Myths and Machines, part 8

Pontus Wärnestål
4 min readMay 20, 2024

Welcome back to part 8 of “Myths and Machines,” where we unravel the connections between ancient myth and cutting-edge technology. This time, we dive into the story of the Oracle in Delphi. What can we learn from this story when it comes to our digital oracles in our own time?

(Previous texts: part 1, part 2, part 3, part 4, part 5, part 6, part 7.)

The Oracle of Delphi: Navigating the Ambiguities of AI Predictions

In ancient Greece, the Oracle of Delphi was revered as a conduit to the divine, a priestess who channeled the cryptic wisdom of Apollo. Seated on a tripod over a fissure in the earth, the Pythia, as she was known, delivered prophecies that were often ambiguous and open to myriad interpretations. Leaders and common folk alike sought her guidance, but the true meaning of her words frequently eluded them, leading to misinterpretations that could alter the course of events dramatically.

A painting of the Pythia, the Delphic Oracle, seated on a tripod. She wears a red veil and robe, holding a branch and a dish. Smoke surrounds her as she delivers a prophecy.
By John Collier — Art Gallery of South Australia Website Webpage PictureOld source [1], Public Domain,

The story of the Oracle of Delphi provides a profound metaphor for our modern interactions with generative language models (LLMs), such as those used in artificial intelligence. These sophisticated systems, like the Oracle, produce outputs that can be both enlightening and enigmatic. The challenge lies in distinguishing clear insights from the potential gibberish — navigating the murky waters of AI-generated information without falling prey to false or misleading interpretations.

The Essence and Warning

The Oracle of Delphi symbolizes the complexity and ambiguity inherent in processing and interpreting vast amounts of information. Her cryptic pronouncements reflect the difficulties faced when deriving clear, actionable insights from complex data. This is particularly relevant today, as we grapple with the outputs of advanced generative AI systems, which can sometimes generate hallucinations — false or nonsensical information that appears plausible on the surface but is ultimately misleading. (Note: research suggest that ethylene and ethane can be identified as the pneuma, the prophetic vapor that was responsible for the trancelike state of the priestesses in Delphi during sessions. The oracle actually hallucinated due to the fumes, and this serves as a connection to what large language models do when they produce erroneous information.)

The Risk of Hallucinations in Generative AI

Generative models, trained on diverse and extensive datasets, have an impressive ability to generate human-like text (and images). They can produce everything from poetry to technical explanations, and they have become valuable tools in various fields. However, their outputs can occasionally include hallucinations — fabricated information that is not grounded in the data they were trained on.

Consider a scenario where an AI system is asked to generate a detailed explanation about a historical event. The AI might produce a narrative that includes accurate details alongside invented or erroneous facts. To an untrained eye, this mixture of truth and fiction can be indistinguishable, leading to misinterpretation and the spread of misinformation.

The Challenge of Assessing Truth

The difficulty in assessing what is true and what is merely gibberish from a generative LLM mirrors the challenges faced by those who sought the Oracle’s counsel. Here are some lessons and parallels:

1. Critical Evaluation: Just as those who received the Oracle’s prophecies had to critically evaluate her words, users of AI-generated content must apply rigorous scrutiny. This involves cross-referencing AI outputs with reliable sources and being wary of information that cannot be verified.

2. Understanding Limitations: Recognizing the limitations of AI is crucial. While generative models can simulate understanding and produce coherent text, they do not possess true comprehension. They generate responses based on patterns in data, which can sometimes lead to plausible-sounding but incorrect information.

3. Avoiding Confirmation Bias: The Oracle’s seekers often heard what they wished to hear, rather than the true meaning of her words. Similarly, users of AI must guard against confirmation bias — the tendency to favor information that confirms pre-existing beliefs. It’s essential to approach AI-generated content with an open mind and a critical perspective.

The Importance of Human Oversight

The role of human oversight in interacting with AI cannot be overstated. AI systems can augment our abilities to process and analyze information, but they are not infallible. The story of the Oracle reminds us that human judgment and interpretation are indispensable in navigating complex information landscapes.

Take-Away Message

The tale of the Oracle of Delphi underscores the challenges of interpreting ambiguous information and the risks of misinterpretation. In the context of generative AI, this myth serves as a cautionary tale about the potential for AI systems to produce hallucinations — false or misleading information that can easily be mistaken for truth. As we integrate AI into our decision-making processes, we must remain vigilant, critically evaluate AI outputs, and ensure robust human oversight. By doing so, we can harness the power of AI while mitigating the risks of misinformation and maintaining a clear grasp on reality.

Next Up: Saturn Devouring his Children…

One of the more horrific stories, and which has a real bearing on the future of the Internet. Read on in the final installment of this series on Myths and Machines –›



Pontus Wärnestål

Deputy Professor (PhD) at Halmstad University (Sweden). Father of two. I ride my bike to work.