"Humans are prone to giving machines ambiguous or mistaken instructions, and we want them to do what we mean, not what we say. To solve this problem we must find ways to align AI with human preferences, goals & values."
- @MelMitchell1 at @QuantaMagazine: quantamagazine.org/what-does-it-m…
“All that is needed to assure catastrophe is a highly competent machine combined with humans who have an imperfect ability to specify human preferences completely and correctly.”
"It’s a familiar trope in #ScienceFiction — humanity threatened by out-of-control machines who have misinterpreted human desires. Now a not-insubstantial segment of the #AI research community is concerned about this kind of scenario playing out in real life."
- @MelMitchell1
"People are often irrational and behave in ways that contradict their values, and values can change over individual lifetimes and generations...it’s not clear whose values we should have machines try to learn."
- @MelMitchell1
on #AIAlignment for @QuantaMagazine
"An essential first step toward teaching machines ethical concepts is to enable machines to grasp humanlike concepts in the first place, which I have argued is still #AI’s most important open problem."
- SFI Prof @MelMitchell1 on "the barrier of #meaning": melaniemitchell.me/PapersContent/…
"Without a better understanding of what #intelligence is and how separable it is from other aspects of our lives, we cannot even define the problem, much less find a solution. [It] won’t be easy; it will require us to develop a broad, scientifically based theory of intelligence."
• • •
Missing some Tweet in this thread? You can try to
force a refresh
"One way to represent the kind of #compositionality we want to do is with this kind of breakdown...eventually a kind of representation of a sentence. On the other hand, vector space models of #meaning or set-theoretical models put into a space have been very successful..."
Today's SFI Seminar by Ext Prof @ricard_sole, streaming now — follow this 🧵 for highlights:
"Why #brains? Brains are very costly...it seems like they are not a very good idea to bring complex cognition to a #biosphere that just needs simple replicators."
"I also want to explore the problem of #consciousness, which is around all the time..."
"We build the geometry directly by thinking about the PATH that our 3D printer takes. There's no intermediate slicing software [to render CAD as "2D" layers]."
"You can start to think about surface textures - like spikes that you can't do with a traditional slicer. Or...here's a path that's a sine wave, but every other layer is rotated. Or...one creature is following THIS path, and the other is chasing it around."
We start with a talk by SFI President David Krakauer:
"Would anyone care to guess why we're so GOOD at building transistors and so CRAP at designing drugs?"
"This thing [points to transistor] lives in a centralized system. This thing [points to cancer drug] lives in US."
"I'm going to pick on economics, because we like to do that at SFI. 'Ooh, look at that cover! So techy. Global, heat maps...' But here's 'Networks' [in the textbook]. THAT'S IT. Here's '#ComplexityEconomics.' NOTHING."