimodal generative AI (integrating text, picture, audio, and potentially sensorimotor input) may start to replicate manipulative abduction: imagine an AI system that can interact iteratively with an external digital or simulated environments—rearranging visual elements on a canvas, altering a 3D model in real time, running physical simulations, or even controlling a robotic arm—instead of just producing a static image or text in response to a prompt. This would allow the system to “see what happens” and generate more creative or explanatory hypotheses. When a designer collaborates with an AI on a challenging engineering problem, for instance, the system could actively manipulate a shared digital prototype by rotating parts, testing structural stresses, and changing materials. Instead of relying solely on linguistic inference, this hands-on exploration could produce new potential solutions. In the same way that a scientist manipulates a physical model or a schematic to find unanticipated affordances, the AI would therefore take advantage of external representations, or of what I call epistemic mediators. In comparison to human eco-cognitive openness, such systems would still be restricted, but they may get closer to an action-based, really placed type of abductive intelligence. Investing in hybrid architectures that closely connect generative models with external loops of manipulation and feedback—whether simulated or embodied—while maintaining the human user’s pivotal position in the cognitive niche is something I would constructively advise. By improving its ability to “think through doing”, this might progressively lessen the existing eco-cognitive constraints of generative AI.

In summary, while GenAI predominantly exhibits locked strategies, there are nuances where it approaches (but does not fully achieve) unlocked characteristics. First of all, we face an effect of exploratory potential because some GenAI applications, such as those used in innovation management or idea generation, show explorative capabilities, producing diverse outputs that can inspire human creativity (e.g., generating novel product designs or process descriptions). However, this exploration is still bound by the model’s training data and lacks the spontaneous, environment-driven adaptability of human cognition. Furthermore, given the fact that human unlocked strategies emphasize the interaction with external cognitive mediators (tools, objects, or social systems), GenAI itself can act as such a mediator, augmenting human creativity by providing raw material (e.g., text or images) that humans can further refine. By using human agency to contextualize AI outputs, this human–AI synergy, as demonstrated in collaborative design processes, partially closes the gap toward unlocked methods. The AI itself, however, is still locked because it is unable to interact with the environment outside of its predetermined parameters on its own.

5. LLMs Are Strong Cognitive Devices That Can Help or Jeopardize Human Creativity

We may now investigate how LLMs operate as potent but conflicting cognitive tools, building directly on the difference between the locked strategies of generative AI and the unlocked, eco-cognitively open nature of high-level human cognition described in the prior sections. As illustrated in the previous sections, it is a high-level of eco-cognitive openness that grants the fantastic creativity many human beings sometimes exhibit. In contrast, generative AI, such as large language models and image generators, operate by taking advantage of “locked strategies”, relying on pre-existing datasets with limited real-time environmental interaction, resulting in constrained creativity. These systems can originate creative results, but the absence of human integral eco-cognitive openness inhibits high-level creative abductive performances, which are instead still the preserve of human beings (at least the most gifted ones). Even if limited in eco-cognitive openness, LLMs surpass, in the case of various specific cognitive tasks, human beings: indeed, most human cognition is repetitive, imitative, not creative, critical, or unpredictable; it seems that people are essentially “stochastic parrots” (a recent lucky academic qualification of LLM programs)³ themselves who never stop repeating patterns they do not examine or comprehend. It is in this perspective that we can see how LLMs reveal the unsettling reality of human intellectual poverty rather than the shortcomings of artificial intelligence⁴.

I am referring to a deeper anthropological and epistemological state when I note that humans themselves frequently act like “stochastic parrots”, continuously repeating language and cognitive patterns they do not truly study or grasp. Large language models’ success does not primarily reveal a technological shortcoming, but rather serves as a potent mirror that highlights an unsettling reality of modern human cognition: the widespread atrophy of true abductive and creative thinking in favor of social parroting and pattern completion. There are a number of frightening implications. First, a lot of what appears to be “thinking”, in a time of rapid information consumption, social media echo chambers, and performative speech, actually consists of sophisticated kinds of repetition and recombinations of pre-existing language pieces. Without profound comprehension or epistemic accountability, humans, like LLMs, often produce products that seem reasonable and cohesive. As a result, intellectual poverty is concealed by what I would refer to as “cognitive mimicry”. Second, these phenomena significantly impair our ability to use high-quality abductive reasoning, which is the foundation of scientific advancement, technological innovation, but also of good moral reasoning and deliberation.

The eco-cognitive niche of human agents gradually deteriorates when automatic pattern-matching takes the place of good cognitive abductions: fewer real surprises are observed, fewer original hypotheses are produced, and ignorance is no longer effectively managed but is instead covered up with fluent but superficial responses. Third, and probably most crucially, this mirroring effect should push us away from complacent criticism of artificial intelligence and toward a thorough self-examination. The true challenge is not to “fix” the machines, but rather to combat the increasing inclination in human cognition toward intellectual laziness, uncritical repetition, and the delegation of thinking to external systems, given how well LLMs mimic and surpass average human linguistic–cognitive performance.

To put it another way, the real danger is not that AI will outperform us, but rather that we are already starting to resemble the constrained cognitive engines we have developed. In this way, the disturbing truth revealed by LLMs is a diagnostic symptom of a larger cognitive and cultural crisis—the progressive deterioration of our ability to exercise genuine epistemic agency and creative abductive intelligence—rather than a judgment on technology alone. We can only develop educational, cultural, and technical practices that actively encourage deeper comprehension, manipulative involvement with the environment, and responsible information management in both people and the artificial systems we construct by recognizing this mirror. Even though it is painful, I think this thought is essential if we want to prevent a future where artificial intelligence and humans are both stuck in increasingly complex kinds of parroting.

Hence, LLMs are de facto strong intellectual devices that can help human beings’ cognitive performances, but that can also jeopardize their most important cognitive prerogatives, such as, for example, “human creativity”. Future AI advancements, like human–AI collaboration and real-time data integration, could enhance human eco-cognitive openness, but risks like bias and overcomputationalization highlight the need for human oversight to ensure meaningful, contextually rich outputs. We also know that other actual or possible damages can be generated, as it is incessantly illustrated in any AI ethics book. The problem is how to counteract all the negative effects: in the following short sections, I will finally elaborate on the fact that I sincerely worry that in our modern societies, the development of legal tools and ethical discourse to address the detrimental effects of the AI tools will not be able to advance to actual implementation. What repressive methods, or “violent arm of the law”, can be used to enforce ethical and legal frameworks? It appears highly likely that one could become entangled in both evanescent “ethics washing” and what is known as “law washing”.

6. Challenges and Ethical Considerations: Jeopardizing Human Discoverability?

I now discuss the wider threats to human discoverability and epistemic agency that result from the same restricted eco-cognitive openness that limits the creative potential of generative AI (as examined in Section 2, Section 3 and Section 4). Beyond the standard warnings currently obsessively traceable in every AI ethics book or article, I would like to stress the attention to a more basic possible threat. With respect to the possible dangers against human high-level creativity and so, to what I called “discoverability” that could be perpetrated by AI tools, I fundamentally have to warn of the risks of what I have called “overcomputationalization” [3], where locked AI strategies could erode human freedom and creativity by overly structuring decision-making. GenAI’s ability to automate tasks (e.g., content creation, cybersecurity) raises similar concerns, as it may reduce human engagement or lead to over-reliance, limiting eco-cognitive openness and thus discoverability. In sum, overcomputationalization can create an excessive dependence on “locked” computer systems and may limit opportunities for open and correctly situated eco-cognitive engagement, which in turn may stifle human innovation.

Improving AI eco-cognitive openness presents difficulties: for example, the “half-automation problem” highlights how GenAI’s partial autonomy can create complacency, undermining human agency. The half-automation problem refers to the challenges and inefficiencies that arise when a system or process is partially automated, combining human and machine efforts in a way that creates friction, errors, or suboptimal outcomes. It occurs when automation is implemented incompletely, leaving humans to handle tasks that are poorly integrated with the automated components, often leading to confusion, delays, or reduced performance.

Of course, AI’s reliance on data might reinforce biases even in the face of more transparency if the training environment is biased. An AI educated on global datasets, for instance, might nevertheless underrepresent voices from underprivileged groups, which would impact its cognitive output. According to my concept of overcomputationalization, artificial intelligence systems that are too autonomous may cause people to interact with their surroundings less. The eco-cognitive openness of human cognition itself may be weakened as a result. Building an AI that truly interacts with dynamic, unpredictable environments requires overcoming technical hurdles, such as real-time learning and ethical safeguards to prevent misuse (e.g., in misinformation). Energy consumption and the production of e-waste are two major environmental effects of the development and application of generative AI and LLMs. The necessity of taking these environmental effects into account and to work toward more sustainable AI techniques is further highlighted by the goal of maintaining human eco-cognitive openness.

I may draw the conclusion that my idea of eco-cognitive openness pushes us to consider the connections between AI systems, people, and the environment critically and to create AI systems that are more aware of their eco-cognitive and ecological implications. By adopting this viewpoint, we can endeavor to develop more cooperative, sustainable, and cognitively empowering AI systems that are advantageous to both people and the environment.

7. Ethics Washing and Law Washing as Nefarious Possible Outcomes

The practical ineffectiveness of present ethical and legal safeguards exacerbates these dangers to human creativity and discoverability, which brings us back to the previously described eco-cognitive and contextual elements. A related important issue of the AI threat to human freedom and creativity in abductive cognitive tasks has been recently underlined by Hildebrandt [6]. She investigated the interference between real-world navigation and recommender systems, with a particular emphasis on the role of intermediaries in their design and the limitations of the deductive and inductive reasoning that form the foundation of these systems. Implicitly referring to my eco-cognitive framework, which I described in this article, she addressed the capacity of human agents to creatively assimilate and interact directly with the physical and institutional components of their environments as the foundation of their real-world navigation. Hildebrandt continued by asserting that the positive outcomes of individuals who pay for recommender systems should not be surprising, as their objective is to increase sales or ad revenue. This relates to the well-known problems of filter bubbles, feedback loops, and echo chambers, and raises the question of whether recommender systems truly produce the desired outcomes while taking into consideration the often-illogical economic incentive structures that affect relevant design decisions. Hildebrandt concluded by talking about the important role that the EU legislative framework may play in reorganizing the financial incentives that influence these systems’ construction. Instead of restricting or displacing human agency, this would allow recommender systems that enhance or support it, so as to foster human creative abduction.

I agree with Hildebrandt that these important aspects of human abductive creativity must be preserved while also defending information freedom. In the book Discoverability [3], I discussed this necessity in detail. These demands ought to be seen as being on par with the freedoms of conscience, private property, individual liberty, and self-determination. How can the rules of our constitutional democracies protect people’s fundamental rights when profiling technology and current AI optimizers are not governed by traditional laws? Despite the Cambridge Analytica data scandal⁵, it seems that the enormous influence of internet firms has not decreased. The harm that the internet and other computational technologies bring to other people is often simply invisible (or untraceable), and we do not know how it shows up or, of course, have the legal tools to stop it.

Hildebrandt appeared to be offering some subtle suspicion regarding the constitutive fragile status of “cyberethical” reflections, which many contemporary authors consider frequently doomed to turn into instances of what is known as ethics washing. The term “ethics washing” is derived from the original “greenwashing”. A standard definition of “greenwashing” does not exist. In general, it involves asserting or promoting the idea that certain actions, goods, and services are more sustainable or ecologically friendly than they actually are. The exact definition of “greenwashing” will change depending on the kind of product and service, as well as between industries, authorities, and legal systems. Additionally, it might differ according to the individual making the claim; what one person values as a sustainability claim can be seen as greenwashing garbage by another. While many also use the phrase to refer to social and governance difficulties, some regard greenwashing as just an environmental matter. This lack of clarification is important because it makes it challenging for organizations to determine what constitutes greenwashing in reality and what measures to take in advance. Given that accusations may have serious negative effects on one’s reputation and finances, it is quite important.

As Tafani [7] (p. 12) pointed out, utilizing the current extensive discussions surrounding artificial intelligence (AI) “With a marketing strategy so widespread it even has a name (‘ethics washing’), tech giants try to escape from legal regulation by approving hundreds of similar abstract and empty declarations or ‘guidelines’ on AI ethical principles. They behave as though respect for individual rights were up to each individual’s good heart, hoping for a gesture of trust that one would not even grant to the butcher or the baker”. Tafani expressed even more negative observations, which may be excessively pessimist, but worth quoting: “Since machine learning systems are not transparent, the introduction of any human control can only serve to legitimise the use of systems that are actually out of control. The role of the human being can then only be to provide ‘human-washed’ predictions or, in the case of proven absurdities or injustices, a scapegoat” [5] (p. 7). Again, “My thesis states that ‘AI ethics’ narratives are based on imposture and mystification: on a false, deliberately deceptive narrative about what ML systems are and about what they can actually do, as well as on a misconception about what ethics consists of. These narratives spread mystification, whereby public issues of structural injustice, whose solution would be costly for tech giants, are substituted by science fiction, relegated to a few experts, whereby law is substituted by industry self-regulation” [7] (p. 3).

I myself greatly fear, even if more moderately, that in our contemporary societies, the ethical discourse and the production of legal tools to deal with the negative consequences of AI and other computational systems do not have a serious chance to move to actual implementation; by what repressive means, by what “violent arm of the law” can one proceed to enforce ethical and legal frameworks? The possibility of being trapped in both a form of ethics washing and what we can call law washing—think of the destiny of the EU’s recent legal framework concerning AI: the Artificial Intelligence Act, formally entered into force in August 2024 and subject to gradual implementation, with most provisions becoming applicable from August 2026 onward—seems highly likely to remain largely ineffective in practice.

Furthermore, upon assigning a job to an “artificial intelligence” system, one presumes that the system is genuinely capable of completing it, regardless of the assignment’s inadequacy or impossibility. As usefully observed by Tafani, putting your faith in the dependability of predictive optimization systems is akin to believing in the fallacy of AI functionality: it is implicitly assumed that, when entrusted to an automated statistic, moving from the available data about a given individual to a specific prediction about that individual is possible, as if by magic, even in the absence of any scientific basis because no causal link has been found between that data and the event or characteristic to be predicted. A “massively large and expensive computing infrastructure doing statistical analysis is” thus “discursively conferred the status of an enchanted object, closing” it “off to other forms of critique” [7] (p. 3).

Due to the fact that machine learning systems can now use automated statistics to accomplish restricted and specialized activities that were previously exclusive to humans, big tech corporations have jumped at the chance to provide an infinite number of “intelligent” goods and services. It is interesting to report here an extreme proposal regarding some especially puzzling AI optimization systems, that nevertheless is worthy not only of our puzzlement but also of our moderate consideration: “The adoption of predictive optimisation systems is incompatible with liberalism and the rule of law because it results in people not being recognised as self-determining subjects, not being equal before the law, not being able to predict which law will be applied to them, all being under surveillance as ‘suspects’ and being able or unable to exercise their rights in ways that depend not on their status as citizens, but on their contingent economic, social, emotional, health or religious status. Under the rule of law, these systems should simply be banned” [5] (p. 12).

8. Conclusions

Eco-cognitive openness draws attention to the differences between AI and human cognition, especially when it comes to generative AI’s closed methods. GenAI’s creative potential is limited by its data-driven, static frameworks, whereas humans use dynamic environmental interactions to generate flexible, context-rich meanings. However, AI may become more open and play a bigger part in global cognition with human–AI cooperation and improvements in real-time environmental integration. It seems that the difficulty lies in making sure AI enhances human eco-cognitive processes rather than replacing them, maintaining the depth of meaning-creation. Because generative AI relies on limited eco-cognitive openness and constrained data environments, it produces weaker types of creative abduction than in the case of high-level creative human cognition, which is consistent with the concept of locked strategies I have described in this article. The outputs of GenAI are based on statistical patterns and lack the adaptable, environment-driven hypothesis creation of unlocking human strategies, even if they can produce novel content and foster human creativity. However, when humans and AI work together, GenAI can act as an epistemic mediator, partially bridging the gap toward unlocked creativity. To avoid the risks of overcomputationalization, and the consequent possible jeopardizing of their high-level creative performances, humans must maintain agency over GenAI outputs, ensuring they are contextualized within broader socio-cultural frameworks. The article finally addressed the issue of “ethics and law washing”; I illustrated the reasons that explain why I sincerely worry that in today’s societies, the development of legal frameworks and ethical discourse to address the unfavorable effects of AI and other computational systems seems not capable of having the opportunity to move to actual implementation. What repressive methods, or “violent arm of the law”, can be used to enforce ethical and legal frameworks? It appears highly likely that one could become trapped in both evanescent “ethics washing” and what is known as “law washing”.