Disclaimer: all claims in this article can be considered incorrect until verified by you—the reader. I just wanted to write something after having read this because I had nothing better to do.
Has any one else noticed that most claims of "AGI in T minus 10 years" are generally made, not by ML/DL engineers or researchers, but by Elon Musk and the philosophers.
Also, notice there is nothing such as an AI engineer. This is just a bogus term used by people of the same ilk as the fungible folk fomenting non fungible .
Years from now ML/DL folk are gonna be miffed by the fact that bogus people and charlatans have claimed the AI [suffix] titles.
This is not unlike—much to the dismay of Astronomers—bogus people and charlatans having claimed the term Astrology. A word that is supposed to mean study of the stars. Although, I do feel that once intergalactic space travel becomes the norm, bogus people and charlatans will cede the term Astrology. But then again, knowing bogus people and charlatans, they'll probably just claim that intergalactic space travel is fake. This is not unlike flat-earthers, who are either really good trolls or bafflingly stupid people.
Getting back to the topic, this is not meant to be a dig at Nick Bostrom or something. I did enjoy Superintelligence, and I do the paper-clip-maximization-bot. But it does feel like most of these claims are just philosophers doing what they do best—exhausting all possible thoughts and ideas in some concept space. Also, the paper-clip-maximization-bot is a nonsense thought experiment, it's like saying what if you write a while loop but don't code in an exit condition.
Until Andrej Karpathy or someone of the Andrej Karpathy type claims that AGI is ten years away we can all worry about aliens instead. But when they do, it would behoove all of us to start the construction of an .
Imagine the year is 2030 and Elon Musk the philosophers have run out of AI related things to worry about, then what? Let me tell you what, AI winter is what. "Ten years ago you said AGI will be here in ten years, now you're telling me it's gonna take another ten years?", "It's 2030, why isn't Jarvis driving me around?", "Self-driving cars? More like ."
Again I'm not taking a dig at Elon Musk and the philosophers or something, in my free time I too like to think about the various ways an AGI would screw us over. But after having delved a bit into the engineering and research behind ML/DL, it feels like what Elon Musk and the philosophers fear is atleast one paradigm shift away.
"Well why can't paradigm shifts take place in ten years?"
In 2012 the paradigm shift that allowed Neural Nets to decimate
competition in the ImageNet challenge was not the software architecture.
It was the
which allowed them to scale up NN architecture. Between LeNet (1988) and
AlexNet (2012) there is no architectural paradigm shift. Heck the
perceptron (precursor to the NN neurons) was invented in 1958.
"You haven't answered the question..."
Essentially
there are limits to how many transistors one can fit in a limited area on
a flat sheet of rock. So probably what needs to happen is someone tries to
facilitate computation in three dimensions, in gray matter—not unlike the
brain. This seems like a reasonable paradigm shift.
"And why can't that take place in ten years?"
Because we don't understand the brain? Shit... That too can take
place in ten years, Elon Musk and the philosophers were right all along, I
should probably start building my air-gapped bunker!
Since I'm a broke developer—which may seem like an oxymoron, but not in India—I have decided to set up a patreon to fund my air-gapped bunker. Please click here to spare some monthly change.