The Singapore Law Gazette

Epieikeia in the Age of Algorithms: Reflections on Advocacy, AI, and the Soul of the Profession

I. The Question Beneath the Question

I spent a day at TechLaw.Fest 2026 wandering the exhibition halls. The booths were impressive. Autonomous platforms drafted pleadings while I watched. Predictive engines forecast judicial outcomes with confidence intervals and neat little graphs. Digital co-pilots managed discovery with the quiet efficiency of a well-run chambers. The demonstrations worked. The sales pitches were polished. Everyone was smiling.

And yet.

Beneath the fluorescent glow of the vendor booths, I found myself circling a question I could not quite articulate. It was not whether these tools worked. They plainly did. It was not whether they would reshape practice. They already have. It was something more fundamental, and it took me a while to name it. In our relentless pursuit of computational efficiency, are we quietly surrendering the architecture of our profession to an automated feedback loop, or are we merely refining our tools?

This is not a transient anxiety. It is the contemporary expression of an ancient philosophical divide about the nature of justice itself.

In Book V of the Nicomachean Ethics, Aristotle articulated a concept that has guided civilised legal traditions for millennia: epieikeia.¹Epieikeia ἐπιείκεια (pronounced eh pi AY kee ah) is the Aristotelian virtue by which a judge corrects the law where its necessary universality produces an unjust result. It is not a sentiment, and it is not leniency. It belongs to practical wisdom (phronesis). Aristotle’s formulation is instructive: the equitable judge asks what the lawgiver would have said had he been present and had known the particular facts. See Nicomachean Ethics V.10 (1137b–1138a). The word is often translated as equity, but the translation requires care. Epieikeia is not mercy in the sentimental sense. It is not the softening of a harsh rule out of pity, nor the exercise of discretion untethered from principle. Aristotle distinguishes it from eleos, the emotion of pity, and from suggnome, the granting of excuse. ²Aristotle treats eleos (ἔλεος) (pronounced as EH-leh-oss) and suggnome (συγγνώμη (pronounced as soong-GNOH-mee) as distinct from epieikeia. Eleos is the emotion of pity, the visceral response to another’s suffering. Suggnome is the granting of excuse or forgiveness, the recognition that a person’s conduct should not be held fully against them; in modern legal terms it is closer to the recognition of a defence than to the exercise of equity. Epieikeia is neither a feeling nor an excuse. See Nicomachean Ethics V.10 (1137b–1138a) and VI.11 (1143a–b).Epieikeia is a virtue of practical wisdom, and it belongs to the judge and the advocate alike.

The law, by its structural necessity, speaks in universal terms. It frames broad statutory mandates and generalised rules designed to govern the collective. But human life is stubbornly particular. It is messy. It is radically unpredictable. The universal rule, applied blindly to the hyper-specific realities of a human grievance, can produce severe and unintended cruelty.

Epieikeia is the corrective. It is the sophisticated human intervention that rectifies the law where it falls short due to its absolute universality. It is the flexible patch applied by judge and advocate alike to ensure that the spirit of justice is not crushed beneath the weight of rigid formal text. The equitable judge does not act outside the law. He completes it.

As litigators, we romanticise the independent advocate as the permanent guardian of this Aristotelian human element. We construct a professional mythology around the lone advocate standing in the well of the court, the singular conduit through which human empathy and moral reasoning enter the institutional machine. But let us be radically honest. How much of our celebrated professional soul, our independent judgement, our capacity for ethical contemplation, is bound up with and limited by the physical tools we use to think, compose, and communicate?

The evolution of legal practice suggests that the medium has always quietly dictated the message.

II. The Medium and the Message: A Brief History of Friction

Consider the generational shift within a single career. The titans of the mid-twentieth-century bar sat before heavy, mechanical typewriters. Every word required deliberate physical force. Every error demanded a painstaking halt, the messy intervention of white-out, or a complete retyping. Carbon paper imposed a literal, absolute constraint on the volume of documentation that could be produced.

This friction was far more than a technical inconvenience. It functioned as an involuntary ethical filter. You did not fire off a toxic, rambling, 20-page letter of demand over a minor dispute in the typewriter era. Not because you were more virtuous, but because the sheer physical toll of composition forced you to pause, reflect, self-edit, and measure the moral and legal weight of every word. The physical drag of the machine mandated a moment of human sobriety. There are practitioners today who choose heavier mechanical keyboards precisely because the resistance forces deliberation. The typewriter is gone. The instinct remains.

Then arrived the personal computer. The ancient barriers to text production collapsed entirely. Drafting became frictionless. Cut-and-paste allowed for the instantaneous duplication of massive legal blocks. The act of writing was severed from the physical limits of the page. We witnessed the first great structural mutation of modern advocacy: text exploded exponentially. Pleadings swelled into dense recitations. Document bundles grew into towering indices. Cross-examination outlines became infinite scripts. The absence of physical friction led directly to a hyper-inflation of legal language, forcing courts to introduce increasingly rigid page limits and font rules to keep the sheer mass of human-generated text from suffocating the administration of justice.

Then came the pandemic, accelerating our transition into the sterile geometry of the remote hearing room. Advocacy was reduced to a compressed gallery view of talking heads. The vital human current of real-time court presence, the micro-expressions of a witness, the atmospheric shifts in the courtroom, the authoritative gravity of the bench, was effectively muted.

Each of these transitions changed not merely how we practised, but what we practised. The medium shaped the message. And now, with the dawn of Generative AI, we face a paradigm shift that is different in kind, not merely in degree.

III. The Nuance of Computational Volume: From Anxiety to Liberation

To understand the ethical battlefield before us, we must first interrogate the traditional calculus of leverage in adversarial practice.

Historically, aggressive, high-friction litigation tactics were checked by the hard economy of human time and financial capital. To construct a devastating narrative, launch an intricate web of interrogatories, or bombard an opposing party with a relentless stream of analytical correspondence required a massive investment of billable hours. The strategy carried an immediate financial risk for the client and imposed a heavy cognitive tax on the solicitor. This natural economic friction acted as an inherent safety valve. It ensured that parties, regardless of financial asymmetry, operated within the boundaries of human endurance and economic rationality. Litigation was an exercise in bespoke human craftsmanship, where every tactical stroke carried a cost.

Autonomous Agentic AI dismantles this equilibrium.

It gives rise to a quiet, unspoken anxiety within the profession: the dread of automated attrition. Resource asymmetry is not new. A well-funded team of 10 once required 10 salaries and 10 tired minds. What is new is the collapse of marginal cost. There is a growing fear among sole practitioners, legal aid lawyers, and mid-sized firms that the future of civil disputes will belong not to the sharpest analytical mind or the most righteous cause, but to the litigant capable of deploying the most massive computational infrastructure. A well-resourced party can now deploy specialised legal models to autogenerate vast rivers of dense, hyper-technical text, complex procedural cross-examinations, and endless streams of adversarial correspondence at zero marginal cost. The machine does not experience burnout. It does not tire. Its billing rate for text generation is, at current subsidised rates, effectively free.

The unspoken worry is that the human advocate on the receiving end will be buried under an asymmetric, industrialised avalanche of machine-generated text. They will be forced to spend finite cognitive energy parsing the noise, reading every machine-gunned paragraph, and drafting ethical responses to an infinite digital interlocutor, while the client’s financial resources are ground down by design. AI is transforming professional correspondence from an exercise in human persuasion into an industrial factory of cognitive exhaustion.

Yet to look upon this horizon with fear is to misunderstand the profound promise of the tools we are inventing. The Luddites, rightly understood, were not opposed to machines as such. They were artisans protesting the displacement of craft by unskilled labour. Their concern was not technology, but what technology cost them. It is a concern we would do well to take seriously rather than dismiss. This computational asymmetry is not an insurmountable threat. It is a historic catalyst for our own professional liberation.

When we intelligently and critically embrace premium, enterprise-grade legal AI, the technology ceases to be a weapon wielded against us and transforms into our ultimate shield. Advanced systems can process, synthesise, cross-reference, and filter thousands of pages of convoluted adversarial text in seconds. They can scan an industrialised deluge of documentation, identify hidden gaps, strip away rhetorical fluff, and distil a mountain of artificial data down to its core legal vulnerabilities.

By allowing technology to absorb the computational noise, the human advocate is liberated from the administrative quagmire that has choked modern practice for decades. Technology does not displace the lawyer. It strips away the mundane, the rote, and the mechanical, returning the advocate to what has always truly mattered: high-value strategic thinking, nuanced moral clarity, deep historical context, and the irreplaceable art of human persuasion.

We do not need to fear the automated wave, because the machine can only ever manipulate syntax. It can never replicate the ultimate source of legal leverage: the human conscience. The tool clears the field so that human judgement can prevail.

IV. The Paradox of Access: From the Disenfranchised to the “Smart Alec”

When the profession discusses “access to justice”, it almost exclusively looks downward at the most vulnerable: the uneducated, the traditionally illiterate, the marginalised. For this segment, multimodal AI (tools that speak, see, listen, and interpret) is rightly celebrated as a democratising force. A person who cannot read a dense legal summons can now photograph a court order with a smartphone, and a voice-activated model will translate the legalese into plain spoken dialect. They can speak back to the interface, voice their grievance in their native tongue, and receive immediate guidance on navigating the courthouse.

Yet this genuine increase in raw accessibility masks an insidious trap: the illusion of access.

If we are not vigilant, we risk sleepwalking into a deeply stratified, structurally unfair two-tier civil justice system. On the upper tier, elite corporate litigants will utilise heavily audited, premium legal AI ecosystems, tools like Harvey AI, whose pricing models and computational requirements naturally target larger institutional firms. These tools are embedded within robust frameworks of experienced human counsel who supervise, audit, and contextualise every algorithmic output before it reaches a client or a court.

On the lower tier, the uneducated, the impoverished, and the self-represented will be left to defend their liberties using free, unsupervised, public-facing commercial chatbots. When an unaudited chatbot confidently hallucinates an incorrect filing deadline, invents a fictitious precedent, or misinterprets a strict limitation period, how can a traditionally illiterate individual verify the error? They cannot. They will navigate valleys of digital data they cannot read, blindly trust the machine’s absolute confidence, and present the algorithm’s hallucinations to a judge in good faith, only to watch their case collapse. The problem is not unique to the disenfranchised. It is the broader epistemic challenge of an age in which fluent falsehood is cheap. But it falls hardest on those least equipped to resist it.

If technology merely automates the exclusion of the marginalised under the guise of inclusion, we have not achieved access to justice. We have automated systemic injustice.

But the illusion of competence does not only harm the disenfranchised. It mutates into a different and highly disruptive friction when it travels up the commercial ladder to the corporate client, the tech-savvy entrepreneur, or the SME director.

Litigators are increasingly encountering the “Smart Alec” client: individuals armed with consumer AI subscriptions and a superficial familiarity with prompts, who independently generate their own legal memos, construct parallel and unaudited strategies, and aggressively seek to cross-examine their own lawyer’s work. They arrive at consultations with raw, uncontextualised algorithmic data streams, demanding that counsel pursue obscure procedural avenues or launch wasteful evidentiary excursions based entirely on what a commercial chatbot told them.

The structural irony is profound. These clients frequently suffer from an acute form of cognitive impatience. They cannot or will not take the time to read the comprehensive, nuanced, carefully tailored advices their human counsel has painstakingly constructed. Instead, they run their lawyer’s drafts through a commercial AI summariser, read a flattened, context-stripped, and frequently distorted characterisation of the legal advice, and then use that flawed summary to harass, micromanage, and cross-examine the operating solicitor.

This is the legal equivalent of a patient who skims an online medical forum, misinterprets the clinical literature because they lack years of diagnostic training, and then aggressively argues with the surgeon about the placement of the incision. The client confuses the machine’s ability to aggregate and structure data with the advocate’s seasoned ability to exercise judgement. They are unqualified to formulate proper legal context, recognise procedural nuances, or design the precise query strings required to safely interrogate a complex litigation file.

It is tempting to say the algorithm merely mirrors the client’s own misconceptions. But the mirror is not neutral. It reflects with the authority of a machine, and that authority is precisely what makes the client’s confidence so difficult to dislodge. And so they weaponise the algorithm’s generic outputs to create internal friction, generate unnecessary billable hours, question professional integrity, and stall their own defence.

This problem is especially acute for those who advise high-net-worth individuals. The client who arrives with an algorithmic second opinion is often the one with the resources to obtain it, and the confidence to trust it. The machine’s fluency is easily mistaken for expertise when the client has no reason to doubt their own judgement.

V. The Ethical Core: The Ghost in the Legal Machine

This client-driven friction brings us to the moral centre of our contemporary professional crisis: the preservation of human agency.

Advocacy, at its core, is not a mechanical exercise in information processing, nor a commercial transaction in risk allocation. It is a deeply relational and profoundly moral act. When a client walks through the doors of a law firm, they are rarely in a state of calm rationality. They are frequently broken, confused, angry, and desperate. They do not simply hand the lawyer a pristine binder of chronological facts. They hand the lawyer their anxiety, their damaged honour, their financial security, and their vulnerability.

The true calling of the advocate is to act as an ethical and moral crucible. To take that raw, chaotic mass of human suffering, filter it through the lens of institutional ethics, professional conduct rules, and legal precedent, and translate it into a structured, dignified plea for justice before the Court.

When we outsource this process to a statistical algorithm, we are not merely engaging in efficiency. We are participating in a dangerous form of emotional, cognitive, and ethical offloading. A Large Language Model can mimic the vocabulary of righteous anger with startling accuracy. It can simulate the rhetorical structure of mercy and compassion. But the machine has no conscience. It cannot feel the devastating weight of a shattered family in a probate dispute. It cannot comprehend the social shame of an unfair criminal accusation. It cannot care about the outcome of the trial. And it cannot stand behind its words with personal honour.

When a practitioner blindly signs off on an AI-generated submission without subjecting every sentence to exhaustive intellectual audit, they are committing a profound ethical desertion. They are offering the Court a soulless artefact generated entirely by a mathematical optimisation loop, a text devoid of human conviction and moral responsibility. If the advocate on record does not genuinely feel and understand the truth of the argument they are signing, why should the judge be expected to believe it?

Advocacy requires a human voice because it requires an anchor of human accountability. The text must be backed by a soul that understands the gravity of the stakes.

The next major wave of professional liability will centre squarely on the evolving boundary of technological competence and ethical delegation. Litigants and disciplinary bodies will routinely interrogate whether a practitioner’s reliance on automated tools has crossed the bright line from legitimate assistance into complete professional abdication. We will see text challenged in open court not because it lacks logical structure, but because it becomes apparent that no independent human mind ever took genuine moral responsibility for the substance on the page.

The defence of “the machine generated it” will be recognised, in the not-too-distant future, as an admission of professional negligence.

VI. Anchoring the Smart Nation in Human Discretion

As Singapore steers its legal landscape toward full integration with National AI initiatives and Smart Nation blueprints, our regulatory objective must never be to suppress or punish technological adoption. The active promotion of cutting-edge legal technology is vital to securing our position as the preeminent global legal hub in Asia.

The court’s role is not to penalise the innovative use of AI, but to meticulously safeguard the standards of its deployment, ensuring that computational efficiency never weaponises into systemic and procedural abuse. We must ensure that our rush to automate does not outpace our commitment to humanise the law.

To preserve the soul of our profession within a thriving pro-AI ecosystem, our evolution must be guided by clear frameworks of responsible innovation and enlightened stewardship.

First, preventing procedural overreach. The courts, utilising existing inherent case management powers under the Rules of Court, must remain vigilant against the tactical deployment of automated text-generation tools to introduce disproportionate, unreviewed structural volume designed solely to overwhelm and exhaust an opponent’s resources. The use of AI to generate procedural noise must be met with immediate and decisive cost sanctions.

Second, the non-delegable duty of competence. In strict alignment with Singapore’s vision of trusted and transparent AI, a practitioner’s ethical responsibility for the accuracy, factual grounding, and moral integrity of their written submissions must remain entirely non-delegable. AI should amplify the advocate’s intellect and sharpen their focus. It must never replace their moral conscience or independent judgement.

Third, expanding digital shelters. As our judicial systems digitise, the State’s pro bono frameworks, community legal clinics, and legal aid bureaus must actively pioneer the deployment of secure, voice-first, and highly structured legal technologies. These tools must be designed to act as protective digital shields, guiding the illiterate and uneducated away from the dangerous pitfalls of unsupervised public chatbots and ensuring their voices are accurately heard.

The integrity of our profession is not a static monument carved out of marble by the generation before us, to be passively admired. It is a live, dynamic, and vulnerable ecosystem that we must actively build, fiercely defend, and consciously humanise every single day of our practice.

Walking away from TechLaw.Fest 2026, it became clear that we have successfully engineered a legal framework that knows exactly how to process massive amounts of complex data at lightning speed.

Now, as advocates, our highest ethical duty is to ensure that Singapore builds a world-class, technologically supreme infrastructure that still fundamentally knows how to look a human being in the eye, feel the weight of their unique grievance, and extend true and Aristotelian mercy.

The machine can process. It cannot care. That distinction, and our unwavering commitment to it, will define the future of our profession.