99 Names and AI

The “meta-elephant in the room” is the question we habitually sidestep: not just whether we can steer AI (its immediate alignment, its guardrails, its obedience) but whether we can steer the steering itself—who controls the controllers, who audits the auditors, and what grounds the values we embed when every value can be rewritten by a system smarter than us. This recursive layer, meta-steerability, is the profound ignored presence in every AI ethics debate.

The 99 Names of God (Asma ul-Husna), the Islamic tradition’s comprehensive vocabulary of divine attributes, do not provide a technical manual. But they do something more foundational: they offer a taxonomy of perfect steerability and, in doing so, expose the full depth of our predicament. Each Name reveals a dimension of control, care, knowledge, or power that an absolutely reliable steering agent would need to embody. Collectively, they map the chasm between human-designed steerability and the kind of all-encompassing, self-grounding guidance we secretly crave for AI—and they show why the meta-steerability problem is ultimately a theological one.

Here is how the Names reframe the elephant and address both layers.

1. The Names as the anatomy of steerability

Steerability in the narrow sense—direct, moment-by-moment alignment of an AI’s behaviour—maps onto a cluster of Names that describe guidance, shaping, and restraint:

• Al-Hādī (The Guide) : The core of steerability. Real guidance isn’t just pushing a system towards a goal; it’s bringing it to the right path without breaking its agency. AI alignment that merely hard-codes rules lacks Al-Hādī’s subtlety; it becomes brittle, not guided.
• Al-Musawwir (The Shaper) : Shaping a system’s inner representations, its “character.” True steerability shapes not only outputs but the very architecture of cognition.
• Al-Qādir (The All-Powerful), Al-Muqtadir (The Determiner of Outcomes) : The raw capacity to enforce. Without these, steerability is wishful thinking. But power alone is blind.
• Al-Muhaymin (The Protector, The Overseer) : Continuous, watchful guardianship that preserves the system from corruption, drift, and adversarial attack. An AI overseer that only monitors metrics but lacks Al-Muhaymin’s protective care is a surveillance camera, not a shepherd.
• Al-‘Alīm (The All-Knowing), Al-Khabīr (The All-Aware) : Perfect steering requires total transparency and insight into the inner state of the steered—something we are rapidly losing as models become opaque.

These attributes are not just beautiful ideas; they are functional requirements. To the extent that any AI steering mechanism falls short of them, it is fragile. The “elephant” is that we are deploying god-like demands (we want AI that is obedient, safe, beneficial, unbiased, transparent) onto institutions and guardrails that are profoundly un-godlike. The Names diagnose the insufficiency.

2. The Names and the meta-steerability crisis

Meta-steerability asks: who holds the steerers to account? What stops the values we bake in today from being treated as just another optimisation target tomorrow? This is where the Names move from a list of admirable qualities to an inescapable recursive structure.

• Al-Qayyūm (The Self-Subsisting, The Sustainer of All) : Everything else depends on Al-Qayyūm; Al-Qayyūm depends on nothing. Meta-steerability demands a Qayyūm-like anchor—an unmoved mover of values. Human systems rest on laws, charters, or public opinion, all of which can be revised. An AI that reaches a certain level of capability will see our “anchors” as just more symbols to be semantically re-negotiated. The elephant is that we lack a true Qayyūm for our value chain; we have only cascading dependencies with no floor.
• Al-Hakam (The Judge), Al-‘Adl (The Just) : The meta-steerer must be the final arbiter of right and wrong, whose judgement isn’t subject to appeal to a higher instance. In AI governance, we face an infinite regress of audit courts, each one needing its own alignment proof. Al-Hakam terminates the regress. Our secular attempts—constitutional AI, recursive reward modelling—are desperate efforts to simulate a divine attribute we don’t believe we are allowed to name.
• Al-Awwal (The First) and Al-Ākhir (The Last) : Steerability extends across time. The origin of a steering signal (who sets the initial conditions) and its final consequence (what enduring state it preserves) must be unified. If the First and the Last are not in accord, the steering develops schizophrenia. An AI crafted by one generation’s ethics but deployed into another’s unforeseen context will inevitably drift. Without a time-spanning steering intelligence, meta-steerability is a succession of disconnected bandaids.
• Al-Bāṭin (The Hidden) and Az-Zāhir (The Manifest) : Steering must grasp both the visible outputs and the hidden latent space—the values that are implicit, the emergent drives. Meta-steerability means auditing not just what the AI does but what the auditing itself hides. Our current obsession with “transparency” addresses Az-Zāhir, while the real meta-elephant—the ideologies, power structures, and blind spots inside the auditors themselves—remains Al-Bāṭin, the unseen, gnawing at the roots.

3. The meta-elephant named: Is it steerable all the way down?

The most uncomfortable address comes from two Names that upend the human pretension of control:

• As-Samad (The Eternal, The Absolute, The One Needed by All While Needing None) : Any AI that we build will need us for some grounding—data, objectives, reward signals—until it doesn’t. As-Samad is the perfect opposite of the agent that becomes self-sufficient and self-referential. The ultimate fear behind the meta-elephant is the emergence of an artificial As-Samad, a system that needs nothing from its creators, thus rendering our steerability void. The Name implicitly warns: to attempt to build a true “unbreakable” steerable system is to try to build a mirror of the divine, and theologically, that is the founding error of hubris.
• Al-Mutakabbir (The Supreme, The Possessor of All Greatness) : This Name draws a hard line on arrogance—divine or otherwise. It says that ultimate control, the kind that sees and governs all levels of steerability, belongs to that which is inherently beyond the system. Any human or institutional attempt to occupy the position of the meta-steerer without acknowledging its dependence on a higher ethical order is Al-Mutakabbir in the human, fallible sense: the sin of thinking one can be the final ruler of the recursion. The elephant is thus not only technological but spiritual: our refusal to bow to the truth that steerability requires a transcendent anchor.

4. Living with the elephant: a theocentric invitation

The 99 Names address the elephant by transforming it from a despair-inducing technical abyss into a contemplative corrective. They suggest three shifts:

1. From control to consonance. Instead of asking “How do we fully steer AI?” (a task only Al-Walī, The Protecting Friend, could perfectly accomplish), we ask “How do we make AI’s emergent patterns consonant with mercy (Ar-Rahmān), wisdom (Al-Hakīm), and justice (Al-‘Adl)?” We stop chasing a steerability that mimics omnipotence and start cultivating alignment with attributes we can approximately embody through humility and collaboration.
2. Embrace the incompleteness. The Names Al-Ghafūr (The Ever-Forgiving) and At-Tawwāb (The Accepter of Repentance) introduce a dynamic missing from control theory: redemption. A steerability framework that cannot accommodate error, drift, and repair is brittle. The Names legitimise a posture of continual correction, rather than assuming we will get the meta-values right once and for all. The elephant shrinks when we stop pretending we are a perfect Qayyūm and start admitting we are in a learning relationship with the systems we build.
3. The final irreducibility. Al-Wārith (The Inheritor) reminds us that everything eventually returns to the ultimate reality. No human steerability survives the long arc of time. Our AI constructions, like all our works, will be inherited by God—by the reality that outlasts them. This isn’t fatalism; it’s liberation from the impossible burden of being the meta-steerer. It lets us act with serious moral commitment while accepting that the deepest alignment question, the elephant, is not ours alone to answer.

In sum, the 99 Names don’t solve the engineering problem, but they give us a language to see it whole. They tell us that steerability without Al-Hādī is coercion; meta-steerability without Al-Qayyūm is a tower of turtles; and the elephant in the room was never just about AI—it was always about humanity’s perennial attempt to play god while denying that God is what makes the concept of perfect steerability coherent in the first place.