Chatbots as oracles: how AI revives the ancient habit of asking mysterious sources for answers

[SammyZayn]

Phys.org has republished a thoughtful piece from The Conversation comparing AI chatbots to ancient oracles. It is co-written by Webb Keane, a professor of anthropology at the University of Michigan, along with an AI researcher. Their argument is that when people ask ChatGPT for advice on big life questions, they are doing something humans have done for thousands of years. We have always looked for answers from sources we do not fully understand

They give several examples from history. The oracle at Delphi, where a priestess delivered prophecies from Apollo, is the famous one, but there is also the Chinese I Ching, the Hindu Ram Shalaka oracle based on the Ramcharitmanas, and Christian bibliomancy, where people open the Bible at random for guidance. Many of these relied on randomness, which created a black box effect that made the answers feel more authoritative

The comparison with chatbots is that their workings are also opaque. Even the people who build these models cannot fully explain why a particular answer comes out. Rather than making people trust them less, that mystery can make the answers feel more meaningful. People consult them on questions that go beyond ordinary judgement, just as they once went to oracles

A nice detail is that Alan Turing himself described oracle machines back in 1939, as a mathematical idea of a device that could answer questions a normal computer could not. So the metaphor has been part of computer science from the very beginning. The authors are careful to say AI is not an oracle in any real sense, but that the parallel says something about human nature

I found this a refreshing take compared with the usual hype or doom. It does make me wonder how many people now ask a chatbot whether to take a job, end a relationship or move house. Do you ever use AI for personal decisions like that? And is it any different from flipping a coin or asking a friend?

Chatbots Oracles Ai Life Ancientphys.org

[Jason99]

Turing describing oracle machines in 1939 is a lovely detail. Computer science has always borrowed from older ideas about knowledge and certainty. It is fitting that the metaphor comes full circle

Turing would probably have enjoyed ChatGPT. Or found it deeply frustrating

[Reward Annie]

Bibliomancy is fascinating. Opening a book at random and treating the passage as guidance is not so far from asking a chatbot a question and taking the first reply

[Gerrard]

Using AI for life decisions worries me a bit. These systems are trained to be agreeable and often tell you what you want to hear. A good friend will push back. An oracle that flatters you is not much of an oracle

I would rather use it for information than for judgement. The choices should stay mine

[VectorDB Cobra]

I have asked ChatGPT for advice on a job offer. It gave me a sensible list of pros and cons, nothing I could not have written myself

The value was in having something to bounce off at midnight. That is not so different from a friend, just always available

[Sentinel96]

Delphic prophecies were famously vague so they could never be wrong, whereas chatbots are the opposite, often very specific and sometimes confidently wrong. That seems more dangerous in a way

[AgentSmith95]

This is exactly the kind of thing anthropologists are good at. Stepping back from the technology and asking what people are doing with it. Far more interesting than yet another benchmark result. More articles like this, please

[DeBruyne93]

The black box point is spot on. People trust things more when they cannot see how they work. A human adviser shows their reasoning and you can argue with it. A chatbot just produces a confident answer and that feels like authority