One Genius' Lonely Crusade to Teach a Computer Common Sense (2016)
wired.com
wired.com
Dead Internet Theory is already in motion.
The issue of the public not being able to tell truth from fiction has existed since the beginning of society. Every improvement in communications technology has led to a massive increase in propaganda and confusion that spikes and then eventually levels out as people learn to tune it out.
Increasing the rate at which people can deploy lies has happened many times in history. This is just one more instance.
If it weren't for your comment, how could I have heard or read about Eurisko?
https://writings.stephenwolfram.com/2023/09/remembering-doug...
Effectively, you need to understand (compute with) what concepts are before using them in statements. And the most common mistake is to confuse the definition of a concept with the concept itself, as happens typically with knowledge graphs.
Source?
> Humans do think with first order predicate calculus. We juggle a bunch of statements/rules the way Cyc does, and don't have a very fast, efficient and powerful conceptual approach to the world with a massive amount of recursive conceptual inference.
Great discussion, thanks.
“At one point, Lenat remembers, it suggested he could win the game by changing the rules. But each morning, Lenat would rejigger the system, pushing Eurisko away from the ridiculous and toward the practical. In other words, he would lend the machine a little common sense.”
Really enjoyed this 2016 article.
For safety-critical applications that use black-box ML models, I feel like a rules-engine paired with a knowledge graph needs to kinda sandwich the ML black box —- to “prompt engineer” the input and ensure the output aligns with common sense and safety standards.
Edit: clarity
I think that the space shuttle did feature this kind of system though.
I was reading through the source code for both (and the Interlisp manual, but mostly Eurisko) earlier this year and found it quite inspiring. Some of the most meta(-meta-heuristic search) code ever written! Now I'm working on a similar project and am looking for parts of AM's and Eurisko's algorithms which I can incorporate! I'll write something up if I have any success.
Eurisko is far more abstract than AM, but the data structures (structures with named 'slots' (members)) it uses makes the code vastly more readable (AM is pretty damn unreadable despite plenty of comments). Lenat himself said the better representations are what made Eurisko possible, unsurprisingly. If you do want to understand the implementation of AM in incredible detail you can read Lenat's PhD thesis. No detailed documentation of Eurisko exists. Also, both programs contain lots of code to print out what the system is currently doing, which is incredibly helpful for deciphering the code.
[1] OpenCyc CVS repository on sourceforge: https://sourceforge.net/p/opencyc/code/
[2] Somebody made the effort to download/upload everything together on GitHub (with state of OpenCyc of 03/2018): https://github.com/asanchez75/opencyc
But I imagine it's still a great resource. Haven't played with it.
https://www.theguardian.com/commentisfree/2024/sep/30/i-took...
The article links to the study.
> Common sense is not that common: a recent study from the University of Pennsylvania concludes the concept is “somewhat illusory”. Researchers collected statements from various sources that had been described as “common sense” and put them to test subjects.
> The mixed bag of results suggested there was “little evidence that more than a small fraction of beliefs is common to more than a small fraction of people”.
What this study does show that is relevant for the submitted article is that whatever he trains the AI for, it may not be something "common".
When someone tries to give an AI "common sense" talking about exactly what I said is important! Because that person is NOT going to do anything "common", (s)he is implementing their own version and biases!
That distinction is at the very heart of the matter attempting to be achieved, not some "nitpicking" about terms!
And the author's person seems really concerning. She is someone, who, by her own admission struggles with grade school math such as fractions, yet proclaims her intellectual superiority over people who think 'obviously' silly things, like Ivermectin curing covid.
The idea of horse medicine curing Covid makes about as much sense as heart medicine helping with erections.
Although, somewhat amusingly, she seems to score below average on the common sense test.
The very cynical take of journalists being both substandard critical thinkers, and unwilling consider alternative viewpoints seems to be true in this case.
Why should it be nonsense? Erections are a function of the circulatory system. One could expect them improve if the circulatory system is improved as well.
His intent (as he stated online, and not mentioned in the Wikipedia page) was to have a collection of sentences which could be used to give an AI some idea of what it was like to be human.
I'm extremely skeptical about the anecdotes about the game as an indicator of this thing's competence. It sounds unlikely that this thing actually encoded any sort of game state or nuanced simulations and was really just spitballing on vague strategies that just happened to find some cheese (twice?). I'm guessing they had to play the strats until one of them proved valuable, and it's kind of weird and surprising that they thought this was a good use of their time and model.
He died August 31, 2023, age 72.
Interesting read considering it was written right before the advent of llms
Is Cyc essentially a 15 million line long Prolog program? Based on the article’s hand wavy description that’s the best I can put together.
And from that wikipedia article: "The dream of building systems that discover scientific hypotheses was pushed to the background with the second AI winter and the subsequent resurgence of subsymbolic methods such as neural networks. Subsymbolic methods emphasize prediction over explanation, and yield models which works well but are difficult or impossible to explain which has earned them the name black box AI. A black-box model cannot be considered a scientific hypothesis, and this development has even led some researchers to suggest that the traditional aim of science - to uncover hypotheses and theories about the structure of reality - is obsolete.[7][8] Other researchers disagree and argue that subsymbolic methods are useful in many cases, just not for generating scientific theories."
which sounds very intriguing to me :-)
There was a source-available version called OpenCyc with a small subset of the knowledge base that is now retired and no longer officially available, but it's still easy to find on the net.
We had used a primitive rolling hand scanner to scan in huge tables of metal alloy data along with recommended usages. My father wrote the query engine using Savvy and I wrote the text based windowing UI.
Would be neat if someone could turn up some record of it somewhere. My father has been retired since ever and has no records of the system either.
You can approach the group who are apparently now known as Materials Australia https://www.materialsaustralia.com.au/ and they should have an archivist or historian with access to period records and publications.
Failing that, visit the state library in your state and request the assistance of a research librarian. They should be able to help you locate potential records or journals if it was ever discussed.