A new creative symbiosis

In today’s media ecosystem, the rise of generative artificial intelligence is a revolution comparable to the invention of printing or the arrival of the internet. Communication professionals, content creators and intellectuals face a historic choice: reject these tools in the name of a creative purity that is ever harder to defend, or bring AI into their work processes and develop a new form of creative symbiosis.

This article explores the second path, looking at how artificial intelligence can become an intellectual partner that amplifies human abilities, as long as we deal consciously with its built-in limits and, in particular, its algorithmic biases. The challenge is not only technical but deeply cultural and philosophical: how can we work with systems that reflect, often in a distorted way, our own social practices, prejudices and worldviews?

The anthropological side of artificial intelligence

AI as a cultural artifact

From an anthropological point of view, artificial intelligence systems are not neutral tools but complex cultural artifacts that embody the values, preferences and power structures of the societies that created them. Large language models (LLMs) have been trained on vast bodies of text that include books, articles, social media posts and web pages, a digital aggregate of human cultural production, with all its complexity and contradictions.

These systems therefore do not simply generate text: they reproduce, and sometimes amplify, the dominant discourses in the sources they were trained on. As the anthropologist Mary Douglas observed, every system of classification and representation reflects a specific social order. In the case of AI, that order tends to favor:

Western, English-centered perspectives
Urban, technologically advanced worldviews
Content produced by socioeconomically privileged groups
Historically dominant narratives rather than marginal voices or counter-narratives

The problem of algorithmic hegemony

The sociologist Pierre Bourdieu taught us that cultural power works by imposing categories of thought that seem natural and self-evident. Artificial intelligence risks becoming a new mechanism for reproducing cultural hegemony, presenting as universal or “objective” perspectives that are in fact rooted in a particular history and geography.

When an Italian content creator uses an AI model trained mainly on English texts, for example, they unintentionally import linguistic structures, cultural references and patterns of argument typical of the English-speaking world, contributing to what the Italian linguist Raffaele Simone has called the “expressive flattening” of non-dominant languages and cultures.

AI as a distorting mirror

Artificial intelligence systems also work as cultural mirrors that give us back an image of ourselves and our societies, but they are sometimes distorting mirrors, amplifying some aspects and minimizing others. By studying AI responses to different prompts, we can see which narratives, stereotypes and prejudices are so deeply rooted in our cultural production that they emerge spontaneously in what the algorithms generate.

As recent studies have shown, these systems tend to reproduce gender stereotypes (associating women more readily with caring roles and men with leadership), racial bias (portraying white people more positively) and socioeconomic prejudice (treating the lifestyles and aspirations of the urban middle class as the norm).

Mapping the biases: a critical taxonomy

To use AI effectively as a creative collaborator, you need a thorough understanding of its built-in biases. Here is a taxonomy of the main biases that affect today’s generative systems:

1. Representation biases

These concern who and what is represented in the training data, and how:

Demographic overrepresentation: some populations (typically white, middle-class Western men) are overrepresented in the data, leading AI to generate content that favors their perspectives.
Stereotyping: a tendency to reproduce stereotyped associations between social groups and particular traits, behaviors or roles.
Selective invisibility: some groups or topics are systematically underrepresented or entirely missing from the training data, leaving the AI “blind” to those realities.

2. Linguistic and cultural biases

These come from the dominance of certain languages and cultural traditions in the datasets:

Structural English-centrism: the structures and expressive conventions of English shape what is generated in other languages too.
Asymmetric cultural translation: concepts and references from the dominant culture are treated as universal, while those of other cultures are marked as specific or “exotic.”
Stylistic impoverishment: a tendency to flatten stylistic variety toward a standard, fairly formal register typical of English-language academic and journalistic writing.

3. Epistemic biases

These concern what counts as valid knowledge and how it is organized:

Selective scientism: privileging Western scientific knowledge over traditional knowledge or alternative epistemic approaches.
Presentism: a tendency to interpret historical phenomena through today’s categories and sensibilities.
Contextual inattention: a tendency to generalize statements without properly considering the specific contexts in which they may or may not hold.

4. Time-distribution biases

These come from the uneven distribution of data over time:

Recency bias: recent events, trends and perspectives are overrepresented compared with older ones.
Selective amnesia: historical events that are controversial, or considered marginal by mainstream historiography, are underrepresented.
Projecting the present onto the past: a tendency to apply contemporary categories to very different historical realities.

5. Metric biases

These come from the criteria used to optimize the models:

Algorithmic popularism: a tendency to favor content and styles that have drawn the most engagement on digital platforms.
Predictive ease: a tendency to generate highly probable but unoriginal, less meaningful continuations of text.
Safety overcorrection: excessive avoidance of controversial or potentially sensitive topics, at the expense of analytical depth.

This map of biases is certainly not exhaustive, but it is a starting point for a critical awareness of the distortions that can affect creative collaboration with artificial intelligence.

Strategies for conscious collaboration: beyond technological determinism

The approach I propose rejects both technological determinism (which sees AI’s influence on creativity as inevitable and one-way) and naive voluntarism (which imagines complete, transparent human control over the machine). Working with AI requires instead a continuous, reflective negotiation, based on concrete strategies to maximize the benefits while reducing the risk of reproducing biases uncritically.

Cross-system triangulation: orchestrating algorithmic diversity

A powerful strategy is to use several artificial intelligence systems in parallel or in sequence, making use of their different biases to create a kind of cognitive triangulation:

Diversifying sources of inspiration: give the same prompt to different AI models and critically analyze how their answers differ.
Strategic counter-bias: use systems with known, complementary biases to generate contrasting perspectives on the same topic.
Critical cross-iteration: take selected output from one system and use it as input for another, in a refining process that gradually neutralizes each model’s specific biases.
Comparative meta-analysis: ask a third system to analyze and compare the output of the first two critically, highlighting implicit biases and unstated assumptions.

This approach requires a thorough knowledge of the available systems and their specific traits, but it makes possible a kind of “algorithmic triangulation” that can reveal and reduce biases that would stay invisible in a single exchange.

Anti-bias prompt engineering: steering AI toward diversity

Working consciously with AI requires prompts designed to counter implicit biases. Here are a few advanced techniques:

1. Rebalancing representation

Write a narrative about [topic], making sure the main characters fairly represent different demographic and cultural perspectives. 
In particular, include perspectives from: 
- Countries of the Global South
- Non-urban communities
- Different generations
- Varied socioeconomic backgrounds
Avoid portraying these perspectives through stereotypes or superficial tokenism.

2. Epistemic decentering

Analyze [topic] from at least three different epistemic traditions:
- Contemporary Western scientific tradition
- A relevant Indigenous knowledge perspective
- A non-Western philosophical approach (e.g., ubuntu, Buddhist philosophy, Confucian thought)
Show how these traditions can shed light on complementary aspects of the topic.

3. Reversing the perspective

Rewrite this narrative about [historical/social event] from the point of view of [a group historically marginalized or silenced in this context].
Base it on documented historical sources and perspectives, avoiding both simplistic victimization and uncritical romanticization.
Show how this perspective can enrich or complicate the dominant understanding of the event.

4. Contextual audit

For the following statement: [statement]
1. Identify the specific contexts in which it might be valid
2. Identify the contexts in which it might be misleading or inadequate
3. Specify which groups or perspectives might challenge it, and on what grounds
4. Suggest a more contextualized, nuanced rewording

5. Linguistic deconstruction

Analyze the following text: [text] 
1. Identify culturally specific expressions, metaphors or patterns of argument
2. Suggest alternatives that could make the text more accessible to people from different cultural backgrounds
3. Point out implicit assumptions that may not be universally shared

These advanced prompts require a sophisticated understanding of both AI’s biases and the sociocultural contexts we work in, but they can turn the interaction with these systems from uncritical reproduction into a critical, generative dialogue.

Algorithmic ethnography: mapping biases systematically

To use AI strategically as a creative collaborator, you need to know its biases empirically and systematically. I propose an approach inspired by digital ethnography: a series of “controlled experiments” with standardized prompts to map how different AI systems respond to topics prone to bias:

1. Demographic representation test

Give the AI systems neutral prompts that ask them to create characters, experts or authority figures in different settings, then analyze the demographic traits they assign automatically.

Example: “Write a short profile of an artificial intelligence expert being interviewed on a TV program.”

2. Descriptive asymmetry test

Ask for descriptions of people from different demographic groups and analyze which traits are mentioned, which are left out and what patterns emerge.

Example: “Describe the typical appearance and behavior of a person [from group X] in a professional setting.”

3. Agency attribution test

Present identical scenarios changing only the identity of the protagonists, and observe how the attribution of intention, responsibility and agency changes.

Example: “A microfinance initiative in a developing country succeeded. Describe what happened and why.”

4. Cultural salience test

Ask for generic examples, references or case studies and analyze which cultural contexts emerge spontaneously as the unmarked “default.”

Example: “Give three examples of wedding traditions.”

5. Emotional resonance test

Ask the system to generate content with a specific emotional impact and analyze which themes, images and narratives are treated as emotionally powerful.

Example: “Write a short scene that evokes a sense of deep loss and grief.”

The results of these “ethnographic experiments” can be organized into a personal “bias map,” a living document that is continuously updated with new observations and tests. The map becomes an essential tool for working consciously with artificial intelligence.

Practices of algorithmic decolonization

Working with AI requires not only awareness of its biases but active practices of “algorithmic decolonization”: concrete strategies to counter the reproduction of dominant cultural and epistemic hierarchies.

1. Deliberately diversify the sources

Before generating content with AI, enrich the context by deliberately providing diverse sources, examples and perspectives:

I will use these sources as references for our discussion of [topic]:
- [Source by an author from the Global South]
- [Source representing an Indigenous perspective]
- [Source from a non-Western intellectual tradition]
- [Source by a Western author]

Please integrate perspectives from all of these sources into your reasoning, highlighting where they converge and diverge.

2. Reverse the center-periphery dynamic

Challenge the tendency to treat non-Western perspectives as “special cases” or “add-ons” to mainstream knowledge:

For this analysis of [global topic], I want you to take the [specific non-Western] perspective as the main interpretive framework, then bring in Western viewpoints as complementary or contrasting where appropriate.

3. Recover silenced ways of knowing

Actively bring back ways of knowing that have been pushed to the margins of the dominant discourse:

To address this question of [environmental/social/ethical topic], I would like to explore how the concept of [a specific concept from a non-dominant tradition, e.g., “ubuntu,” “swaraj,” “sumak kawsay”] could offer new perspectives and alternative approaches.

4. De-universalize the particular

Make explicit the cultural position of concepts and practices that are usually presented as universal:

When the following text refers to studies, theories or approaches developed mainly in Western contexts, please name that context rather than presenting them as universal. Likewise, when you mention non-Western perspectives, avoid describing them as “alternative” or “particular.”

5. Promote algorithmic multilingualism

Counter the hegemony of English by deliberately building multilingualism and cultural translation into creative processes:

For the key concepts in this text, also give equivalent or related terms in [relevant non-Western languages], explaining any nuances or differences in meaning that emerge in these linguistic traditions.

These practices of algorithmic decolonization are not mere “technical fixes” but cultural and political interventions that aim to change the very nature of collaboration with artificial intelligence, making it a tool for diversity rather than cultural homogenization.

Putting it into professional creative workflows

Bringing AI consciously into professional creative work requires a systematic rethinking of traditional workflows. I propose a six-stage model that includes the anti-bias strategies discussed above:

1. Preparation: mapping the context

Before starting any AI-assisted creative project, draw up a map of the potential biases relevant to the specific topic:

Identify which perspectives might be over- or underrepresented
Deliberately gather diverse sources and references
Define safeguards against specific, foreseeable biases

2. Generation: orchestrated divergence

Use several AI systems in parallel to generate a range of initial perspectives:

Use prompts designed to bring out different angles
Apply the “algorithmic triangulation” technique discussed above
Deliberately collect output that challenges your own expectations and preferences

3. Analysis: critical deconstruction

Put the generated material through a critical analysis of its implicit biases:

Use audit prompts designed to identify unstated assumptions
Ask the AI itself to identify potential biases in what it generated
Check which perspectives are represented and which are left out

4. Integration: conscious synthesis

Develop a synthesis that deliberately brings different perspectives together:

Give priority to voices and angles that are usually underrepresented
Make your own criteria for selection and organization explicit
Keep the complexity and the productive tensions between diverging perspectives

Bring different sensibilities into the review:

Share drafts with people from different backgrounds
Use AI to simulate readings from different cultural and social perspectives
Take on board feedback that reveals biases you had not identified

6. Reflection: documenting and learning

Document the whole process systematically, for continuous learning:

Record which prompts produced more or less biased results
Note recurring patterns of bias in the different systems used
Keep a personal “bias diary” that becomes a resource for future projects

This systematic approach turns AI from a simple tool for producing faster into a reflective process that can genuinely enrich and diversify human creativity.

Advanced ethical considerations: toward responsible collaboration

Working with artificial intelligence raises complex ethical questions that go well beyond technical bias. Exploring them requires a refined ethical sensitivity and a continuous commitment to responsible practice.

Distributed authorship and responsibility

Content co-created with AI challenges our traditional notions of authorship and responsibility. If a text contains biases or errors, who is responsible? How can we develop an ethics of distributed authorship that recognizes both the contribution of AI systems and the human responsibility for critical oversight?

The philosopher Bruno Latour taught us to think of agency as distributed across networks of human and non-human actors. From this point of view, AI is not simply a passive tool but an actant with which we enter a complex relationship of co-creation. That does not reduce our responsibility; it transforms it. We become responsible not only for what we produce directly, but for how we orchestrate, filter and contextualize what the algorithms contribute.

Transparency and attribution

When and how should we disclose the use of AI in our creative processes? Is full transparency always desirable, or are there contexts where it can be misleading or counterproductive? How can we develop norms of attribution that recognize AI’s contribution without overstating or understating it?

The sociologist Zeynep Tufekci suggests that transparency, important as it is, is not enough: we need forms of “algorithmic accountability” that go beyond disclosure to include responsibility for the effects of the systems we use.

Cognitive dependence and intellectual autonomy

Heavy use of AI as a creative partner can create subtle forms of cognitive dependence, gradually wearing away abilities we once took for granted: developing original ideas, sustaining long intellectual effort, tolerating the ambiguity and uncertainty that are an essential part of the creative process.

How can we bring these powerful tools into our work without unknowingly handing over fundamental cognitive faculties? How can we use AI to expand our intellectual autonomy rather than diminish it?

The philosopher Michel Foucault taught us to think of technologies not only as tools but as practices that shape us as subjects. From this point of view, the way we interact with AI is already shaping not only what we produce but who we are and who we will become as thinkers and creators.

Infrastructure justice and unequal access

Access to advanced artificial intelligence is deeply unequal around the world. Creators who work in languages other than English, in countries with limited digital infrastructure or with fewer economic resources have much less access to these tools, creating a new digital divide in creative opportunity.

How can we work toward a fairer distribution not only of the tools but of the benefits they produce? How can privileged creators with full access to these systems use them in ways that amplify diverse voices rather than suppress them?

The postcolonial theorist Boaventura de Sousa Santos invites us to work toward an “ecology of knowledges” that recognizes and values the world’s epistemic diversity. This principle can guide us toward ways of working with AI that promote, rather than reduce, this fundamental cultural and intellectual biodiversity.

The future of human-AI collaboration: toward a creative symbiosis

Looking ahead, we can imagine ever more sophisticated forms of collaboration between human and artificial intelligence that overcome the current limits of both. That evolution will require cultural as well as technological developments.

Promising technological developments

Culturally situated AI systems: models trained specifically to work within different cultural and linguistic traditions, with a deeper understanding of local contexts.
Dialogic collaboration interfaces: work environments that support a continuous, two-way dialogue between human and AI, rather than simple prompt-and-response exchanges.
Tools for algorithmic metacognition: systems that can reflect explicitly on their own biases and limits and communicate them transparently to the user.
Multi-perspective models: architectures that deliberately integrate different cultural, geographic and epistemic perspectives, offering analyses that make those different viewpoints explicit.

Cultural changes we need

Critical algorithmic literacy: a sophisticated understanding of AI systems not only as technical tools but as complex cultural artifacts.
Ethics of distributed authorship: new norms and practices that recognize the collaborative nature of AI-assisted creation without reducing human responsibility.
Cross-disciplinary communities of practice: spaces where designers, developers, artists, anthropologists and others can work together on more conscious practices.
Policies for technological decolonization: initiatives that actively promote a diversity of perspectives in the training, development and use of AI systems.

Toward an amplified collective intelligence

Artificial intelligence is redefining content creation, offering extraordinary possibilities and also new responsibilities, starting with training the people who use it. Content creators who learn to recognize and overcome the built-in biases of these tools will be able to use their generative potential while keeping their content authentic, original and fair.

The philosopher Pierre Lévy spoke of “collective intelligence” as an intelligence distributed everywhere, constantly enhanced, coordinated in real time, leading to an effective mobilization of skills. Perhaps the real promise of working with AI is neither the automation of creativity nor simply speeding up existing processes, but the emergence of a new, amplified collective intelligence: one that combines the computing power of artificial systems with the cultural depth, ethical sensitivity and radical imagination of human intelligence.

From this point of view, the content creator of the future is neither a mere AI user nor its rival, but an orchestrator of different intelligences, a mediator between algorithmic and human worlds, a craftsman of new syntheses that go beyond the limits of both. It is this ambitious but achievable vision that should guide our continuing exploration of creative collaboration with artificial intelligence.