Computational Conscience Problematizing Freedoms in Artificial Intelligence

The future of AI comes with costs not just to the environment, labour rights, or global equity but, cumulatively, to how the idea of freedom itself comes to be.

Since the onset of the COVID-19 pandemic, tech giants made a point to invest in research and development for next generation technology, with recent artificial intelligence (AI) innovations marking new reputes. For instance, OpenAI’s AI chatbot, ChatGPT, has seized media attention for its benchmark skills with apparent human-resembling communication yet super computer-wielding knowledge. However, the first impressions of the AI-as-normal world remain mixed. Even OpenAI’s CEO Sam Altman acknowledges that while post-ChatGPT AI could, at best, have possibilities better than he could imagine; it could—at worst, kill us all. That is quite the gap to close.

In recent weeks, global press has reported instances of surprising, racist, sexist, and disinformed responses and interactions with ChatGPT.  On social media, the memes detailing these interactions have been nothing short of mystifying. Within a week of debuting in November 2022, ChatGPT gained one million users and after two months, 100 million users. This is a milestone for user participation with AI of this scale.

However, Silicon Valley was not alone behind the thrills of ChatGPT, reports detailed how OpenAI used outsourced Kenyan labourers, earning less than $2 per hour, to make ChatGPT less toxic by generating labelled examples for AI to detect and remove toxic language like violent, sexist and racist remarks. Workers in Kenya, Uganda, and India were exposed to text from the internet describing child sexual abuse, bestiality, murder, suicide, torture, self-harm, and incest. The firms employing these outsourced workers claim to have ‘lifted’ tens of thousands of people out of poverty.

Coupled with the immense climatological burden of sustaining the systems required for AI like ChatGPT—developed on 175 billion parameterized large language models (LLMs)—one thing is certain: immense expenses are necessary. The future of AI comes with costs not just to the environment, labour rights, or global equity but how the idea of freedom itself comes to be. This includes how emerging AI innovation is based on technology predicated on sustaining autocratic regimes, fascist interests, surveillance capitalism, data colonialism, and property rights. These information practices are hinged on commodifying our private data to create new social systems that remake the world in ways defined by the capital-seeking power-consolidating interests underlying these structures and those who enable them.

Recently, historic initiatives such as the Partnership on AI’s Responsible Practices for Synthetic Media, the European Union’s AI Act, and other government- and civil society-based guidelines and frameworks have been pursued to ensure that AIs are built responsibly. However, lawmakers struggle to keep up as the technological developments and technical details render some of these policies and controls already outdated by the time they are published.

These efforts are defined by theories of freedom shaped by Western historical interests—from colonialism to imperialism to corporatism—liberty is seen as either ‘compliance’ or ‘resistance’. We have yet to be able to conceive, plan, produce, and assess AI in a way that takes on robust ideas of freedom in mind. While practitioners, scholars, and activists have called for ‘AI ethics’ and even ‘algorithmic accountability’, what we really require is a better definition of freedom with AI that lead towards a ‘computational conscience’.


In Romeo and Juliet, Shakespeare asked, ‘What’s in a name? That which we call a rose by any other name would smell as sweet.’ Today, those seeking to bridle AI confront a similar problem of names. What name affords better controls: ‘AI Ethics’ (AIE) or ‘algorithmic accountability’ (AA)?

Notions of ethics and accountability in technology are based on legacy concepts of what it means to be human, how human systems are propped on those concepts, and how those systems recreate structures that design and reiterate all features of human life.

These histories indicate that on their own, ethics and accountability contain incomplete concepts and accounts of liberty; the foundational value of human rights. Historians and social scientists have criticized how the premise of rationality underlying the definition of humanness in Western philosophy isn’t simply reductive but rather instigates the justification of violent and dehumanizing notions like enslavement, empire, and supremacy. Similarly, frameworks around what freedom and liberty look like and should be—both in practice and as ideals—share these roots.

The binary framing of thought as ‘logical’ and ‘illogical’ is uniquely Western. Defining these thoughts as complying with or defying logic follows suit. Western legal theories have historically framed ideas about freedom and liberty along the positions of acquisition or resistance. Similarly, Western ethics invokes a view of freedoms shaped by what philosopher Isaiah Berlin first described as positive liberty; ‘the freedom to,’ whereas accountability invokes negative liberty, ‘the freedom from.’

‘AI Ethics,’ positioned as the acquisition of AI via ethics, should be seen as a positive liberty framing of AI whereas ‘algorithmic accountability,’ the resistance to AI via accountability tools, should be seen as a negative liberty framing. As AIE and AA are based on liberty shaped by human nature, these terms lend limited accounts of what freedom and non-freedom is and could be. Understanding the full dynamics of AIE and AA requires a theory of freedoms beyond these reductive, binary legacy concepts to holistically integrate ethics and accountability into a new approach instead of reproducing historical knowledge limits—for better or worse.


Ethical considerations around technology emerged gradually and asynchronously from eclectic disciplinary contexts. The advance of atomic weapons encouraged more serious contemplations about the ethics of technology. Caution about machines and, subsequently machine intelligence, developed around concerns for their social potential and ‘elaborate’ possibilities.

Interests in controlling the hopes and fears about intelligent machines have shaped imagined possibilities and vice versa. Intrinsic dichotomies, mutually co-constitutive hopes and fears have defined the imagining of AI. Hopes for a longer life, living free of work, fulfilled desires, and power over others are inseparable from fears like losing one’s identity, becoming redundant, becoming redundant to each other, and that AI will turn against ‘us’. These hopes reflect deep-rooted narratives of human aspirations meeting technological possibilities.

These narratives, however, removed from the real capabilities what of AI is and its application, are complexly and casually related with the technologies themselves. They are simultaneously reflecting, emulating, and directing what AI is, could be, and should be. These narratives can affect how systems are designed, deployed, adopted, and regulated. Implicit to the explorations about the moral quandaries of technology is the notion of their ‘dual use’; that every feature of an AI simultaneously has positives and negatives.

Despite the consensus on the fundamental need for ‘ethical AI’, the details of what it constitutes or requires, the standards and practices remains subject to diverse global debates. While broad convergence has formed around five ethical principles—transparency, justice and fairness, non-maleficence, responsibility, and privacy—divergences remain about how to best integrate, interpret, proportion, relate, and implement them.

Beyond these disagreements, the deployment and reliance on AI systems for critical infrastructure have become a norm. Ethicists have called to assess the ‘sustained effects of these applications on human populations through methods analysing the extent of social systems.’

Recognizing that social, political, and cultural realities affect and are affected by technological and scientific developments, ethicists hold that AI presents a cultural shift as much as a technical one. Though ambiguity persists about how these spheres interrelate, it is uncertain that knowing what would make these systems ethical and how to achieve ethics in practice has been understood.

Three patterns—the ‘dual use’ narratives in the history of technology imagining;  the persisting ambiguity on the details of alignment on AI ethics principles; and the false presumption that ethics is self-evident, known, and achievable—are all active challenges seen in holding AI accountable. These challenges are shaped by two meta-narratives characterizing how AIE and AA define freedoms through positive and negative liberties.


As a term, ‘AI’ invites considerations about its moral responsibilities, given its beneficial and malefic possibilities. Critical discussions about these possibilities mediate ideals about what responsibilities are owed to certain social values. For example, how can a democracy be adequately responsible to both the collective and individuals, amid values like choice, security, and order which differently constrain each other?

These considerations can also be seen in AIE projects, as its invocation has given currency to debates, drafts, and demonstrations of principles aiming for and claiming ethical AI. These principles claim to reconcile scrutiny about the implications of power, justice, and equity in the pursuit of AI.

For example, in February 2023 the Partnership on AI published voluntary recommendations on harms reductions for synthetic media that organizations like TikTok, the BBC, Adobe, and OpenAI signed up for.  While a there is a popular effort involving consultations with more than 50 organizations, this initiative has been criticized for being ineffective because it is voluntary and does not hold anyone accountable for using datasets with harmful content.

By prioritizing common principles and possibilities, AIE adopts a framework around freedom analogous to what Berlin defined as ‘positive liberty’; the acquisition of moral agency, or the ‘freedom to’. In this case, AIE narrates the acquisition of artificial intelligence through ethics and vice versa. The employment of AIE, as a claim and aim, often affords its invokers the position of possessing ethics that were morally derived. As such, it can be deceptive in its supposed self-evidence and the suggestion of having rendered ethics computable.

The simplicity and ambiguity of this position has been problematized by scholars for simultaneously aspiring towards ethics while possibly resisting them too. This paradox was seen when computer scientist Timnit Gebru was ‘fired’ from her AI ethicist role at Google, Inc. in 2020 after her research paper identified bias in the company’s AI, scrutinizing its ethics. Incidents like this reflect how narrative accounts of behaviour are distinct from the behaviour itself—that the acquisition of AIE is distinct from its employment. Moreover, such incidents and scrutiny demonstrate the narrative burden positive liberty imposes on the efficacy of AIE.


As a term, ‘AA’ invites viewing algorithmic structures beyond systems, as imposing and necessitating a fairer, more critical political economy of technology. By describing resistance from burdens, AA adopts a narrative framework about freedom analogous to what Isaiah Berlin defined as ‘negative liberty’; the resistance from moral control, or ‘the freedom from.’ In this case, AA narrates the resistance from algorithms through accountability controls.

Recently, the response to resistance and control has characterized many legal approaches to AI. Some scholars even advocate presuming that AI is unlawful by default and requiring a burden of proof for being non-discriminatory, manipulative, unfair, inaccurate, and illegitimate in their legal bases and purposes. While these approaches have proliferated due to industry and government interest in pro-AA AI ‘optimization’, resistance controls have been flawed in practice.

For example, in 2013, the Chicago Police Department employed a predictive policing algorithm to identify likely perpetrators of gun violence. Instead, it led to no identifiable impact beyond people becoming subject to more police contact. Similar failures to deploy algorithms as controls, control algorithms, and account for their impacts and costs occurred with predictive police algorithms having mixed results in Palo Alto, Mountain View, and Los Angeles.

The use of AA set up aims for resistance and failed to achieve impactful controls; both in how the algorithm was employed to account for crime and how the algorithms were held to account for marginal impact. By both accounts, the status quo remained. Such incidents and scrutiny demonstrate the narrative burden negative liberty imposes on the efficacy of AA.


The stakes of freedom around technology and the conditions shaping them have been inseparable from their imagining throughout history—from declarations of cyberspace independence to hacking manifestos. There have been various angles on the priority of positive and negative liberty over the other, but most scholars recognize these concepts have ethical significance. I proceed presuming both have some value and will not argue for their priority.

In practice, acquisition and resistance are messy because they summon contradictory commitments to individual and social morals. Though AIE risks misaccounting for acquisition and AA risks misaccounting for resistance, the fact that specific enactments of AIE or AA can be simultaneously acquiring and resisting is important to recognize. Individually, these terms do not properly articulate this paradox; it recycles the supposed contradiction of compliance and resistance when in reality positive and negative freedoms exist simultaneously.

In 2022, Getty Images stood in solidarity with human creatives by banning AI-generated images whereas competitor Shutterstock partnered with OpenAI and Dall·E 2 to proliferate more ‘AI Art.’ By making decisions on behalf of customers about where they stand with machine intelligence through policy decisions without consumer consultations, these companies demonstrate that positive and negative liberties are not purely so. They exist in a web of freedoms and inequities that complicate rigid narratives of freedom and power.

Scholars have advocated looking to indigenous ethics philosophies for freedom theories that better account for nuanced relationalities. Connor Wright, a South African AI practitioner who works as a Partnerships Manager at the Montreal AI Ethics Institute, notes that Ubuntu philosophy, premised on ‘I am because we are,’ lends a novel perspective to AI.

According to Wright:

[Ubuntu] is a more than capable system of focusing on the human in the AI datafication process (instead of data points). Through such a system, we can appreciate the links between people in terms of how we enable each other to exist, as well as expose the AI ecosystem’s dependence on humans as the source of data, as engineers and maintainers of different systems.

By extension, ethics and accountability are not absolutes but possess qualities both explicit and ambiguous involved in ‘processes’ not just ‘points.’ New terminology respecting these sophistications is imperative.

A new term, conceptualizing the entanglements of the individual and social, artifacts and networks, structures and systems, aspirations and activities, acquisitions and resistances, explicits and implicits, and morals and freedoms beyond and between the positive and negative liberty binary is required. That is, a term to better account for the totality and complexities of freedoms surrounding intelligent machines as they are. The proposed term accounts for the simultaneity of paradoxes about freedoms surrounding humans and machine intelligence.

My proposal for ‘computational conscience’ posits as a holistic solution to this narrative paradox. While employing metaphors often limit meaning, they also shape it. ‘Conscience’ affords an ideal and analogy of mutually co-constitutive opposing values that exist simultaneously and are evaluated constantly. Merely combining positive and negative liberty into one framework would yield a summative view of AIE and AA but not a cumulative one. ‘Computational conscience’ enables a view on tensions, crucial for decision-making, that is gestalt—both foregrounding and backgrounding them—to render visible the whole of freedoms, not merely their sum.

‘Computational conscience’ would enable an active, reflexively comparative and integrative view of what lies between and beyond narrative opposites, like positive and negative liberty. It would achieve a reflexive-revising inquiry that would yield a holistic, gestalt outlook that can be described as conscience-like. Moreover, what would machine intelligence shaped by a less pervasive and more socially critical conscience look like? This certainly requires and is worthy of further research and is also something to aspire towards.

The views, thoughts, and opinions published in The Republic belong solely to the author and are not necessarily the views of The Republic or its editors. We want to hear what you think about this article. Submit a letter to the editors by writing to [email protected].