"What's in a name? From Norbert Wiener's Teleology to Artificial Intelligence" For those interested in my teleology-cybernetics-S/AI explorations, I have a talk upcoming at SCiOs UK virtual meeting 16 Nov systemspractice.org/events/s…

A recurring pattern in science: eminent scientists treating a consequence as so implausible that nature must somehow forbid it. Eddington: stellar collapse? No. Rutherford: atomic energy? “Moonshine.” Newcomb: powered flight? No. Pauling: quasicrystals? “No such thing.”

One major result from @OpenAI's release that's been a bit slept on: *No algorithm can solve Hilbert’s Tenth Problem over Q* (004). I think it exhibits remarkable out-of-the-box thinking. A major *human* approach for almost 80 years, growing out of Tarski's questions to Raphael Robinson & building on the work of Julia Robinson '49, Bjorn Poonen '03-09, and others, was to first *define the integers inside the rationals using existential conditions*. This proof sidesteps that obstacle entirely — it gets undecidability without ever needing such a definition! As I currently understand it, it sets up a sequence of finite rational tests that, if they all pass, produce an “integer-like” solution in a nonstandard setting. Elliptic-curve/size constraints then show that an ordinary integer solution must exist, which in turn, is enough to transfer undecidability to Q from Z (where we already knew HTP to be undecidable by Davis-Putnam-Robinson-Matiyasevich '70). github.com/openai/math/blob/…

I'm speaking about getting alignment from misaligned systems today at RESI/MIT today at 3pm in Stata (G449) and tomorrow at the Harvard CS Colloquium events.seas.harvard.edu/even… If you can't make it, I gave ~ 2/3 of this talk at Cornell last week: youtu.be/WUHF_dWdkww

Perhaps matrix multiplication is an example of (Manuel) Blum’s Speed-up Theorem.

Many staggering results here. But this is a particularly amazing one: the exponent for matrix multiplication is no more than 2.25. The previous world record had been something like 2.37. This leap in progress is like Bob Beamon's long jump. github.com/openai/math/blob/…

The recursive loop of history continues. Almost a century ago, Warren McCulloch helped Eilhard von Domarus complete and translate his doctoral thesis, "The Logical Structure of Mind". Once he found the math (i.e. Walter Pitts) a decade later he published the first artificial neural network model, "A Logical Calculus of the Ideas Immanent in Nervous Activity". From this, along with a paper by Rosenblueth, Wiener and Bigelow, "Behavior, Purpose and Teleology" and a book by Craik, "Nature of Explanation" Cybernetics (nee Teleology) was born, and within a decade more would give rise to cognitive science, AI, systems science and more. Last night I had the distinct pleasure to host two of McCulloch's students at our latest scientific salon, along with other luminaries in cybernetics and semiotics and some of my own students. Today, we'll have a reading group discussion of von Domarus' thesis in the Phenomenology Center at Duquesne. History continues recursively.