In the red corner, weighing in using US standard measurements, the challenger: NYU’s Gary “Down with Deep Learning” Marcus. And in the Blue Corner, weighing in using le système métrique, Facebook’s Yann “Show Your Work” LeCun. This fight will decide the fate of AI research once and for all.
Since we have DL but don't yet have AGI, it's quite obvious to everyone that we are missing something.
I repeat: the interesting question is precisely what.
I repeat my claim: DL (defined as gradient-based learning of non-linear functions) will be part of the solution.
— Yann LeCun (@ylecun) November 26, 2018
For those unfamiliar with the term: AGI is human-level intelligence in a machine. The debate centers around whether one specific AI development paradigm is necessarily the path to AGI. It’s a lot more complex than that, but the gist of it is: LeCun feels strongly that deep learning will be a part of the solution. And, of course, Marcus (sort of) disagrees.
Show me the deep net that can invent math and astronomy and I will eat my proverbial hat, or make a large donation to your favorite charity.
— Gary Marcus (@GaryMarcus) November 24, 2018
It might not seem like a big deal, especially if you measure by likes and retweets, but sometimes history isn’t sexy to the mainstream when it happens.
The primary agitators in this debate, LeCun and Marcus, may not be the kind of social media influencers you’re used to, but in the world of machine learning they’re as influential as they come.
And, when titans of AI clash, there’s always going to be some drama. This one even has an extra layer: it’s a classic big tech versus academia match up.
In his capacity as Facebook’s supreme AI hive-mind ruler (not his official title) LeCun oversees the development of cutting-edge machine learning advances. At least he used to, we hear he stepped back from his role in day-to-day operations to work on taking Sophia the Robot down a peg or two full time.
And Gary Marcus teaches psychology and neural science at NYU, which makes him an expert on intelligence. He wrote a scathing criticism of deep learning. It was not called “Deep Learning: What have you done for me lately?” The subtitle was not “ImageNET: yawn.”
It’d be a lot cooler if it was. You can read it, with its boring title “Deep Learning: A Critical Appraisal,” here. If you want to see some of LeCun’s work … actually, I dare you to try and not see some of LeCun’s work. My name is Facebook: for we are many.
It wasn’t just LeCun and Marcus trading tweets during the Great AI Debate of November, though. Like any good show there were plenty of celebrity cameos.
It'll speak with a lisp.. Or at least that's a training scheme to do so.
— Jeff Dean (@JeffDean) November 21, 2018
Google’s Jeff Dean showed up. He helped create the Google Brain program. And we’re all on the edge of our seats waiting for Google to add a heart and some courage (sorry Jeff, couldn’t resist).
And Jack Clark made an appearance as well, if only because Marcus summoned him. He’s the policy director at OpenAI, but more importantly he puts out the Import AI newsletter. There’s no joke here — you should subscribe to it if you’re an AI enthusiast.
The debate also brought tweets from Judea Pearl (one for you to Google), and Thomas Dietterich – but I don’t think it’s an official debate until he shows up, so that wasn’t unexpected.
I'm defensive? I don't need to be. But why are you so offensive?
CherryPi (the FAIR StarCraft bot) is largely rule based too, with a bit of ML/RL. It's also open source. The Samsung bot is not "still better" than CherryPi since it didn't even exist until now.
— Yann LeCun (@ylecun) November 27, 2018
I don’t think Starcraft is considered operational code for Facebook.
— Thomas G. Dietterich (@tdietterich) November 27, 2018
This discussion seemingly sprang up out of nowhere, but I suspect all the AGI talk of late is related to Martin Ford’s new book “Architects of Intelligence.” In it, he interviews 25 of the world’s leading AI experts, including LeCun, Marcus and Dean.
what’s better than a twitter debate on AI, causality & deep learning? a new book w in-depth interviews with @ylecun, @yudapearl, me, @drfeifei , Geoff Hinton, @demishassabis, @rodneyabrooks, @etzioni, @DaphneKoller, @JeffDean, Yoshua Bengio & many more https://t.co/cnhEuO7VWt
— Gary Marcus (@GaryMarcus) November 27, 2018
TNW spoke with Ford about the book. He told us his biggest takeaway from the experience of writing it was seeing how differently the experts perceived the field of AI. And you can certainly see what he means on Twitter.
And, if you can credit Ford with stoking the conversation, then Marcus and LeCun deserve mad respect for opening the doors on a fascinating debate.
Obviously the path forward wasn’t figured out in a week’s debate on Twitter, but the conversation alone is — hyperbole alert — a matter of historical record that could find a natural home in a digital museum one day.
Say what you will about social media, but it’s nothing short of amazing that anyone with an internet connection can watch these brilliant minds discuss the future of AGI. It’s like watching physicists discuss the Manhattan Project in its early conception.