Dr Fazal Ali
Truth is not enough in the Intelligent Age because AI models don’t “know” what truth is. AI Assemblages function by predicting patterns and harmonising text rather than consulting reality. A factually correct statement can lack the context, nuance, or ethical judgement required to be genuinely useful. Vigilance is the only safeguard to avert conflating truth with verisimilitude.
We may never be the masters of truth. Truth is tentative. It conquers us. It is subtle. It inspires us. As we embrace truth, we find it open and inviting. Truth is a part of trustworthiness. But achieving trustworthiness does not guarantee truthfulness. An AI can be highly factual and accurate in its outputs (it tells the “truth”), but if it is programmed to discriminate, uses leaked personal data, or cannot explain its reasoning, it is not trustworthy. Conversely, an AI can reliably be wrong; that is, it predictably generates false information, meaning it is technically reliable but neither true nor trustworthy.
While truth or accuracy is about whether the output is factually correct, trustworthiness is a much broader, multidimensional standard that evaluates whether the system’s behaviour is safe, fair, transparent and accountable. The AI evangelists see the creative class as priests before the oracle. We have found ourselves on the edge of the industrialisation of good-enough truth.
“Truth” is not listed as a single, standalone word in any taxonomy of ethical principles of Artificial Intelligence (AI) because AI does not have a concept of objective reality. Instead of “truth,” frameworks prioritise proxies like accuracy, reliability, and verifiability. The reason “truth” cannot be an AI ethical principle is rooted in how AI-Assemblages work.
Generative AI models do not “know” facts. They use statistical probabilities to predict the next plausible word or pixel based on their training data. They prioritise correlation over comprehension. The present models mimic patterns rather than checking facts against an objective baseline. With confidence, they can generate plausible yet entirely false information or hallucinations.
What humans consider “truth” often requires context, reasoning, and judgment. Demanding that an AI adhere to “truth” would require it to resolve philosophical or domain-specific debates universally, which is an impossible task for an algorithm. AI divorces problem-solving from human intelligence because machines bypass the cognitive friction of intuition, contextual judgment, and emotional understanding that builds true expertise.
This automation allows computers to generate solutions without ever possessing or experiencing the conscious intelligence that underpins human problem-framing. It is unclear whether it is Dr Frankenstein or his monster against whose maleficence we should be guarding as we gaze across the ravine of differing epistemic standards. Algorithmic fairness often relies on statistical correspondence. Because AI prioritises the statistical majority rather than truth, it often flattens nuanced, localised human experiences, erasing minority viewpoints in favour of dominant ones.
Instead of “truth,” standard frameworks established by organisations like the National Institute of Standards and Technology (NIST) and the European Commission focus on trustworthiness. Most ethical AI frameworks invoke the core pillars of Accuracy, Fairness, Reliability, Transparency, Explainability, and Accountability. Leading frameworks such as the OECD and UNESCO focus on ensuring that AI systems do not deceive users, clearly state their limits, and use verifiable data.
An LLM can distinguish between “true,” “false”, and a third state, “neither true nor false,” underscoring that its relationship to truth is never binary and rarely certain. This mechanism favours formal coherence rather than truth. When a response is correct, it’s the effect of a fortunate alignment of probabilities, not the result of a will to verify.
In the absence of precise facts, models default to predicting a statistically likely pattern rather than accepting that the answer eludes them. And when it’s false, it’s not a lie but the ordinary output based on the training data. Large language models don’t consult reality; they predict the word most likely to follow, based on billions of instances.
There are no social contours or inner cartography of the world in their architectures. If an AI lacks sufficient information on a niche topic, it will guess to fill the gaps, weaving patterns that sound logical but are factually void. Models can inherit biases from their training data, leading them to invent patterns or exaggerate falsehoods that align with those biases. AI models do not “understand” truth. Without grounding mechanisms, they rely solely on statistical probabilities.
Generative AI models are assemblages of fragments, sometimes correct, sometimes false, sometimes indeterminate. They don’t plan to mislead any more than they intend to tell the truth. They extend forms, without a necessary connection to the reality they inscribe. Because AI can produce plausible yet incorrect or hallucinated information (especially in highly subjective or specialised fields), blind reliance on these tools amplifies the risks of intellectual atrophy and moral attrition.
Dr Fazal Ali completed his Master's in Philosophy at the University of the West Indies. He was a Commonwealth Scholar who attended the University of Cambridge, Hughes Hall; the Provost of the University of Trinidad and Tobago; the acting President of UTT; and the Chairman of the Teaching Service Commission. He is the President of NIHERST and an external services consultant with the IDB.
