Traditional explanatory resources, such as user manuals and textbooks, often contain content that may not cater to the diverse backgrounds and information needs of users. Yet, developing intuitive, user-centered methods to effectively explain complex or large amounts of information is still an open research challenge. In this paper we present ExplanatoryGPT, an approach we devised and implemented to transform textual documents into interactive, intelligent resources, capable of offering dynamic, personalized explanations. Our approach uses state-of-the-art question-answering technology to generate on-demand, expandable explanations, with the aim of allowing readers to efficiently navigate and comprehend static materials. ExplanatoryGPT integrates ChatGPT, a state-of-the-art language model, with Achinstein’s philosophical theory of explanations. By combining question generation and answer retrieval algorithms with ChatGPT, our method generates interactive, user-centered explanations, while mitigating common issues associated with ChatGPT, such as hallucinations and memory shortcomings. To showcase the effectiveness of our Explanatory AI, we conducted tests using a variety of sources, including a legal textbook and documentation of some health and financial software. Specifically, we provide several examples that illustrate how ExplanatoryGPT excels over ChatGPT in generating more precise explanations, accomplished through thoughtful macro-planning of explanation content. Notably, our approach also avoids the need to provide the entire context of the explanation as a prompt to ChatGPT, a process that is often not feasible due to common memory constraints.
2. Outline
The contents of this presentation will be
organised as follows:
• Introduction
• Background
• Proposed Approach
• Experiment
• Discussion
• Conclusions
2
3. Outline
The contents of this presentation will be
organised as follows:
• Introduction
• Background
• Proposed Approach
• Experiment
• Discussion
• Conclusions
3
4. Outline
The contents of this presentation will be
organised as follows:
• Introduction
• Background
• Proposed Approach
• Experiment
• Discussion
• Conclusions
4
5. Outline
The contents of this presentation will be
organised as follows:
• Introduction
• Background
• Proposed Approach
• Experiment
• Discussion
• Conclusions
5
6. Outline
The contents of this presentation will be
organised as follows:
• Introduction
• Background
• Proposed Approach
• Experiment
• Discussion
• Conclusions
6
7. Outline
The contents of this presentation will be
organised as follows:
• Introduction
• Background
• Proposed Approach
• Experiment
• Discussion
• Conclusions
7
8. Outline
The contents of this presentation will be
organised as follows:
• Introduction
• Background
• Proposed Approach
• Experiment
• Discussion
• Conclusions
8
9. Problem and Motivation
• Problem: Too much static information. How
to explain it?
• Advancements in NLP, such as models like
ChatGPT, can provide dynamic explanations.
• However, these models introduce challenges
like hallucinations.
• Memory constraints are another challenge
associated with using NLP models for
dynamic explanations.
9
10. Problem and Motivation
• Problem: Too much static information. How
to explain it?
• Advancements in NLP, such as models like
ChatGPT, can provide dynamic explanations.
• However, these models introduce challenges
like hallucinations.
• Memory constraints are another challenge
associated with using ChatGPT.
10
11. Problem and Motivation
• Problem: Too much static information. How
to explain it?
• Advancements in NLP, such as models like
ChatGPT, can provide dynamic explanations.
• However, these models introduce challenges
like hallucinations.
• Memory constraints are another challenge
associated with using ChatGPT.
11
12. Problem and Motivation
• Problem: Too much static information. How
to explain it?
• Advancements in NLP, such as models like
ChatGPT, can provide dynamic explanations.
• However, these models introduce challenges
like hallucinations.
• Memory constraints are another challenge
associated with using ChatGPT.
12
13. Proposal: ExplanatoryGPT
• ExplanatoryGPT is a combination of ChatGPT with
answer retrieval algorithms.
• It’s designed to address memory constraints and
hallucinations
• ExplanatoryGPT integrates question generation
and answer retrieval algorithms. It’s based on
Achinstein's theory of explanations to address the
abovementioned issues.
• Hallucination: the generation of content that does
not maintain fidelity to a given context or source.
13
14. Proposal: ExplanatoryGPT
• ExplanatoryGPT is a combination of ChatGPT with
answer retrieval algorithms.
• It’s designed to address memory constraints and
hallucinations
• ExplanatoryGPT integrates question generation
and answer retrieval algorithms. It’s based on
Achinstein's theory of explanations to address the
abovementioned issues.
• Hallucination: the generation of content that does
not maintain fidelity to a given context or source.
14
15. Proposal: ExplanatoryGPT
• ExplanatoryGPT is a combination of ChatGPT with
answer retrieval algorithms.
• It’s designed to address memory constraints and
hallucinations
• ExplanatoryGPT integrates question generation
and answer retrieval algorithms. It’s based on
Achinstein's theory of explanations to address the
abovementioned issues.
• Hallucination: the generation of content that does
not maintain fidelity to a given context or source.
15
16. Proposal: ExplanatoryGPT
• ExplanatoryGPT is a combination of ChatGPT with
answer retrieval algorithms.
• It’s designed to address memory constraints and
hallucinations
• ExplanatoryGPT integrates question generation
and answer retrieval algorithms. It’s based on
Achinstein's theory of explanations to address the
abovementioned issues.
• Hallucination: the generation of content that does
not maintain fidelity to a given context or source.
16
17. Background
• Achinstein: Explaining is an illocutionary act with
the intention to provide new understandings to the
explainee by answering questions.
• Illocution (in explanations):
– Equivalent to pertinently anticipating implicit
(archetypal) questions, e.g., Why? What for? How?
When? etc.
– It’s the difference between answering questions
and explaining.
• Example: “How are you?”
– “Alright”: not an explanation.
– “I’m good, because I just successfully defended my
PhD, etc.”
17
18. Background
• Achinstein: Explaining is an illocutionary act with
the intention to provide new understandings to the
explainee by answering questions.
• Illocution (in explanations):
• Equivalent to pertinently anticipating implicit
(archetypal) questions, e.g., Why? What for? How?
When? etc.
• It’s the difference between answering questions
and explaining.
• Example: “How are you?”
• “Alright”: not an explanation.
• “I’m good, because I just successfully defended my
PhD, etc.”
18
19. Background
• Achinstein: Explaining is an illocutionary act with
the intention to provide new understandings to the
explainee by answering questions.
• Illocution (in explanations):
• Equivalent to pertinently anticipating implicit
(archetypal) questions, e.g., Why? What for? How?
When? etc.
• It’s the difference between answering questions
and explaining.
• Example: “How are you?”
– “Alright”: not an explanation.
– “I’m good, because I just successfully defended my
PhD, etc.”
19
20. Background
• Achinstein: Explaining is an illocutionary act with
the intention to provide new understandings to the
explainee by answering questions.
• Illocution (in explanations):
• Equivalent to pertinently anticipating implicit
(archetypal) questions, e.g., Why? What for? How?
When? etc.
• It’s the difference between answering questions
and explaining.
• Example: “How are you?”
• “Alright”: not an explanation.
• “I’m good, because I just successfully defended my
PhD, etc.”
20
21. Background
• Achinstein: Explaining is an illocutionary act with
the intention to provide new understandings to the
explainee by answering questions.
• Illocution (in explanations):
• Equivalent to pertinently anticipating implicit
(archetypal) questions, e.g., Why? What for? How?
When? etc.
• It’s the difference between answering questions
and explaining.
• Example: “How are you?”
• “Alright”: not an explanation.
• “I’m good, because I just successfully defended my
PhD, etc.”
21
22. Background
• Achinstein: Explaining is an illocutionary act with
the intention to provide new understandings to the
explainee by answering questions.
• Illocution (in explanations):
• Equivalent to pertinently anticipating implicit
(archetypal) questions, e.g., Why? What for? How?
When? etc.
• It’s the difference between answering questions
and explaining.
• Example: “How are you?”
• “Alright”: not an explanation.
• “I’m good, because I just successfully defended my
PhD, etc.”
22
23. Why Achinstein’s Theory?
• This definition of illocution is computer-friendly.
We can use linguistic theories to define and
generate implicit archetypal questions and AI can
help us estimate how pertinently they are
answered.
• Our tech: textbook or documentation
information is extracted and organised by
question-answering algorithms.
• ChatGPT is used to refine the retrieved
information, improving readability, coherence,
and cohesion.
23
24. Why Achinstein’s Theory?
• This definition of illocution is computer-friendly.
We can use linguistic theories to define and
generate implicit archetypal questions and AI can
help us estimate how pertinently they are
answered.
• Our tech: answer retrieval algorithms which can
extract information out of textbooks or
documentation.
• ChatGPT is used to refine the retrieved
information, improving readability, coherence,
and cohesion.
24
25. Why Achinstein’s Theory?
• This definition of illocution is computer-friendly.
We can use linguistic theories to define and
generate implicit archetypal questions and AI can
help us estimate how pertinently they are
answered.
• Our tech: answer retrieval algorithms which can
extract information out of textbooks or
documentation.
• We use ChatGPT to refine the retrieved
information, improving readability, coherence,
and cohesion.
25
49. Discussion
• ChatGPT could be fed with the whole
explanation context necessary for
generating appropriate responses.
• But without a filtering mechanism we
would hit ChatGPT's memory constraints.
• ChatGPT cannot process more than a few
pages.
• Our YAI doesn’t have this problem. We use
answer retrieval to generate accurate and
context-specific outputs.
49
50. Discussion
• ChatGPT could be fed with the whole
explanation context necessary for
generating appropriate responses.
• But without a filtering mechanism we
would hit ChatGPT's memory constraints.
• ChatGPT cannot process more than a few
pages.
• Our YAI doesn’t have this problem. We use
answer retrieval to generate accurate and
context-specific outputs.
50
51. Discussion
• ChatGPT could be fed with the whole
explanation context necessary for
generating appropriate responses.
• But without a filtering mechanism we
would hit ChatGPT's memory constraints.
• ChatGPT cannot process more than a few
pages.
• Our YAI doesn’t have this problem. We use
answer retrieval to generate accurate and
context-specific outputs.
51
52. Discussion
• ChatGPT could be fed with the whole
explanation context necessary for
generating appropriate responses.
• But without a filtering mechanism we
would hit ChatGPT's memory constraints.
• ChatGPT cannot process more than a few
pages.
• Our YAI doesn’t have this problem. We use
answer retrieval to generate accurate and
context-specific outputs.
52
53. More Discussion
• The answer retrieval mechanism is
used for macro-planning: the
selection of information.
• ChatGPT is then used for surface
realisation.
• Thus, ExplanatoryGPT can provide
specific and context-aware
explanations, avoiding broad or
incorrect responses.
53
54. More Discussion
• The answer retrieval mechanism is
used for macro-planning: the
selection of information.
• ChatGPT is then used for surface
realisation.
• Thus, ExplanatoryGPT can provide
specific and context-aware
explanations, avoiding broad or
incorrect responses.
54
55. More Discussion
• The answer retrieval mechanism is
used for macro-planning: the
selection of information.
• ChatGPT is then used for surface
realisation.
• Thus, ExplanatoryGPT can provide
specific and context-aware
explanations, avoiding broad or
incorrect responses.
55
56. Conclusions
• Answer retrieval and ChatGPT complement each other.
• This methodology has potential for improving
explanations in intelligent textbooks and
documentation.
• Future research:
– experiments to verify whether ExplanatoryGPT
helps enhancing educational materials and
software documentation;
– experiments to verify whether ExplanatoryGPT is
better than ChatGPT.
56
57. Conclusions
• Answer retrieval and ChatGPT complement each other.
• This methodology has potential for improving
explanations in intelligent textbooks and
documentation.
• Future research:
– experiments to verify whether ExplanatoryGPT
helps enhancing educational materials and
software documentation;
– experiments to verify whether ExplanatoryGPT is
better than ChatGPT.
57
58. Conclusions
• Answer retrieval and ChatGPT complement each other.
• This methodology has potential for improving
explanations in intelligent textbooks and
documentation.
• Future research:
• experiments to verify whether ExplanatoryGPT
helps enhancing educational materials and software
documentation;
• experiments to verify whether ExplanatoryGPT is
better than ChatGPT.
58
59. Conclusions
• Answer retrieval and ChatGPT complement each other.
• This methodology has potential for improving
explanations in intelligent textbooks and
documentation.
• Future research:
• experiments to verify whether ExplanatoryGPT
helps enhancing educational materials and software
documentation;
• experiments to verify whether ExplanatoryGPT is
better than ChatGPT.
59