Blog post - What does interpreting with AI provide?

[hilary]

What does interpreting with AI provide?



My first encounters with AI interpretation came as a surprise. I hadn’t expected it to be able to work with metaphor, but it could – very well. I had expected it, given two hexagrams, to be able to work out which lines were changing, but it couldn’t. And that makes me wonder what it understands a hexagram to be – and then it occurs to me that this is anthropomorphic nonsense on my part, because what do I mean, ‘understand’? Am I speculating on what ChatGPT pictures in its mind’s eye when it imagines a reading??

So I asked Yi a simple, open question: ‘What does interpreting readings with AI provide?’
Yi answered with Hexagram 26, Great Tending, changing at lines 1, 4 and 6 to Hexagram 32, Lasting.

And then I had the idea of feeding this as a prompt to various AIs – the ones that don’t cost money, or my phone number, to create an account! – to see what they would come up with. I’ll share some of their output first, and then what the reading brought to mind for me.

The AIs​

Perplexity​

Perplexity.ai gets marks from me for linking to its sources, and choosing LiSe’s site as one of them. Unfortunately, its other source is a bit rubbish, and it doesn’t quote that much text. No hornboard at line 4 or heaven’s highway at line 6. It manages the entirety of 26.1, though, and says,

Line 1 changing suggests initial caution: “There is danger. Fruitful to stop”. This may indicate the need for careful consideration before fully embracing AI interpretation.

That’s a pretty representative sample of its interpretation – generic and feeble. It gets more points, though, for being the only one to suggest that Hexagram 32 might refer to the enduring quality of the Yi itself, not of AI!

CoPilot​

CoPilot – the one you can access via Bing – is worse: anodyne, uninspiring and a text-free zone. Here’s its complete reading:

Interpreting I Ching readings with AI offers a unique blend of ancient wisdom and modern technology.
Hexagram 26 (Great Taming) suggests the importance of restraint and nurturing strength over time. Changing lines add layers to the reading:

The final Hexagram 32 (Duration) emphasizes consistency and steadiness over the long term.

So, interpreting I Ching readings with AI might symbolize blending patience and strength with the steadiness of technology to gain profound, enduring insights. It highlights the importance of continuous learning and adaptability. Quite the fascinating combination! What are your thoughts on this?

How depressing. I couldn’t muster the motivation to nudge it with questions about specific images to see if it could do better. Imagine if this were someone’s first (non-)encounter with the Yi – what are the odds they’d ever want to bother with it again?

ChatGPT​

ChatGPT (the vanilla version, not Remo’s YiShi) produces a much longer, more in-depth response. As you can see, though, it doesn’t actually know the text – which is a bit odd when it does know the Chinese names of the hexagrams. But by default, someone asking ChatGPT for a reading will get no hexagram text and inaccurate versions of the lines.

(Update: someone asking ChatGPT, vanilla version, for a reading will get some random collection of hexagrams and lines, not generated with anything like yarrow or coin odds, and probably with the wrong relating hexagram. Pam reported this in response to my last post and Remo confirms it.)

At this point I tried asking it to use Legge’s translation exclusively (since that one’s public domain and freely available online). It still hallucinated the line texts. I told it as much. It apologised:

You’re right, and I apologize for the oversight. Let me correct that and provide an interpretation strictly based on James Legge’s translation of the I Ching.

Well, that sounded intelligent and promising. Here’s what it came up with next:

• Line 1 (First Changing Line): Legge’s translation reads: “At the beginning, the danger is great. Do not act yet.”
  This suggests a moment of caution when starting a process. When using AI, it indicates that one should not rush into conclusions or actions without first laying a solid foundation. It reminds us to carefully evaluate the situation before proceeding with interpretations or decisions.
• Line 4 (Fourth Changing Line): Legge translates this line as: “Perseverance leads to success. Go not to the east, go not to the west. The south brings good fortune. The north brings misfortune.”
  This line emphasizes that the correct guidance will ensure progress. Here, it suggests that interpreting with AI requires following a methodical, aligned approach. Moving in the right direction, under careful guidance, brings success. Straying or acting without direction will lead to difficulty.
• Line 6 (Sixth Changing Line): The sixth line says: “The strong power remains hidden. It is a time for restraint and caution. Success will come when the time is right.”
  In this line, Legge describes how maintaining restraint in times of power is key to eventual success. When applied to AI interpretation, this stresses that it’s essential not to rush the results but to allow for the development of understanding. Great wisdom comes when the right moment arrives.

As you can see, this lot bears even less resemblance to the Legge translation than its first attempt; it reads as if I’d asked, ‘Please generate some random twaddle in imitation of the I Ching’. (Oddly, given the copyright situation, it did much better when I asked it to use Wilhelm. I suppose copyright is not something AI worries about much. Would it reliably give the correct text every time if asked for Wilhelm, though? Who knows?)

It’s not bad at interpreting the text it hallucinated, but this is not quite the point. ChatGPT should take some advice from the Image of Hexagram 26…

YiShi​

Remo’s achievement in persuading ChatGPT to use the actual I Ching now seems quite remarkable. Since he is concerned about copyright, he’s using a machine-generated translation from Wilhelm’s German, which has its limitations.

Still, even a machine-generated translation is a great deal better than none. Here’s its version of 26.4:

Changing Line 26.4: The protective board of a young bull. All hail!
Explanation: AI functions as a safeguard, providing structure and support for those navigating the complexities of interpretation. Its guidance is steadfast, reliable, and serves to shield from potential errors or missteps.

This is our first glimpse of the bullock. As for the interpretation… does AI look more like the intelligently-applied hornboard, to you, or the dumb ruminant in need of taming? (Perhaps the interpreter is biased.)

Thanks to its use of a translation, YiShi is also the only AI to have heard of trigrams.

AI mirrors the “sky within the mountain” by integrating and embodying vast knowledge in a grounded and accessible form.

That’s not a bad response to the imagery, is it?

Gemini 2.0​

(When I opened up Google Gemini, it invited me to try its very latest version, so that’s what this is…)

Gemini, rather like YiShi, thinks that Hexagram 26 could refer to

“- AI’s capacity to store and process vast amounts of information: AI can “contain” and analyze far more data related to the I Ching (commentaries, interpretations, historical uses) than any human could.”

That would be a good application of the trigrams (which it didn’t mention), and seems like a plausible idea that will surely be true one day… but at the moment, Gemini has no text to offer for the hexagrams, and had mixed success with the lines:

• Line 1: “Danger. It furthers one to desist.” This suggests that at the beginning, caution is advised. AI’s role in I Ching interpretation should be approached with care, acknowledging its limitations and potential for misinterpretation.
• Line 4: “The taming power of the great. Good fortune.” This line indicates that with proper guidance and application, the “great power” (AI) can be harnessed for positive outcomes. It suggests that AI can be a valuable tool for understanding the I Ching, as long as it’s used responsibly.
• Line 6: “He has attained to the way of heaven. Success.” This is a very auspicious line, suggesting that the use of AI in I Ching interpretation can lead to a deeper understanding of universal principles and truths.

Two out of three… could be worse, though we’ve lost our bullock again. But look what happened when I fed it the actual text:

Me:
You haven’t used the correct translation of Hexagram 26, line 4. The full text (from the Wilhelm/Baynes translation) reads: ‘The headboard of a young bull. Great good fortune.’ How does that change your interpretation?

Gemini:

You are absolutely right! My apologies, I seem to have oversimplified that line. The Wilhelm/Baynes translation of Hexagram 26, line 4, is indeed: “The headboard of a young bull. Great good fortune.”

This is a much more specific and evocative image. The “headboard” refers to a piece of wood placed across the horns of a young bull to prevent it from butting and causing damage. It symbolizes a controlled, yet powerful force.

• Emphasis on Control and Guidance:The image of the headboard highlights the need to carefully manage and direct the power of AI. It’s not about unleashing it unchecked, but about providing a framework and boundaries for its use in interpreting the I Ching. This could mean:
  • Careful selection of data and algorithms: Ensuring that the AI is trained on reliable and accurate information about the I Ching.
  • Human oversight and validation: Recognizing that AI is a tool, and human intuition and understanding are still essential in the interpretive process.
  • Ethical considerations: Establishing guidelines for how AI should be used in this context, ensuring that it respects the traditions and nuances of the I Ching.
• Potential for Great Benefit: The “great good fortune” associated with this line reinforces the idea that with proper control and guidance, AI can be a very valuable asset in I Ching interpretation. It suggests that this approach can lead to significant breakthroughs in understanding and application.
• The “Young Bull” Analogy: This image also suggests that AI’s role in I Ching interpretation is still in its early stages. It has great potential, but it needs to be nurtured and guided to reach its full maturity.

Here’s how this more accurate translation changes the interpretation in the context of AI and I Ching interpretation:

So, with this corrected understanding of Line 4, the overall message becomes even clearer: AI can be a powerful tool for understanding the I Ching, but it must be used with careful control and guidance, much like placing a headboard on a young bull. When used responsibly, it can bring great benefit and lead to a deeper understanding of this ancient wisdom.

Thank you for pointing out my mistake! It’s crucial to use accurate translations when interpreting the I Ching, and I appreciate you helping me to refine my understanding.

I thought it worth quoting all of the above because I think it’s a genuinely good interpretation. A human could do better, but I’ve also seen humans do worse. You get this calibre of response fairly reliably from Gemini and ChatGPT, provided you feed them the real text or trigrams. You’ll get all kinds of gibberish if you don’t, and I really don’t like the thought that a newcomer could be misled into believing this to be the real thing.

I also want to give Gemini extra credit for a) that final comment about the importance of accurate translations and b) the note it tacked on the end of its initial interpretation:

Important Note: As a large language model, I can provide interpretations based on my understanding of the I Ching. However, it’s essential to remember that the I Ching is a complex and nuanced system, and ultimately, the most meaningful interpretation will come from your own intuition and understanding.

So, what does AI interpretation provide?​

26.1.4.6 to 32…

Great Tending: a way to farm and cultivate, grow and nurture, our understanding.

‘Great tending,
Constancy bears fruit.
Not eating at home, good fortune.
Fruitful to cross the great river.’

Asking an artificial intelligence to interpret a metaphor is certainly a way of ‘not eating at home’. It’s going to ‘feed’ us applications and ideas we would not have come up with ourselves, and this is a good thing. It’s worth taking the plunge.

This comes with a context of Lasting: Great Tending that endures over time. I like Perplexity.ai’s suggestion that this is the quality of the oracle itself, which has lasted through the inventions of coinage, printing, fruit and nut chocolate, the internet, etcetera, and hasn’t broken yet.

I can also imagine Lasting as a way of doing Great Tending: with constancy, over time, feeding and taming the animal daily with a view to growing it into something better. In this kind of lasting, we need to contribute the ‘direction to go’:

‘Thunder and wind, Lasting.
A noble one stands firm and does not change his bearings.’

(A question to ask before starting a conversation with AI about a reading: where are you going with this?)

The dangers at line 1​

‘There is danger.
Fruitful to stop it.’

To start with, there’s the one I mentioned in my last post: getting ‘what it says’ handed to you without ever engaging with the reading or experiencing a real connection with the oracle. This line changes to Hexagram 18, Corruption: definitely this is not a new danger.

There’s also the basic problem that AI is inclined to deceive people with wrong hexagrams and fake text. Not a problem if you already know the real thing, but a royal mess for those who don’t.

And also… while the AIs came up with some robust interpretations generally, I can’t help noticing that not one of them had a serious suggestion for what this danger might be that it would be fruitful to stop. But isn’t that the first thing a human being looking at this reading is going to ask? Danger? What? Where? (Also, how come there’s this danger and yet also ‘heaven’s highway’?)

This isn’t a sophisticated response: the birds in my garden do much the same when they hear an alarm call. We’re wired to pay attention to warnings, so these questions naturally come to mind. Only – of course – for an AI, they don’t. Gemini may have begun its response with ‘Interesting reading!’ but it can’t be interested; it can’t be alarmed; it can’t wonder what it means. There’s still no-one behind the curtain.

Gemini seemed quite proud of being objective, as it talked about Hexagram 32:

AI’s ability to provide consistent and unbiased interpretations: Unlike human interpreters who may be influenced by personal biases or emotional states, AI can offer consistent readings based on its programmed parameters.

That sounds enticing – or possibly like something entirely missing the point. Where does such pure, unbiased interpretation lead? (I have another post in the pipeline about that.)

The hornboard and the bullock​

This reading’s got me thinking about bullocks. The ones I walked past this morning were more interested in the grass than in the humans, though I’ve also met some that were inconveniently inquisitive. Full-grown bulls, on the other hand, are not left to graze on public footpaths. They’re anything up to a ton of probably-angry animal that can charge at over 30mph. (Fewer than 1 in 20 people attacked by a bull survive, according to Wikipedia.) So if a bull has it in mind to gore you, you need to be elsewhere – or, of course, you need to have controlled what it was going to have in mind a year or so earlier.

And this is where the hornboard comes in: imposing conscious, deliberate control and guidance on the great lumbering beast, early on, while you still can.

There is another layer of meaning to this line, because hornboards were fixed to especially good, spotless young bulls to mark them out as a potential offering to the spirits. So the hornboard does not only prevent harm; it also tends and nurtures a Great Possession (the hexagram this line changes to), something that could become sacred as well as dangerous.

This line combines strongly with line 1 –

‘There is danger.
Fruitful to stop it.’
‘Young bull’s hornboard
From the source, good fortune.’

Stop the danger by applying the hornboard; this might bear fruit as a sacred offering, the source of good fortune.

The hornboard looks to me like all we can do to control and direct the AI – and also our own mindless impulse to jump in and use it. Our desire to get all our questions answered is another animal that’s both potentially sacred and also quite stupid and dangerous. Affixing the hornboard early on means using AI deliberately, with an intelligent goal in mind.

How could we do that? (Besides preferably not handing control to Skynet…)

To start with, we need to cast the reading ourselves – AI, by default, does not know how, it only thinks it does – and then bolt on a real translation-hornboard, like Remo has done. Probably the only reliable way to do this with a ‘vanilla’ AI is to paste the translation into the chat, step by step. Given real text to work with, instead of random Yijing-adjacent inventions, the output is much better.

Apart from that, I think the prompt we use is our best hornboard. I’ve found noticeably better results when I ask specific follow-up questions: what can I learn from the trigrams; what might this specific image mean; how are these two moving lines connected to the resulting hexagram?

Is this heaven’s highway?​

This is not a line I would have expected to receive – but then I also wouldn’t have expected AI to respond intelligently to 3,000 year old imagery. I think we need to read it in the context of the whole – that is, lines 1 and 4 come first, and then we might reach heaven’s highway, if we use AI in this directed, controlled, carefully-cultivated way.

It will generate immediately-applicable commentary, after all, and come up with applications we might never have thought of. Besides being startlingly good with imagery, it’s also organised and thorough: it will interpret all the text is has, without skipping bits, and come up with a good, portable summary at the end.

Also…

Heaven’s highway rolls on: it always has and always will. This line changes to Hexagram 11, Flow, which I cast back when I asked how the Yi was made. It’s got along fine without AI for the past three millennia, and I expect it will manage for the next three, as well, with or without it. In other words, I think this line goes beyond the question and sets it in a little perspective. AI may be cluttering all the headlines at the moment, it may be in the process of rewriting our society and rotting our brains (or vice versa), but there are things it can’t break.

Some provisional suggestions for getting closer to line 6:

• cast a real reading and type a real translation into the chat (or use YiShi)
• stop and do your own interpreting first (it can be hard to connect with a reading when your mind is cluttered with someone – or something – else’s ideas)
• tell the AI about the trigrams and ask what they might mean
• ask it more questions about specific images

I should think there must be a lot of improvement still to come. What if AI read more books? (What if it could work directly from the original Chinese?) What if it could understand hexagrams and how they change into one another? What if it learned about nuclear trigrams, hexagram pairs, complements, the Sequence? It seems as though all that should be within its capabilities.

There are some things I suspect it will never manage, though, no matter how carefully it’s guided. More on that in the next post…

 

[my_key]

Reading the different replies here, it's clear that the AI in general needs to be taught a host of add on skills to make them more accomplished..

When remod was experimenting with YiShi, to find how to best 'calibrate' it, I was impressed by the graphic images that were produced to represent a couple of the readings. Being a visual person, I found this particularly insightful.
From a reading interpretation perspective I like the idea that AI readings provided a response that is free from any of my personal bias and prejudice that may unknowingly send me down a 'rabbit hole'. Equally, it will provide a safeguard against any bias and prejudice projected onto the reading by any human, which can be a useful thing should others be doing a reading for you.

As things stand at the moment, an AI interpretation would provide for me an impartial picture. I would always do my own interpretation first and not take the picture as gospel. Reflection on both readings would allow me to see a balanced perspective between my real and an artificial intelligence that may add a greater degree of clarity.

Finally, I think interpreting with AI could provide a useful learning tool for new starters. Although the quality of the AI 'understanding' may need some tweaking to make this a failsafe situation.

If the trend of accessing an AI reading moves towards it habitually being used as a quick and dirty, low time commitment, low personal investment, minimal self reflection tool then, human kind may just as well shut the door on this type of divination with the last one leaving switching off the light.

 

[hilary]

the AI in general needs to be taught a host of add on skills to make them more accomplished

Absolutely. At the moment it doesn't know what hexagrams are, so can't cast readings, deduce changing lines from hexagrams or vice versa, or 'understand' what it's looking at all. They're also far from consistent in getting the text right.

So until all that changes, I completely disagree with this:

Finally, I think interpreting with AI could provide a useful learning tool for new starters.

The trouble is, AI speaks with such perfect confidence. Here, said ChatGPT, is what Legge wrote about this line. There's nothing in its message to tell you when these bald assertions of fact are true and when they're complete inventions.

From a reading interpretation perspective I like the idea that AI readings provided a response that is free from any of my personal bias and prejudice that may unknowingly send me down a 'rabbit hole'. Equally, it will provide a safeguard against any bias and prejudice projected onto the reading by any human, which can be a useful thing should others be doing a reading for you.

Ah yes... the 'impartiality' and bias-free reading. Is this of benefit? I have one more post in the pipeline about this aspect.

If the trend of accessing an AI reading moves towards it habitually being used as a quick and dirty, low time commitment, low personal investment, minimal self reflection tool then, human kind may just as well shut the door on this type of divination with the last one leaving switching off the light.

True. But I'm not vastly worried about this. It seems to me not qualitatively different from all the 'Simple Positive I Ching For The Modern Era' type books, which strip out the imagery and just tell you what it means. They must put some people off for life - but every now and then they're also the gateway to getting a real translation.

 

[my_key]

Hi Hilary
There are many ways to skin a cat. but you do need a sharp knife. I agree with you that AI is not sharp enough as things stand.

The important thing, for me, is not to blindly follow gibberish, however cool, confident and well packaged it may seem. That calls for discerning engagement.

Impartiality is always a good mirror to hold up to anything, I Ching or otherwise. The filters that each of us humans erect to keep us safe and secure on many occasions go unnoticed or become invisible over time. A well polished mirror provides opportunity for a re balance. Again, though, I would question whether AI has a good enough reflective sheen on it to be totally effective here. I suspect though with the right question, in some cases, it would be a good enough tool for consideration in the uncovering of potential blind spots.

I'm sure this is a topic that will go on and on as AI is not going away any time soon.