[AI-interop] Thoughts following ADASS
Tim Jenness
tim.jenness at noirlab.edu
Fri Nov 6 06:43:03 CET 2026
I’ve just attended ADASS and I’m currently at the IVOA meeting. I was
chatting with Marco about the role of the IVOA in the AI future and I have
some thoughts.
At ADASS there were at least 3 separate presentations on MCP servers that
provide a way for AIs to ask about VO resources. Kelle Cruz set up a Slack
channel on her ORCA Slack that seems to list even more. It’s clearly not
worth the IVOA getting involved in MCP server interface specifications but
I do think there is a path forward for some kind of IVOA Note about the
topic.
Specifically, we should look at each MCP server and see what their pain
points are and whether there is anything that can make the implementation
of these servers more straightforward. I know that the NOIRLab MANNA server
is having to work around bad TAP implementations, but those clearly need
fixing regardless of whether an AI is involved or not.
The other side of this is thinking how we can make it easier for an AI
agent to interpret the results. In many cases we do not provide enough
information in our TAP services for agents to be able to understand whether
or how flux measurements from one instrument can be compared with another.
It seems to me that we need to make sure that we have the ability for TAP
services to be able to link their results from filters to filter profiles
and also to give enough citation information (DataOrigin per column?) to
determine which algorithm was used for a set of columns. Obviously TAP
results with DataOrigin information are more useful to an AI than those
without since they can download relevant description papers. It’s always
been true that the VO services improve the more semantic linkage there is
but now AIs can really make use of those linkages to improve how they use
the results they find.
Thoughts?
—
Tim Jenness
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ivoa.net/pipermail/ai-interop/attachments/20261105/ab772aca/attachment.htm>
More information about the AI-interop
mailing list