Situation Analysis
CHAT-GPT offers powerful language processing that driven by reasoning and inference. Thus, it would be natural to explore the degree to which said functionality applies to data incorporated into a questions and answering session.
In this post, and the reply that follows, I provide an experiment testing the nature of reasoning and inference offered by CHAT-GPT (using the GPT4 Large Language Model [LLM]) when applied to structured data represented using RDF-Turtle Notation.
Note, you can visualize the structured data used in this post via your browser, if you choose, courtesy of the OpenLink Data Sniffer (OSDS) Browser Extension for your preferred browser.
Session Transcript
## Turtle Start ##
## Entity Relationships
### TBox
@prefix rel: <http://purl.org/vocab/relationship/> .
@prefix owl: <http://www.w3.org/2002/07/owl#> .
rel:parentOf a owl:TransitiveProperty .
### ABox
<http://dbpedia.org/resource/Prince_William_of_Wales> a <#RoyalPerson>, <#MalePerson> ;
schema:name "Prince William" ;
rel:siblingOf <http://dbpedia.org/resource/Prince_Harry_of_Wales> .
<http://dbpedia.org/resource/Elizabeth_Bowes-Lyon> a <#RoyalPerson>, <#FemalePerson> ;
schema:name "Queen Mother" ;
rel:parentOf <http://dbpedia.org/resource/Elizabeth_II_of_the_United_Kingdom> .
<http://dbpedia.org/resource/Elizabeth_II_of_the_United_Kingdom> a <#RoyalPerson>, <#FemalePerson> ;
schema:name "Queen Elizabeth II" ;
rel:parentOf <http://dbpedia.org/resource/Charles,_Prince_of_Wales>,<http://dbpedia.org/resource/Anne,_Princess_Royal>,
<http://dbpedia.org/resource/Prince_Andrew,_Duke_of_York>, <http://dbpedia.org/resource/Prince_Edward,_Earl_of_Wessex> ;
rel:siblingOf <http://dbpedia.org/resource/Princess_Margaret,_Countess_of_Snowdon> .
<http://dbpedia.org/resource/Charles,_Prince_of_Wales> a <#RoyalPerson>, <#MalePerson> ;
schema:name "Prince Charles" ;
rel:parentOf <http://dbpedia.org/resource/Prince_William_of_Wales> , <http://dbpedia.org/resource/Prince_Harry_of_Wales> .
<http://dbpedia.org/resource/Princess_Margaret,_Countess_of_Snowdon> a <#RoyalPerson>, <#FemalePerson> ;
schema:name "Princess Margaret" .
<http://dbpedia.org/resource/Anne,_Princess_Royal> a <#RoyalPerson>, <#FemalePerson> ;
schema:name "Princess Anne" ;
rel:parentOf <http://dbpedia.org/resource/Peter_Phillips>, <http://dbpedia.org/resource/Zara_Phillips> .
<http://dbpedia.org/resource/Zara_Phillips> a <#RoyalPerson>, <#FemalePerson> ;
schema:name "Princess Zara Phillips" .
<http://dbpedia.org/resource/Princess_Beatrice_of_York> a <#RoyalPerson>, <#FemalePerson> ;
schema:name "Princess Beatrice" .
<http://dbpedia.org/resource/Princess_Eugenie_of_York> a <#RoyalPerson>, <#FemalePerson> ;
schema:name "Princess Eugenie" .
<http://dbpedia.org/resource/Prince_Andrew,_Duke_of_York> a <#RoyalPerson>, <#MalePerson> ;
schema:name "Prince Andrew" ;
rel:parentOf <http://dbpedia.org/resource/Princess_Eugenie_of_York>, <http://dbpedia.org/resource/Princess_Beatrice_of_York> .
<http://dbpedia.org/resource/Prince_Edward,_Earl_of_Wessex> a <#RoyalPerson>, <#MalePerson> ;
schema:name "Prince Edward" .
<http://dbpedia.org/resource/Prince_Harry_of_Wales> a <#RoyalPerson>, <#MalePerson> ;
schema:name "Prince Harry" .
<http://dbpedia.org/resource/Peter_Phillips> a <#RoyalPerson>, <#MalePerson> ;
schema:name "Prince Peter Phillips" .
## Turtle End ##
Q:
Given the dataset above, and the nature of relations defined in the TBox section, list all siblings.
A:
Based on the provided Turtle dataset, here is a list of all the siblings:
- Prince William and Prince Harry are siblings.
- Prince William: About: William, Prince of Wales
- Prince Harry: About: Prince Harry, Duke of Sussex
- Queen Elizabeth II and Princess Margaret are siblings.
- Queen Elizabeth II: About: Elizabeth II
- Princess Margaret: About: Princess Margaret, Countess of Snowdon
- Princess Beatrice and Princess Eugenie are siblings.
- Princess Beatrice: About: Princess Beatrice
- Princess Eugenie: About: Princess Eugenie
These sibling relationships are identified based on the “rel:siblingOf” property in the dataset.
Q:
List all the aunties.
A:
Based on the provided Turtle dataset, here is a list of all the aunts (aunties):
- Princess Anne is the aunt of Prince William and Prince Harry.
- Princess Anne: About: Anne, Princess Royal
- Prince William: About: William, Prince of Wales
- Prince Harry: About: Prince Harry, Duke of Sussex
- Princess Margaret is the aunt of Prince Charles, Princess Anne, Prince Andrew, and Prince Edward.
- Princess Margaret: About: Princess Margaret, Countess of Snowdon
- Prince Charles: About: Charles III
- Princess Anne: About: Anne, Princess Royal
- Prince Andrew: About: Prince Andrew, Duke of York
- Prince Edward: About: Prince Edward, Earl of Wessex
- Princess Anne is the aunt of Princess Beatrice and Princess Eugenie.
- Princess Anne: About: Anne, Princess Royal
- Princess Beatrice: About: Princess Beatrice
- Princess Eugenie: About: Princess Eugenie
These aunt relationships are identified based on the “rel:parentOf” and “rel:siblingOf” properties in the dataset, where an aunt is the sibling of a parent.
Q:
Redo, grouping aunties in easy to read form.
A:
Certainly! Here is a list of all the aunts (aunties) grouped in an easy-to-read form:
- Aunt: Princess Anne (About: Anne, Princess Royal)
- Niece/Nephew: Prince William (About: William, Prince of Wales)
- Niece/Nephew: Prince Harry (About: Prince Harry, Duke of Sussex)
- Niece/Nephew: Princess Beatrice (About: Princess Beatrice)
- Niece/Nephew: Princess Eugenie (About: Princess Eugenie)
- Aunt: Princess Margaret (About: Princess Margaret, Countess of Snowdon)
- Niece/Nephew: Prince Charles (About: Charles III)
- Niece/Nephew: Princess Anne (About: Anne, Princess Royal)
- Niece/Nephew: Prince Andrew (About: Prince Andrew, Duke of York)
- Niece/Nephew: Prince Edward (About: Prince Edward, Earl of Wessex)
These aunt relationships are identified based on the “rel:parentOf” and “rel:siblingOf” properties in the dataset, where an aunt is the sibling of a parent.
Q:
Redo, but for Uncles.
A:
Certainly! Here is a list of all the uncles grouped in an easy-to-read form:
- Uncle: Prince Charles (About: Charles III)
- Niece/Nephew: Princess Beatrice (About: Princess Beatrice)
- Niece/Nephew: Princess Eugenie (About: Princess Eugenie)
- Uncle: Prince Andrew (About: Prince Andrew, Duke of York)
- Niece/Nephew: Prince William (About: William, Prince of Wales)
- Niece/Nephew: Prince Harry (About: Prince Harry, Duke of Sussex)
- Uncle: Prince Edward (About: Prince Edward, Earl of Wessex)
- Niece/Nephew: Prince William (About: William, Prince of Wales)
- Niece/Nephew: Prince Harry (About: Prince Harry, Duke of Sussex)
These uncle relationships are identified based on the “rel:parentOf” and “rel:siblingOf” properties in the dataset, where an uncle is the sibling of a parent.
Observation
The Uncles response above is incorrect, and I opted not to prompt further.
Conclusion
CHAT-GPT is primarily useful to operators that are knowledgeable in the subject-matter domain being operated on. Basically, it will dangerously mislead anyone seeking an all-answering oracle for a vacuous mind!
Recreating this exercise in your own setup
You need to use the GTP4 engine rather than GPT 3.5 (or lower) to replicate these results.
OpenLink Data Sniffer (OSDS) Browser Extension Links
- Google Chrome Store for Chromium-based Browsers
- Mozilla Store for Firefox 1
- Mac App Store for Safari 1
- Github Open Source Repository
Related
- Custom Reasoning & Inference using British Royal Family Relationships – Part 2
- Creating Custom Inference Rules using the SPIN Vocabulary and Virtuoso 8.0 | by Kingsley Uyi Idehen | OpenLink Virtuoso Weblog | Medium
- ChatGPT Prompt Collection
- ChatGPT Prompt Collection RSS Feed
- ChatGPT Prompt Collection Atom Feed
- ChatGPT Nanotations Collections – these docs combine prompts and the structured data generated by answers that double as LOD Cloud Knowledge Graph and Semantic Web entry points