Note
If you haven't set up a horizon scanning process yet, check out this horizon scan process setup guide first
Note
If you haven't set up a horizon scanning process yet, check out this horizon scan process setup guide first
Trying to find a balance between building a longterm scan hit library that's usable + useful VS project-specific requirements.
Useable + Useful
You want your tagging to be focused enough to be useful for a pointy search - eg. I’m trying to find the examples I had of “ attribution protocols” for a paper on how web3 is changing provenance structures or social graphs.
Comprehensive Rollups
Broad enough that I can grab all the scan hits relating to “Education”
Fit for Purpose
Comprehensive + Flexible enough that I can work with / review / present all the scan hits relating to say “Art Futures” (a project) , but still retain a master scan library with data integrity.
Trying to find a balance between building a longterm scan hit library that's usable + useful VS project-specific requirements.
Useable + Useful
You want your tagging to be focused enough to be useful for a pointy search - eg. I’m trying to find the examples I had of “ attribution protocols” for a paper on how web3 is changing provenance structures or social graphs.
Comprehensive Rollups
Broad enough that I can grab all the scan hits relating to “Education”
Fit for Purpose
Comprehensive + Flexible enough that I can work with / review / present all the scan hits relating to say “Art Futures” (a project) , but still retain a master scan library with data integrity.
Above is an example of a basic tagging taxonomy. When you start wanting to run more complex AI workflows, having a clear structure helps enormously.
Above is an example of a basic tagging taxonomy. When you start wanting to run more complex AI workflows, having a clear structure helps enormously.
You can see in the example above, that
tags can belong to one or more categories
projects can select the categories that are relevant + any additional outlier tags
scan sprints can do the same
You can see in the example above, that
tags can belong to one or more categories
projects can select the categories that are relevant + any additional outlier tags
scan sprints can do the same
Note
The below workflow seems like a lot when you look at it all at once, but once you set it up (and build on it slowly over time), it does enable you to generate quick starting synthesis very quickly.
Note
The below workflow seems like a lot when you look at it all at once, but once you set it up (and build on it slowly over time), it does enable you to generate quick starting synthesis very quickly.
Start Small. Create a list of tags and use a simple AI prompt to generate "Tag Headlines" for each tag. This helps later on when you're asking AI to auto-suggest which tags are relevant to a scan hit.
A simple prompt like:
Provide the most common definition of the for those records where this field is currently empty. Pay particular attention to crafting a definition that will be helpful in defining this tag for application of futures and foresight signals of change.
Start Small. Create a list of tags and use a simple AI prompt to generate "Tag Headlines" for each tag. This helps later on when you're asking AI to auto-suggest which tags are relevant to a scan hit.
A simple prompt like:
Provide the most common definition of the for those records where this field is currently empty. Pay particular attention to crafting a definition that will be helpful in defining this tag for application of futures and foresight signals of change.
Next we link Tags to Categories using an AI Auto-suggest function. This is where the "Tag Headline" that we generated in the previous step provides additional context for the AI agent to suggest the most relevant categories.
You can see in the image above, that we direct the AI agent to reference the [Tag] field and the [Tag Headline] field when suggesting categories.
Next we link Tags to Categories using an AI Auto-suggest function. This is where the "Tag Headline" that we generated in the previous step provides additional context for the AI agent to suggest the most relevant categories.
You can see in the image above, that we direct the AI agent to reference the [Tag] field and the [Tag Headline] field when suggesting categories.
You can link your tags to any other meaningful categorisation or rollup. Here I've taken the sustainable development goals of the public UN website (both the goal and its descriptor), then I've asked the AI to auto-suggest based on the [Tag], [Tag Headline] and [Category] which SDG the tag most likely belongs to.
I've also asked the AI agent to suggest how this tag might impact the SDG. Again this is a context step which provides further reference down the track when we want to ask an AI agent - how this scan hit might impact development (or sustainability) for example.
You can link your tags to any other meaningful categorisation or rollup. Here I've taken the sustainable development goals of the public UN website (both the goal and its descriptor), then I've asked the AI to auto-suggest based on the [Tag], [Tag Headline] and [Category] which SDG the tag most likely belongs to.
I've also asked the AI agent to suggest how this tag might impact the SDG. Again this is a context step which provides further reference down the track when we want to ask an AI agent - how this scan hit might impact development (or sustainability) for example.
Note
Note
I've also down the same thing with Signature Solutions + Enablers - breaking them down to the subtopics listed on the public UN website - which I've called [SS tags]and then asking AI to auto suggest which [SS tags] match my scan hit.
So rather than connecting a Signature Solution to the scan hit, the AI suggest operates at a much more granular level by reading the scan hit data and then matching to the [SS Tags]. This took a couple of hours to set up but once it was done, it made it really easy to identify potential connections.
I've also down the same thing with Signature Solutions + Enablers - breaking them down to the subtopics listed on the public UN website - which I've called [SS tags]and then asking AI to auto suggest which [SS tags] match my scan hit.
So rather than connecting a Signature Solution to the scan hit, the AI suggest operates at a much more granular level by reading the scan hit data and then matching to the [SS Tags]. This took a couple of hours to set up but once it was done, it made it really easy to identify potential connections.
Note
We're already getting into a multi-agentic AI workflow here so there is obviously the risk that errors or miscategorisation is being carried further down the line - the important thing is to check, review and revise (especially in the first stages) but at the very least, it offers some useful prompts rather than starting each tagging session anew.
Note
We're already getting into a multi-agentic AI workflow here so there is obviously the risk that errors or miscategorisation is being carried further down the line - the important thing is to check, review and revise (especially in the first stages) but at the very least, it offers some useful prompts rather than starting each tagging session anew.
What did you think of this content?
It was interesting
It was not interesting
I have feedback
What did you think of this content?
Helpful
Not helpful
Feedback
Last updated on
Dec
4,
2024
Last updated on
Dec
4,
2024