Beautifully written article Beth. The joy of the expressive narrative is matched by the thoughtful illumination of the things the AI hype is missing. And missing it is not just losing the thread it's failing to reveal the depth and novel interconnections that make the effort worthwhile and create value for those who are spending their valuable time on it.
Thanks sharing this. I'm finding too frequently that the intellectual effort you're describing which grounds the knowledge in context and provenance is just missing. And sadly unnoticed.
Yes, lets. I think what's your doing is becoming increasingly in the age of agenticAI with agents representing their carbon-based twin in autonomous decision-making supposedly for the benefit of their DNA coded partner ;-)
Are we allowed to talk about this out loud now? I will never forget that you recommended the Master and his Emmisary. I am writing a new piece on thinking - Hannah Arendt informed - stay tuned.
Thank you Chris for this - I already pinged Thanh the git repo so we can play, what a gift to start off our new year with. We will see if we can run some experiments—my lazy thoughts for a Saturday.
- Using CurateGPT’s entity and relationship extraction as an upper or T-shaped ontology layer feels like a good A/B comparison (wondered why they didn't do this in the paper) for patient and caregiver materials.
- My next curiosity is agentic control — MCP / ACP style chains where curators explicitly run extract → ground → evidence → approve, instead of inheriting whatever the model felt like doing that day.
I think the reason they don't go down the path of establishing an upper level ontology in the paper is that their work operates in a richer, more fully developed context where foundational ontologies are already well established. They are not building new ontologies, but adding to existing ones. They do so using established infrastructure in the OBO Foundry. https://obofoundry.org/
Other work mentioned in the paper gives deatil on the foundations of this work. Specifically the papers on DRAGON-AI and SPIRES. (I haven't read either.)
Beautifully written article Beth. The joy of the expressive narrative is matched by the thoughtful illumination of the things the AI hype is missing. And missing it is not just losing the thread it's failing to reveal the depth and novel interconnections that make the effort worthwhile and create value for those who are spending their valuable time on it.
Thanks sharing this. I'm finding too frequently that the intellectual effort you're describing which grounds the knowledge in context and provenance is just missing. And sadly unnoticed.
But you noticed and that makes me very happy. Been a minute P1, let’s catch up soon please.
Yes, lets. I think what's your doing is becoming increasingly in the age of agenticAI with agents representing their carbon-based twin in autonomous decision-making supposedly for the benefit of their DNA coded partner ;-)
Are we allowed to talk about this out loud now? I will never forget that you recommended the Master and his Emmisary. I am writing a new piece on thinking - Hannah Arendt informed - stay tuned.
Interesting, though yet unfinished read.
You might be interested in what Chris Mungall’s group of curators is doing with GPT.
https://arxiv.org/pdf/2411.00046
Thank you Chris for this - I already pinged Thanh the git repo so we can play, what a gift to start off our new year with. We will see if we can run some experiments—my lazy thoughts for a Saturday.
- Using CurateGPT’s entity and relationship extraction as an upper or T-shaped ontology layer feels like a good A/B comparison (wondered why they didn't do this in the paper) for patient and caregiver materials.
- My next curiosity is agentic control — MCP / ACP style chains where curators explicitly run extract → ground → evidence → approve, instead of inheriting whatever the model felt like doing that day.
I think the reason they don't go down the path of establishing an upper level ontology in the paper is that their work operates in a richer, more fully developed context where foundational ontologies are already well established. They are not building new ontologies, but adding to existing ones. They do so using established infrastructure in the OBO Foundry. https://obofoundry.org/
Other work mentioned in the paper gives deatil on the foundations of this work. Specifically the papers on DRAGON-AI and SPIRES. (I haven't read either.)
https://pmc.ncbi.nlm.nih.gov/articles/PMC11484368/
https://pmc.ncbi.nlm.nih.gov/articles/PMC10924283/