Philippine President Ferdinand Marcos Jr. has called for a "careful study" of proposals to lower the minimum age of criminal liability for minors, as reported by Inquirer net. While this is fundamentally a legal and social policy debate, it presents a critical case study for engineers and technologists working at the intersection of governance, data science, and fair systems. The question of how-and whether-to deploy algorithmic decision support in youth justice is not just a courtroom issue; it's an engineering challenge that demands the same rigor Marcos is asking for.
This article reframes the juvenile Justice reform debate through an engineering lens. We'll explore how risk assessment algorithms, data pipelines. And AI governance frameworks intersect with the principle of doli incapax (the presumption that minors can't form criminal intent). Marcos' call for "careful study" is exactly the approach we need when building software that affects children's liberty. If we can't explain the model, we shouldn't use it on a child,
The Age of Criminal Liability: A Complex Socio-Legal Problem
Currently, the Philippines sets the minimum age of criminal liability at 15 years old (Republic Act 9344, as amended by RA 10630). Proposals to lower it to 12 or even 9 have been repeatedly debated. Opponents argue that incarceration harms adolescent brain development and that poverty, not malice, drives many offenses. Supporters cite a rise in heinous crimes involving minors as justification for stricter accountability.
From a systems engineering perspective, this is a classic multi-constraint optimization problem. You have competing variables: public safety, child welfare, limited state resources. And the capacity of rehabilitation programs. Any change in the legal "parameter" (age threshold) will ripple through juvenile detention centres, courts. And social services. A "careful study" must model these dynamics before deployment-just as you would in a production environment.
Why Software Engineers Should Care: Algorithmic Risk Assessments in Court
In many jurisdictions-including the U. S and increasingly in Southeast Asia-judges use algorithmic risk assessments to predict a minor's likelihood of reoffending or failing to appear in court. Tools like the Youth Assessment and Screening Instrument (YASI) or the Ohio Youth Assessment System score minors based on static and dynamic factors (age, criminal history, family stability, school attendance). These scores influence detention decisions, custody levels, and program assignments,
The engineering problemThese models inherit biases from historical arrest data. Which often reflect systemic discrimination. For example, a study by ProPublica (2016) on the COMPAS recidivism algorithm found that Black defendants were twice as likely as white defendants to be misclassified as high risk for violent recidivism. If the Philippines lowers the age of liability, the data fed into such models will include younger, more vulnerable cohorts, amplifying the potential for harm. Marcos' call for careful study should be a mandatory step before any automated system touches a child's case.
The Promise and Peril of AI in Juvenile Justice
AI could help match minors to evidence-based rehabilitation programs, flag potential abuse in detention. Or identify patterns that human judges miss. For instance, a machine learning model trained on longitudinal data might reveal that first-time property offenders under 14 almost never reoffend if offered family counselling-a finding that could shape policy. That's the promise.
The peril comes from opaque "black box" models. If a decision denies a 12-year-old liberty based on an AI score that neither the judge nor the minor's lawyer can interpret, due process is violated. The European Union's proposed AI Act classifies such tools as "high-risk" and mandates transparency - human oversight. And bias audits. The Philippines lacks such legislation. Marcos' "careful study" directive should include explicit calls for algorithmic auditing standards for any tech used in juvenile courts.
Case Study: COMPAS and ProPublica's Bias Analysis
The COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) tool developed by Northpointe remains the most infamous example of algorithmic risk assessment gone wrong. ProPublica's 2016 analysis "Machine Bias" showed that the tool's false positive rate for recidivism was 45% for Black defendants versus 23% for white defendants. Yet Northpointe defended the tool by claiming calibration parity-a statistical property that doesn't ensure fairness.
This controversy highlights a key engineering lesson: metrics like accuracy, AUC. And calibration aren't enough. You must also measure "equal opportunity" (equal false positive/negative rates across groups) and "counterfactual fairness" (the decision would be the same if the protected attribute were different). Any "careful study" of changes to minors' criminal liability that involves software must mandate these fairness audits publicly.
Data Privacy and the Rights of Minors
Collecting data on minors for risk assessments introduces additional privacy obligations. Under the Philippine Data Privacy Act (RA 10173), personal information of minors is considered sensitive. When building a system that tracks juvenile records, engineers must implement data minimization, strict access controls. And automatic deletion upon the minor reaching a certain age or completing rehabilitation.
Contrast this with the current state: many juvenile records remain in fragmented, legacy databases with weak security. If the age of liability is lowered, the volume of sensitive data will surge, increasing attack surface. A "careful study" must include a security architecture review-what RFC 3552 calls the "threat model. " Without encryption at rest and in transit, and without periodic penetration testing, the system is unsafe for production.
What 'Careful Study' Looks Like in Practice: Simulation, Audit, Transparency
Marcos used the phrase "careful study"-but what does that mean for a technologist? It means running simulations on historical data before deploying any policy change. Build a digital twin of the juvenile justice system. Vary the age threshold. Observe predicted effects on detention occupancy, case processing time, and recidivism rates. The study must be auditable: every assumption, every feature weight, every training dataset version must be documented and reproducible.
It also means transparency. If a government contracts a vendor to supply a risk assessment tool, the contract must mandate open publication of performance metrics across demographic subgroups. The vendor shouldn't hide behind trade secrets. The Philippine Department of Justice should publish an algorithmic impact assessment, similar to Canada's Directive on Automated Decision-Making. Which requires peer review of any system that makes high-impact decisions.
Lessons from Software Engineering: Iterative Development vs. Policy Iteration
Software engineers know that shipping a product on day one is almost never correct. We use iterative development, beta testing, canary releases, and A/B experiments. The same philosophy should apply to policy changes like lowering the age of criminal liability. Start with a pilot region, and monitor outcomes for six monthsCollect data on recidivism, mental health, and family impact before scaling nationwide.
But there's a key difference: in software, we can roll back a bad release. In justice, a wrongful detention can't be undone. That's why the bar for "careful study" must be higher. Marcos' administration should fund a longitudinal study that uses matched cohort analysis-comparing minors in the new liability regime with a control group under the old system-before any legislative change is codified.
The Role of Open Source and Community Oversight
One way to build trust in risk assessment systems is to make the source code and training data sets public (with careful redaction of personal information). Open-source projects like the FairML library and IBM's AI Fairness 360 provide toolkits that any technical auditor can use to evaluate bias. If the Philippines mandates that any algorithmic tool used in juvenile justice be open-source, independent researchers-both local and international-can verify its fairness.
This aligns with the spirit of Marcos' "careful study": it crowd-sources the scrutiny. No single agency or consultant can catch every bias or data leakage. The open-source community acts as a continuous audit. However, this requires political will to resist vendor lock-in and proprietary claims. The Global Data Justice project has shown that governments that embrace open-source models for criminal justice experience fewer scandals and higher public trust.
FAQ: Addressing Common Questions on Minors' Criminal Liability and Technology
Q: Does lowering the age of criminal liability automatically mean more minors will be put in jail?
A: Not necessarily-it expands the group eligible for prosecution. However, most proposals include diversion programs. The real impact depends on how the justice system's software (case management, risk assessment) handles the increased volume and younger population.
Q: Can AI replace judges in juvenile cases?
A: No. At best, AI can provide a decision support score. The final decision must remain with a human judge who can consider nuanced factors the model may miss, such as a minor's trauma history or artistic talent. The European Commission's Ethics Guidelines for Trustworthy AI explicitly state that automated systems must not override human discretion in sentencing.
Q: What is the biggest technical risk of using algorithms in juvenile justice,
A: Bias amplificationHistorical arrest data often over-represents marginalized communities. A model trained on that data will perpetuate the same inequalities, locking minors into cycles of surveillance and punishment. The "careful study" must include a disaggregated analysis by income, ethnicity. And geography.
Q: How can engineers ensure data privacy for minors?
A: Use pseudonymization, strict access control based on role. And add automatic data retention limits. Follow the principle of "data minimization" - only collect what is directly needed for the decision. Encrypt all data in transit (TLS 1,? And 3) and at rest (AES-256)
Q: What lessons can the Philippines learn from other countries?
A: Canada's Algorithmic Impact Assessment, the UK's Office for Artificial Intelligence guidelines, and New Zealand's Oranga Tamariki (ministry for Children) experience with predictive analytics all stress the need for mandatory pre-deployment audit, public consultation. And a sunset clause that requires re-approval every two years.
What do you think?
Should the Philippine government mandate open-source auditing of any risk assessment tool used in juvenile justice, even if it delays procurement?
Is lowering the age of criminal liability a socio-economic issue that no amount of algorithmic fairness can truly fix?
If you were tasked with designing a "careful study" framework for this policy change, what three key metrics would you track before and after implementation?
Conclusion: President Marcos' call for a "careful study" before changing minors' criminal liability is exactly the right approach-not just for legislators but for engineers and data scientists. Building fair, transparent, and effective systems for youth justice requires the same iterative rigor, bias auditing. And security-first mindset that we apply to any high-stakes software. The law may set the age. But technology will define how that age is enforced, and as the Inquirernet report suggests, this debate is far from over. Let's ensure the study is thorough enough to protect both public safety and the futures of our youngest citizens.
This article was inspired by news coverage from Marcos urges 'careful study' of changes in minors' criminal liability - Inquirer. And net and related reports from Philstar, and com and The Manila Times,
Need a Custom App Built?
Let's discuss your project and bring your ideas to life.
Contact Me Today →