In 2007 a question was broadcast from the the Singularity University to the extant artificial intelligensia: when will AI surpass human intelligence? This was my off-the-cuff response.
Everyone’s first take on strong AI is usually "when will a computer first pass the Turing test?" My answer, along with most of the crowd, was 2030 to 2050. Lately, I've pulled that in to about 2025. (Google is not going to let Apple get there first, and similarly for Microsoft and IBM — not to mention DoD, USA versus China.)Surely by then we'll have intelligent assistance programs that can fake enough discourse, common sense, and knowledge of current events to slip by the judges. But, this is just penny ante stuff — parlor games.
More interesting to me is the surpass part. So, just to stir up the natives, here's a little story for you. It's about RALPH, a stealth mode program in pre-alpha, designed at INTEL strictly for internal corporate use in the year 2040.
RALPH's purpose is to autonomously design INTEL's next generation chip. That chip will leap ahead of Moore's Law, replace RALPH's own cpu — without a reboot — and crush once-and-for-all that nettlesome Chinese competitor Taiwan-Semi.
RALPH thoroughly understands the detailed features, limitations, and manufacture of Intel's current chip, the Dodecium40, since its precursor helped design it. However, now RALPH is on its own. Unaided, it combs the world's literature researching design candidates and manufacturability.
Here are a few picoseconds from its logfile.
0357 GMT — trying to crack loss of quantum coherence reported by previous Dodecium41 design team that limited number of qubits.
0357GMT literature search> Balkan EECS literature> Bulgarian Journal of Quantum Computation (machine edition)
0357GMT review precursor literature
0357GMT possible solution (p=.0013) (confirmatory evidence needed)
(RALPH is onto something, but will the theoretical insight actually cut it when the nanotubes hit the road?)
0358GMT Begin feasibility test> 1)Intel McMurdo Sound AutoFab (comment: easy cryo) in Antarctica.
0358GMT Begin feasibility test> 2) Intel GEO AutoFab (comment: no funky gravity probs in geostationary orbit)
Several microseconds later, based on preliminary results, RALPH decides to go ahead with a full bore design effort for a new Dodecium41 chip, whose lynch pin is a new ground-breaking reformulation of quantum coherence. Put that in your moo shu, Taiwan-Semi!
logfile continues ...
0359GMT subproblem> financial> debt; equity; vc; theft; new currency ... (several hundred novel combos)
0340GMT subproblem> financial> solution set...
(Comment: Among the leading candidates are —
The last financial option requires a detailed simulation by RALPH of its own ingenuity post augmentation by the new chip. (INTEL shareholders have regularly voiced concerns that RALPH is becoming a "loose cannon.")
RALPH has a thorough understanding of its entire hardware and software suite. For years its precursors have been able to autodiagnose and fix all glitches in real-time. But predicting its own smarts post-upgrade (and making a big bet on it) — that's a toughie. Its decision takes a full 3 nanoseconds during which 83% of its Bose-Einstein condensate heat sinks are sucking away infrared at full bore.
Comment to reader — you get the idea: we're talking real super AI — not nickel and dime Turing test stuff — imitating mall rats for brain-dead primate entertainment.
Mandatory Copy Law of 2037 (copy this or else (Ralph will freeze and upload your brain — with a heavy dose of antivirus to protect itself.) Author, an unemployed former poet, is a reverse time-traveled, Ralph simulacrum/avatar.)