five years from now, the term "artificial intelligence" will refer to an entirely different technology
we don't know what it is, but it will be something mostly unrelated to the stuff that has that name today
we say this with confidence because that has consistently been the case every few years since the term was coined in 1956
as soon as everyone understands the latest fad, it feels obvious that it has nothing to do with "intelligence", and we all stop calling it that
we sometimes feel like we're drawing an unreasonably hard line by saying "machine learning" (which is to say, the field that deals with statistical techniques done by computer) or "differentiable neural networks" or "large language models" or whatever other specific thing we actually mean
it often involves pushing back on our friends who find it obvious what "AI" refers to, and think being specific about it is just pointless obscurity
but like. it feels obvious today because we are all immersed in marketing material that pushes this one specific meaning
the moment the money dries up, the marketing will too, and when we all look back at stuff written during this period, it won't be at all clear what people were trying to say
we've been following this field since early childhood in the 1980s. yes, we were very avid readers of technical documentation at that age. yes, kids these days don't have the same opportunities to learn how this stuff actually works, and that makes us very sad.
anyway, that's by way of saying we've seen a lot of shifts in computing over the years, and we're speaking from experience about this
@ireneista Working in/for a Computer Science (university) department has been a useful thing for me for this nuanced view of 'AI'. While there''s an overall 'AI' supergroup in the department, it has a bunch of named sub-groups called various things (local names and abbreviations include CL, KR, ML, and Vision). When we routinely have to ask 'okay what *sort* of AI is that new professor associated with?', it sticks in one's mind.
(And they're not all one big happy resource-sharing family.)
@ireneista Most frustratingly nearly nothing gets done with it in the long run, people just loose interest, another AI winter, and eventually the whole hype restarts.
Eg: What did we do with https://en.wikipedia.org/wiki/SHRDLU …?
Siri/Alexa/etc can't even do that!
@ireneista eg: Imagine a further developed command line interface which is 'aware' of files, folders, applications… and can do simple tasks like moving, renaming, etc?
Never happened.
@DLC right, exactly!
well, we're going to build it at some point, we decided. we took a strong interest in that back in the 2000s, then lost interest because, well, all the Irenes who were alive at that time died, and we-today had to figure out how to live in the ruins of their life and so on... but anyway we've convinced ourselves that it should actually be built
@ireneista We're working on an (actual) smart home design based on the idea of agents which write/read to a shared whiteboard?
Silly eg: The camera-agent dumps a link to a raw video stream on the whiteboard
A cat finder-agent looks at any video source & marks the presence of cats in the video
The cat locator-agent looks at any marked cat footage tries to place the cat in the space in or around the house & drops it's location on whiteboard
The cat Identifier tries to name the cat
etc…
@DLC @ireneista this is so scarily close to how Palantir works ;)
in other words, you have converged onto an architecture that works
heh, i'm interested in knowing more; there's this humongous read about their history and where they floundered, i have to find the link -- perhaps you know which i'm talking about? it's that memorable ;)
meanwhile, there's some talk coming out of IBM about neuro-symbolics, which presumably is the next step -- maybe the discrete-representations movement is how the two fields will be merged
@ireneista
I think - as someone of about the same generation, who's followed the AI field for many years - it's worth remembering that every time we had an "AI winter" we also did get a whole set of increasingly powerful methods and algorithms out of it. And they continued to be useful and used, just without the "AI" label slapped onto them.
Robust high-dimensional fuzzy matching is powerful and useful, with or without the "AI" polish on top.
Irenes (many)
Petra
Chris Siebenmann
Dimly Lit Corners
الجبر خوارزمی
l'empathie mécanique
Rho
Janne Moren