This makes little sense to me. Ontologies and all that have been tried and have always been found to be too brittle. Take the examples from the front page (which I expect to be among the best in their set): human_activity => climate_change. Those are such a broad concepts that it's practically useless. Or disease => death. There's no nuance at all. There isn't even a definition of what "disease" is, let alone a way to express that myxomatosis is lethal for only European rabbits, not humans, nor gold fish.
Given we've tried to develop such ontologies constantly for thousands of years now. What do you think the cause for such hopeless optimism might be? If only we had a database of causal relationships to consult...
Democritus (b 460BCE) said, “I would rather discover one cause than gain the kingdom of Persia,” which suggests that finding true causes is rather difficult.
"According to the Greek historian Herodotus, Xerxes's first attempt to bridge the Hellespont ended in failure when a storm destroyed the flax and papyrus cables of the bridges. In retaliation, Xerxes ordered the Hellespont (the strait itself) whipped three hundred times, and had fetters thrown into the water."
Not so sure one should take stories about who said something in ancient times at face value ;)
If you think you can use logic to determine human behavior in the past, well, it doesn't even work for modern behavior lol You'd be surprised what kind of beliefs about the world led to what kind of actions in history
Felix, qui potuit rerum cognoscere causas. [0]
Virgil.
[0] https://en.m.wikipedia.org/wiki/Felix,_qui_potuit_rerum_cogn...
Virgili, hoc postremo dico: mihi nomen non est Felix.
Or is less of a hassle.
I totally agreed that in the past years of hammering out an ontology for a particular area just results in a common understanding between those who wrote the ontology and a large gulf between them and the people they want to use it ( everyone else ).
What's perhaps different is that the machine, via LLM's, can also have an 'opinion' on meaning or correctness.
Going fully circle I wonder what would happen if you got LLM's to define the ontology....
>what would happen if you got LLM's to define the ontology.
https://deepsense.ai/resource/ontology-driven-knowledge-grap...
>hammering out an ontology for a particular area just results in a common understanding between those who wrote the ontology and a large gulf between them and the people they want to use it
This is the other side of the bitter lesson, which is just the empirical observation of a phenomenon that was to be expected from first principles (algorithmic information theory): a program of minimal length must get longer if the reality it models becomes more complex.
For ontologists, the complexity of the task increases as the generality is maintained while model precision is increased (top down approach), or conversely, when precision is maintained the "glue" one must add to build up a bigger and bigger whole while keeping it coherent becomes more and more complex (bottom up approach).
But “disease => death” + AI => surely at least few billion in VC funding.
The best thing about this statement is that it can be read as 'the fact that disease causes death, plus the application of AI, will surely lead to billions VC funding' but it can also be read as 'disease is to death as AI is to a few billion in VC funding'. :D
Koller and Friedman write in "Probabilistic Graphical Models" about the "clarity test", so that state variables should be clear for an all seeing observer.
States like "human_activity" are not objectively measurable.
Fairly PGMs and causal models are not the same, but this way of thinking about state variables is an incredible good filter.
> States like "human_activity" are not objectively measurable.
Well, or at least they would need a heavy dose of operationalisation.
Exactly. In some cases disease causes death. In others it causes immunity which in turn causes “good health” and postpones death.
Contradictory cause-effect examples, each backed up with data, are a reliable indicator of a class of situations that need a higher chain-effect resolution.
Which is directly usable knowledge if you are building out a causal graph.
In the meantime, a cause and effect representation isn't limited to only listing one possible effect. A list of alternate disjoint effects, linked to a cause, is also directly usable.
Just as an effect may be linked to different causes. Which if you only know the effect, in a given situation, and are trying to identify cause, is the same problem in reverse time.
It is my opinion that if we examine any factor closely, it will have multiple disjoint effects. As in nothing is absolutely unilateral in its effects. Some of those effects will depend on certain conditions. If it is possible to specify condition, annotations, and other nuances such as levels of confidence or source of the opinion, such a database might be pretty useful.
It's pretty easy to outline a high level ontology and let LLMs annotate/link it into something pretty useful, you can even have a benchmark suite using that ontology via LLM as a judge to progressively optimize it.
What is an ontology exactly? I see Palantir talking about it all the time and it just sounds like vague marketing.
It comes from "the knowledge of being," and has been used to describe real-world knowledge representation, in particular hierarchical(-ish) semantic networks in AI since its early days.
When I see Palantir talk about it in a press release is that something real or just fluffy marketing?
These days anyone can spin up a developer account and check it out. Near as I could tell, you can create abstract 'objects' and link them to datasets/columns in the environment. And then you can link objects together. It's basically just an ER modeling tool, but they have great sales and seemed to have convinced people that they are constructing ontologies.
Ontology, not ontologies, have been tried.
We have quite a good understanding that a system cannot be both sound a complete, regardless people went straight in to make a single model of the world.
> a system cannot be both sound a complete
Huh, what do you mean by this? There are many sound and complete systems – propositional logic, first-order logic, Presburger arithmetic, the list goes on. These are the basic properties you want from a logical or typing system. (Though, of course, you may compromise if you have other priorities.)
My take is that the GP was implicitly referring to Gödel’s Incompleteness Theorems with the implication being that a system that reasons completely about all the human topics and itself is not possible. Therefore, you’d need multiple such systems (plural) working in concert.
A collection of systems is itself a system. The theorem would not recognize the distinction.
That doesn't make much sense.
If you take multiple systems and make them work in concert, you just get a bigger system.
> If you take multiple systems and make them work in concert, you just get a bigger system.
The conclusion may be wrong, but a "bigger system" can be larger than the sum of its constituents. So a system can have functions, give rise to complexity, neither of its subsystems feature. An example would be the thinking brain, which is made out of neurons/cells incapable of thought, which are made out of molecules incapable of reproduction, which are made from atoms incapable of catalyzing certain chemical reactions and so on.
This is just emergence, though? How is emergence related to completeness?
This happens over and over with the relatively new popularization of a theory: the theory is proposed to be the solution to every missing thing in the same rough conceptual vector.
It takes a lot more than just pointing in the general direction of complexity to propose the creation of a complete system, something which with present systems of understanding appears to be impossible.
> How is emergence related to completeness?
I didn't make that argument. I think, the original conclusion above isn't reasonable. However, "a concert" isn't "just" a bigger system either, which is my point.
It just depends on your definition of system, doesn’t it?
Sort of, the guardrail here IMO is you have an ontology processor that basically routes to a submodule, and if there isn't a submodule present it errors out. It is one large system, but it's bounded by an understanding of its own knowledge.
Concerts - again plural. And naturally you only bring in appropriate instruments.
Turtles all the way down?
I believe, neither the expansion of Gödel's theorems to "everything", non-formalized systems, nor the conclusion of a resolution by harnessing multiple systems in concert, are sound reasoning. I think, it's a fallacious reductionism.
What is a non-formalized system?
I am very curious on this. In particular, if you are able to split systems into formalized and non formalized, then I thinks there are quite some praise and a central spot in all future history books for you!
I am not a native speaker, so please don't get hung up on particular expressions.
I meant, the colloquial philosophies and general ontology are not subject of Gödel's work. I think, the forgone expansion is similar to finding evidence for telepathy in the pop-sci descriptions of quantum entanglement. Gödel's theorems cover axiomatic, formal systems in mathematics. To apply it to whatever, you first have to formalize whatever. Otherwise, it's an intuition/speculation, not sound reasoning. At least, that's my understanding.
Further reading: https://en.wikipedia.org/wiki/G%C3%B6del's_incompleteness_th...
Yep - when you use a multiplum of systems, then some systems can be regarded complete while other systems are sound.
This is in contrast to just one system that attempts to be sound and complete.
Could you define sound and complete in this context ? IIRC Rust's borrow checker is sound (will not mark something dysfunctional as functional) but not complete: some programs would take too long to verify, the checker times out, and compilation fails even though the program is potentially correct.
The meaning of the word person is ~sound (ie. Well defined) when two lawyers speak.
But when a doctor tells the lawyer that they operated a person, the lawyer can reasonably say "huh" - the concept of a person has shifted with the context.
Agreed. About the strongest we can hope for are causal mechanisms, and most of those will be at most hypotheses and/or partial explanations that only apply under certain conditions.
Honestly, I don’t know understand how these so-ontologies have persisted. Who is investing in this space, and why?
Even more importantly, it's not even a simple probability of death, or a fraction of a cause, or any simple one-dimensional aspect. Even if you can simplify things down to an "arrow", the label isn't a scalar number. At a bare minimum, it's a vector, just like embeddings in LLMs are!
Even more importantly, the endpoints of each such causative arrow are also complex, fuzzy things, and are best represented as vectors. I.e.: diseases aren't just simple labels like "Influenza". There's thousands of ever-changing variants of just the Flu out there!
A proper representation of a "disease" would be a vector also, which would likely have interesting correlations with the specific genome of the causative agent. [1]
Next thing is that you want to consider the "vector product" between the disease and the thing it infected to cater for susceptibility, previous immunity, etc...
A hop, skip, and a small step and you have... Transformers, as seen in large language models. This is why they work so well, because they encode the complex nuances of reality in a high-dimensional probabilistic causal framework that they can use to process information, answer questions, etc...
Trying to manually encode a modern LLM's embeddings and weights (about a terabyte!) is futile beyond belief. But that's what it would take to make a useful "classical logic" model that could have practical applications.
Notably, expert systems, which use this kind of approach were worked on for decades and were almost total failures in the wider market because they were mostly useless.
[1] Not all diseases are caused by biological agents! That's a whole other rabbit hole to go down.
You're losing interpretability and scrutability, but gaining detail and expressiveness. You have no way to estimate the vectors in a causal framework, all known methods are correlational. You have no clean way to map the vectors to human concepts. Vectors are themselves extremely compressed representations, there is no clear threshold beyond which a representation becomes "proper".
That was very well said.
One quibble, and really mean only one:
> a high-dimensional probabilistic causal framework
Deep learning models aka neural network type models, are not probabilistic frameworks. While we can measure on the outside a probability of correct answers across the whole training set, or any data set, there is no probabilistic model.
Like a Pachinko game, you can measure statistics about it, but the game itself is topological. As you point out very clearly, these models perform topological transforms, not probabilistic estimations.
This becomes clear when you test them with different subsets of data. It quickly becomes apparent that the probabilities of the training set are only that. Probabilities of the exact training set only. There is no probabilistic carry over to any subset, or for generalization to any new values.
They are estimators, approximators, function/relationship fitters, etc. In contrast to symbolic, hard numerical or logical models. But they are not probabilistic models.
Even when trained to minimize a probabilistic performance function, their internal need to represent things topologically creates a profoundly "opinionated" form of solution, as apposed to being unbiased with respect to the probability measure. The measure never gets internalized.
What’s the relationship between what you’re saying and the concepts of “temperature” and “stochasticity”? The model won’t give me the same answer every time.
The probability distribution that the model outputs is deterministic. The decoding method that uses that distribution to decide what next token to emit may or may not be deterministic. If we decide to define the decoding method as part of "the model", then I guess the model is probabilistic.
It's also worth noting that the parameters (weights and biases) of the model are random variables, technically speaking, and this can be considered probabilistic in nature. The parameter estimates themselves are not random variables, to state the obvious. The estimates are simply numbers.
As I understand it, this is a dataset of claimed causation. It should contain vaccines->autism, not because it's true, but because someone, in public, claimed that it was.
So, by design, it's pretty useless for finding new, true causes. But maybe it's useful for something else, such as teaching a model what a causal claim is in a deeper sense? Or mapping out causal claims which are related somehow? Or conflicting? Either way, it's about humans, not about ontological truth.
Also, it seems to mistake some definitions as causes.
A coronavirus isn't "claimed" to cause SARS. Rather, SARS is a name given to the disease cause by a certain coronavirus. Or alternatively, the name SARS-nCov-1 is the name given to the virus which causes SARS. Whichever way you want to see it.
For a more obvious example, saying "influenza virus causes influenza" is a tautology, not a causal relationship. If influenza virus doesn't cause influenza disease, then there is no such thing as an influenza virus.
Yes, I agree there are a lot of definitions or descriptions masquerading as explanations, especially in medicine and psychology. I think maybe insurance has a lot to do that. If you just describe a lot of symptoms, insurance won't know whether to cover it or not. But if you authoritatively name that symptom set as "BWZK syndrome" or something, and suddenly switch to assuming "BWZK syndrome" is a thing, the unknown cause to the symptoms, then insurance has something it can deal with.
But this description->explanation thing, whatever the reason, is just another error people make. It's not that different from errors like "vaccines cause autism". Any dataset collecting causal claims people make is going to contain a lot of nonsense.
Isn't this like Cyc? There have been a couple of interesting articles about that on HN:
https://news.ycombinator.com/item?id=43625474 "Obituary for Cyc"
https://news.ycombinator.com/item?id=40069298 "Cyc: History's Forgotten AI Project"
Cyc is hardly [1] mentioned in modern work under the knowledge representation and reasoning umbrella, because most [2] of it was/is unavailable or unknown to most researchers. It's hard to build on something that's primarily marketing material.
[1] I could be wrong, but even those that mention Cyc use it only as a historical example of early work in KRR / symbolic AI. [2] OpenCyc being the small subset which is available, tho I haven't met anyone who worked with it.
Seems like a subset of CYC - attempting to gather causal data rather than declarative data in general.
It's a bit odd that their paper doesn't even mention CYC once.
Everything old is new again
I read it as "casual" rather than "causal", got very dissapointed while reading the article!
An inventory of casual knowledge would be really fun, although it's hard to think what it would consist of now that I think about it...
There is this concept of "hidden knowledge" about all the things you know at work that no one really thinks about is knowledge so it's hard to let newcomers know about it.
But that does sound different than "casual knowledge", and so does "trivia".
Oh well!
I did, too! And it reminded me of a project idea I had a while ago:
A time traveller's wiki that collects casual knowledge for different times (and different places).
Such as: "Buying a train ticket in Paris in 1972".
But it was a shower thought and it's pretty hard to imagine how this knowledge should be collected and especially presented.
In a way, wikipedia is already doing this by keeping records of articles as they change over the years :)
The article about train tickets wasn't so good as an example but "computer monitor" from 2004 is kind of fun to read :)
Unfortunately, "casual knowledge" is often omitted when writing informative articles. In this example, there is no mention that power buttons are often located somewhere in the back of the monitor, which was good to know in 2004. Also, some monitors are drawing power from the computer, thus they won't power up before the computer will. And speaking of that: You may want to turn of your computer after shutdown!
Edit: This would probably be useful for novelists and filmmakers (in addition to the casual time traveller)
it's as simple as precisely describing "common sense"
Yes, that very simple task :D
But I also wonder if that is actually an equivalent.
I think of "common sense" as think you should know to function well in your current society.
I really don't know what "casual knowledge" would mean. In my head it's some kind of low stakes knowledge for everyday life (but more 'useful' than trivia).
Maybe the order is "common sense", "casual knowledge" and trivia?
The sample set contains:
{
"causal_relation": {
"cause": {
"concept": "boom"
},
"effect": {
"concept": "bust"
}
}
}
Plus, regardless of what you might think of how valid that connection is, what they're actually collecting, absent any kind of mechanism, is a set of all apparent correlations...
This might be of at least some value to augment training LLMs? I spent a lot of time in the 1980s and early 1990s using symbolic AI techniques: conceptual dependency, NLP, expert systems, etc. While two large and well funded expert system projects I worked on (paid for by DARPA and PacBell) worked well, mostly symbolic AI was brittle and required what seemed like an i finite amount of human labor.
LLMs are such a huge improvement that the only real use I see in projects like Cause et, the defunct OpenCyc project, etc. the only possible practical use might be as a little extra training data.
Reminds me of the cyc project. https://en.wikipedia.org/wiki/Cyc
Organizing all knowledge requires a flexible system of organization (starting with how the categories are organized and accessed, not the data).
Random thoughts about organizing knowledge:
- Categories need fractal structures.
- Categories need to be available as subcategories to other categories as a pattern.
- Words need to be broken down into base concepts and used as patterns.
- Social information and context alter the meaning of words in many cases, so any semantic web without a control system has limited use as an organization tool.
Might as well go ahead and add https://tylervigen.com/spurious-correlations?page=135 from the looks of it.
The associated paper references Judea Pearl's theories on causality, but curiously doesn't mention the DoWhy implementation [0], which seems to have some recognition in the causal inference space.
This reminds me of an article I read that was posted on HN only a few days ago: Uncertain<T>[1]. I think that a causality graph like this necessarily needs a concept of uncertainty to preserve nuance. I don't know whether this would be practical in terms of compute, but I'd think combining traditional NLP techniques with LLM analysis may make it so?
I get some vibes of fuzzy logic from this project.
Currently a lot of people research goes in the direction that there is "data uncertainty" and "measurement uncertainty", or "aleatoric/epistemic" uncertainty.
I foumd this tutorial (but for computer vision ) to be very intuitive and gives a good understanding how to use those concepts in other fields: https://arxiv.org/abs/1703.04977
Right. The first example on the site shows disease as a cause, and death as an effect. This is wrong on several levels: There is no such thing as healthy or sick. You’re always fighting off something, it just becomes obvious sometimes. Also, a disease doesn’t necessarily lead to death, obviously.
Since you're always going to die, the problem is solved - the implication is true by the right side always being true, and the left side doesn't matter.
Then it’s correlation instead of causation and the entire premise of a causation graph is moot.
Why not use PROLOG then, is the essence of cause and effect in programming. And also can expound syllogisms.
The conditional relation represented in prolog, and in any deductive system, is material implication (~PvQ), not causation. You can encode causal relationships with material implication but you’re still going to need to discover those causal relationships in the world somehow.
Conditional statements don't really work because "if A, then B" means that A is sufficient for B, but "A causes B" doesn't imply that A is sufficient for B. E.g. in "Smoking causes cancer", where smoking is a partial cause for cancer, or cancer partially an effect of smoking.
"A causes B" usually implies that A and B are positively correlated, i.e. P(A and B) > P(A)×P(B), but even that isn't always the case, namely when there is some common cause which counteracts this correlation.
Thinking about this, it seems that if A causes B, the correlation between A and B is at least stronger than it would have been otherwise.
This counterfactual difference in correlation strength is plausibly the "causal strength" between A and B. Though it doesn't indicate the causal direction, as correlation is symmetric.
"The map is not the territory" ensures that bias and mistakes are inextricable from the entire AI project. I don't want to get all Jaron Lanier about it, but they're fundamental terms in the vocabulary of simulated intelligence.
Reminds me of the early attempts at hand categorising knowledge for AI
I wonder how they will quantize causality. Sometimes a particular cause has different, and even opposite, effects.
Alcohol causes anxiety. At the same time it causes relaxation. These effects depend on time frame, and many individual circumstances.
This is a single example but the world is full of them. Codifying causality will involve a certain amount of bias and belief. That does not lead to a better world.
It's nice to see more semantic web experiments. I always wanted to do more reasoning with ontologies, etc., and it's such an amazing idea, to reference objects/persons/locations/concepts from the real world with uris and just add labeled arrows between them.
This is such a cool schemaless approach and has so much potential for open data linking, classical reasoning, LLM reasoning. But open data (together with RSS) has been dead for a while as all big companies have become just data hoarders. And frankly, while the concept and the possibilities are so cool, the graph databases are just not that fast and also not fun to program.
semantic web/OWL was always way too heavy to imagine humans using, you could imagine AI doing the heavy lifting here though..
As a trader, I’m always trying to piece together all the causal factors that move markets. It’s like building a mental map of cause and effect so I can make sharper, faster decisions
A cool idea, in desperate need of an example use case.
Can't an LLM extract this type of information with reasonably high accuracy?
This is difficult, but then I just had someone earnestly inform me that the covid virus doesn't cause covid, so I think there's a need here, if only to have an automated way of identifying idiots.
Wittgenstein is calling
I was hoping this would be actual normalized time series data and correlation ratios. Such a dataset would be interesting for forecasting.
I know it's a reductive take to point to a single mistake and act like the whole project might be a bit futile (maybe it's a rarity) but this example in their sample is really quite awful if the idea is to give AI better epistemics:
{
"causal_relation": {
"cause": {
"concept": "vaccines"
},
"effect": {
"concept": "autism"
}
}
},
In the precision dataset, there are the sentences that led to this, some are:
>> "Even though the article was fraudulent and was retracted, 1 in 4 parents still believe vaccines can cause autism."
>> On 28 February 1998 Horton published a controversial paper by Dr. Andrew Wakefield and 12 co-authors with the title "Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children" suggesting that vaccines could cause autism.
>> He was opposed by vaccine critics, many of whom believe vaccines cause autism, a belief that has been rejected by major medical journals and professional societies.
All that I've seen don't actually say that vaccines cause autism
Oh, ouch, yeah. We already know that misinformation tends to get amplified, the last thing we need is a starting point full of harmful misinformation. There are lots of "causal beliefs" on the internet that should have no place in any kind of general dataset.
It's even worse than that, because the way they extract the causal link is just a regex, so
"vaccines > autism"
because
"Even though the article was fraudulent and was retracted, 1 in 4 parents still believe vaccines can cause autism."
I think this could be solved much better by using even a modestly powerful LLM to do the causal extraction... The website claims "an estimated extraction precision of 83% " but I doubt this is an even remotely sensible estimate.
the cyc of this current ai winter
Causality is literally impossible to deduce...
> CauseNet aims at creating a causal knowledge base that comprises all human causal knowledge and to separate it from mere causal beliefs
Pretty bold to use a picture of philosophers as your splash page and then make a casual claim like this. To say the least, this is an impossible task!
The tech looks cool and I'm excited to see how I might be able to work it into my stuff and/or contribute. But I'd encourage the authors to reign in the rhetoric...
Indeed. I can't take an epistemology project seriously if it has no humility.
Building a perfectly accurate model of the world isn't possible. We need to create tools that make it easier for regular people to build more accurate models, not delude ourselves with dreams of perfection.
Well of course because no such model of the world can or does exist
I wonder what is this for.
I don’t know if it’s inadvertent, but it’s headed toward just becoming an engine for over fitted generalizations. Each casual pair will just emerge based on frequency, which will reinforce itself in preemptively and prematurely classifying all future information.
Unfortunately, frequency is the primary way AI works, but it will never be accurate for causality because causality always has the dynamic that things can happen just “because”. It’s hacked into LLMs via deliberate randomness in next-token prediction.
this will be super cool if it can be done!
I think this is many years old