AI Ethics and Artificial Intelligence: What the Current Frameworks Miss

AI ethics and artificial intelligence governance have generated an extensive literature of guidelines, principles, and recommendations over the past decade. According to a 2023 UNESCO analysis, more than 160 national and international AI ethics frameworks have been published since 2016. Almost none created enforceable obligations. This post examines what that means and what genuine AI ethics requires.

What separates AI ethics principles from AI ethics law?

AI ethics principles are voluntary commitments. They produce reports, frameworks, and roadmaps. They do not produce legal liability for AI systems that cause documented harm. The gap between AI ethics principles and AI ethics law is not a staging problem — as if law will eventually catch up to the principles. It is a power problem: the entities best positioned to enforce AI ethics obligations are also the entities most invested in avoiding them.

The EU AI Act is the most significant legislative attempt to bridge this gap. It creates binding obligations for high-risk AI systems, with fines up to €35 million or 7% of global annual turnover for serious violations. But its conformity assessment mechanism — the process by which AI systems are evaluated before deployment — still permits notified bodies with commercial relationships with the companies they assess. This is what I document as a normative simulation: a legal structure that creates the appearance of independent oversight while structurally permitting its opposite.

Why does the UNESCO AI Ethics Recommendation fall short of what the moment requires?

The 2021 UNESCO Recommendation on the Ethics of Artificial Intelligence is the most comprehensive multilateral AI ethics instrument in existence. It covers 11 core values and 10 policy areas. Its signatories include 193 member states. And it is entirely non-binding. Countries are invited to implement its recommendations through domestic legislation at their own pace and discretion.

This is not a criticism of UNESCO's institutional constraints — any instrument requiring consensus from 193 states will reflect their lowest common denominator. It is an observation that the voluntary multilateral model is insufficient for technologies that operate across borders in real time. An AI system trained in one jurisdiction and deployed in another faces no enforceable global accountability standard. The Neurological Birth Certificate model I propose addresses this directly: a binding global certification that crosses jurisdictions the way the technology itself does.

What does the Caribbean and Latin American perspective add to AI ethics?

In my research, I have identified a structural blind spot in the global AI ethics debate: it is conducted almost entirely by scholars and policymakers from OECD countries, about AI risks that primarily affect OECD populations, through governance mechanisms designed by and for OECD institutional contexts. The Dominican Republic, which launched the first national AI strategy in the Caribbean and Central America (ENIA, Decreto 498-23, 2023), is building AI governance capacity in an environment where civil liability frameworks are weaker, independent audit institutions are younger, and the populations most exposed to AI bias have fewer legal resources.

This is not an argument for lower AI ethics standards in the Caribbean and Latin America — it is an argument for higher ones. When institutional capacity to remediate AI harm after the fact is limited, the case for mandatory pre-deployment auditing is stronger, not weaker. The Humanoide Framework I propose was designed with this context in mind.

Photo by Igor Omilaev on Unsplash

What would genuine AI ethics enforcement require?

Genuine AI ethics enforcement requires four elements that no existing framework fully implements. First, mandatory pre-deployment review of every high-impact AI system by an auditor with no financial relationship with the system's developer or operator. Second, legal veto power in the hands of that independent auditor — not advisory authority, not recommendation authority, but the ability to prohibit deployment. Third, global jurisdictional reach that matches the global reach of the systems being regulated. Fourth, ongoing post-deployment monitoring with the same independence requirements as pre-deployment review.

The Neurological Birth Certificate model I propose in Humanoide en la Torre de Babel is my attempt to specify what meeting all four requirements would look like institutionally. It draws on analogies from pharmaceutical regulation (where independent review before market authorization is standard), aviation safety (where accident investigation is conducted by bodies independent of the airline industry), and clinical medicine (where treatment protocols are reviewed by ethics boards independent of the treating institution).