He’s the PT Barnum of tech, the clown prince of wannabe edgelords, and the laughingstock of the synthetic intelligence group.
Elon Musk not too long ago went on the document with what is perhaps his most ridiculous prediction but:
2029 appears like a pivotal 12 months. I’d be stunned if we don’t have AGI by then. Hopefully, folks on Mars too.
— Elon Musk (@elonmusk) May 30, 2022
Greetings humanoids
Subscribe now for a weekly recap of our favourite AI tales
“2029 appears like a pivotal 12 months. I’d be stunned if we don’t have AGI by then. Hopefully, folks on Mars too.”
To the layperson unfamiliar with laptop science, this in all probability simply looks as if a extremely sensible particular person being optimistic. However to most individuals who perceive the fundamentals of AI, that is the form of assertion you’d anticipate from a mainstream journalist with a restricted understanding of the subject.
I’m undecided there’s a single severe particular person working within the discipline of AI (except for traders) who’d co-sign Musk’s perception — no offense meant to the summary dreamers on the market.
The truth is, a coalition of consultants has been shaped with the only real intent of asking the world’s richest blusterer to place his cash the place his mouth is.
How a lot are you prepared to guess? https://t.co/9pxVgyMMqm
— Gary Marcus 🇺🇦 (@GaryMarcus) May 30, 2022
Betting towards the world’s richest particular person
Synthetic intelligence luminary and co-author of the best-selling e book “Rebooting AI,” Gary Marcus, initially laid the problem down.
Marcus, an NYU professor who’s made his bones debating the likes of Yoshua Bengio and Yann LeCun (two people who find themselves absolute rock stars within the discipline of machine studying), first supplied Musk a $100,000 wager.
The phrases, per a Substack post from Marcus, are as follows:
Listed here are 5 predictions, organized from best for many people to those who would require a excessive diploma of experience:
- In 2029, AI will be unable to observe a film and inform you precisely what’s going on (what I known as the comprehension problem in The New Yorker, in 2014). Who’re the characters? What are their conflicts and motivations? and many others.
- In 2029, AI will be unable to learn a novel and reliably reply questions on plot, character, conflicts, motivations, and many others. Key might be going past the literal textual content, as Davis and I clarify in Rebooting AI.
- In 2029, AI will be unable to work as a reliable cook dinner in an arbitrary kitchen (extending Steve Wozniak’s cup of espresso benchmark).
- In 2029, AI will be unable to reliably assemble bug-free code of greater than 10,000 traces from pure language specification or by interactions with a non-expert consumer. [Gluing together code from existing libraries doesn’t count.]
- In 2029, AI will be unable to take arbitrary proofs from the mathematical literature written in pure language and convert them right into a symbolic type appropriate for symbolic verification.
Right here’s what I suggest, should you (or anybody else) manages to beat a minimum of three of them in 2029, you win; if just one or two are damaged, we will’t very nicely say we now have nailed the final half in synthetic common intelligence, can we? In that case, or if none are damaged, I win.
Right here’s the factor people, Marcus (in my view) isn’t attempting to make a buck off Musk and it positive doesn’t appear to be he’s attempting to drum up controversy to promote one thing.
He, like many different AI consultants, seems to be involved with the adverse results Musk’s rhetoric may have on the remainder of the sector.
AI isn’t magic; it’s science. And it’s typically unimaginable to tell most of the people on the precise state of AI applied sciences (my job, right here at TNW’s Neural vertical), when folks like Musk are on the market making claims so outlandish, they fall into fantasy fan-fiction territory.
We’ve been right here earlier than
Keep in mind when Musk declared driverless automobiles have been simply across the nook — that they’d be right here inside two years? Then he stated they’d be right here the subsequent 12 months. And the subsequent. And the one after that.
Right here’s Musk promising driverless automobiles in 2014 (and yearly after that!):
After which there’s Hyperloop. Musk was going to dig tunnels all around the world, fill them with magnetic levitation know-how, and ferry passengers underground in specialised autonomous autos able to taking pictures via the Boring Firm’s tubes sooner than a Boeing Jet.
Sadly, no person bothered to inform Musk that maglev tech, as he envisions it, doesn’t exist. What did we get? A tunnel the place 1-3 folks at a time can get right into a Tesla automobile, pushed by a Tesla worker, that slowly meanders its means via a crowded tube the place automobiles are queued up resulting from site visitors. It’s been called a deathtrap.
The truth is, it looks as if each single considered one of Elon Musk’s endeavors, aside from perhaps PayPal and SpaceX, are seemingly constructed on a bedrock of bullshit.
Neuralink? The corporate Musk stated was going to “treatment autism?” Nevermind that 1000’s of scientists, technologists, and engineers have devoted their complete lives over the previous century to growing mind implants to deal with medical circumstances (and that autistic folks don’t should be “cured,” as a result of autism isn’t a disease).
And that brings us to the top of cow manure-based know-how that’s the Tesla “Optimus” robotic.
Robotics are tremendous simple, wink-wink
Optimus was introduced final 12 months as a Tesla undertaking to create a humanoid robotic able to doing boring and harmful chores people don’t need to.
The above video demonstrates how a lot progress Tesla made on the undertaking earlier than the announcement: precisely zero. What you see in that video is a human carrying spandex. Impressed?
Now, we’re to imagine that the corporate may have a functioning prototype to point out off by September:
Tesla AI Day pushed to Sept 30, as we could have an Optimus prototype working by then
— Elon Musk (@elonmusk) June 3, 2022
I’ll go on the document proper now: there’s completely no means Tesla demonstrates a humanoid robotic able to something greater than MIT-level feats of robotics in September.
That’s not an insult to MIT’s robotics lab. It’s a jab at anybody who thinks constructing robots is straightforward. And it’s a 15-minutes-long belly-laugh within the face of anybody who believes Tesla can exhibit a machine that, in any means, convinces the AI group it’ll be able to safely working autonomously in unmapped house areas by 2029.
However no person offers a crap what I believe. Elon Musk is the richest man alive and I’m only a journalist who ought to learn to code.
I requested Gary Marcus what he thought in regards to the upcoming Tesla AI Day, and whether or not Optimus was going to be prepared for primetime by then. He appeared a bit skeptical too:
*IF* Tesla could make main progress in the direction of general-purpose robots quick, I might be impressed.
There are dozens of extremely tough issues to be solved, starting from the management of bodily movement, to the language understanding in deciding what folks need, to the planning in deciding methods to do no matter is needed safely and effectively.
However I additionally notice that we’re nonetheless ready on Tesla’s autonomous taxi fleets. Common-purpose robots are in some ways much more difficult, as a result of the number of human expertise is so huge.
We’ve all seen the cool movies of dancing robots from Boston Dynamics; constructing really common function humanoid robots at a scale, with all of the AI that may require, could be one thing else once more. I don’t see it occurring this decade.
The total $500K guess
To that finish, a coalition of sanity-speakers has sprung up round Marcus. Right here’s a snippet from an article that was simply printed on Fortune as we speak by Marcus and Vivek Wadhwa, an AI skilled from Carnegie Mellon:
Vivek Wadhwa thought the guess was terrific, honest and provocative, and one thing that would transfer the sector of A.I. ahead. So Wadhwa determined to match Marcus’ wager.
Inside a pair hours there was a flurry on Twitter and Marcus’s substack had near 10,000 views, and shortly different consultants within the discipline additionally supplied their assist for the wager, rising the pool to $500,000. However not a phrase from Musk.
Then author and futurist Kevin Kelly, who co-founded the Lengthy Now Basis, supplied to host the guess on his web site facet by facet with an earlier, and associated guess that Ray Kurzweil made with Mitch Kapor.
That form of cash looks as if rather a lot to a lowly author like myself, however Elon Musk may blow $500K a day for the remainder of his life and nonetheless have a fortune left over.
So what’s the maintain up Musk? It’s just about unimaginable to think about that you simply haven’t heard in regards to the guess. For somebody who spends a big quantity of their time trolling folks on Twitter and trashing the media, you’ve at all times been extremely sheepish in relation to debating (and even disputing) the phrases of anybody who even barely seems to know what they’re speaking about within the discipline of AI.
That, pricey readers, is why I’m declaring Musk the official “CEO” of tech: the Hen Government Officer.
Cluck cluck, Musk followers. Cue the memes.
