MIT is in the process of converting its brand's value into money. It's interesting, because the brand is so strong that they probably can announce even more stupid "quests" and "initiatives" and "labs" (like the IBM "collaboration") to extract value from industry without seriously harming the quality of the students who apply. But as a current student it's kind of heartbreaking. This place has been so special for so long. But even MIT cannot hold out against the corporate cancer that's been spreading through higher education for the last twenty years.<p>EDIT: just wanted to add, I'd love to ask the marketing team behind this site how they came up with <i>quest</i>. What the fuck is a quest, and how is it different than all of these researchers just doing whatever they wanted, anyway? MIT isn't funding these labs directly in most cases and the work has been ongoing, it's just a branding exercise.
This is the high water mark, the summer solstice if you will, of the current wave.<p>Buckle up, Serious People are going to rediscover the fundamental Hard Problems and relocate the current Hot Topics into their appropriate ontologies.
MIT is also giving the course Artificial General Intelligence at <a href="https://agi.mit.edu/" rel="nofollow">https://agi.mit.edu/</a>
List of recommended articles and papers (reading material for the AGI course): <a href="https://agi.mit.edu/vote-ai/" rel="nofollow">https://agi.mit.edu/vote-ai/</a>
Sounds like they are solving the exodus of brains to Industry by effectively creating Industry-funded departments located on campus. “You will be an MIT employee, fully funded by Google.”
> And today, by tapping the united strength of these and other interlocking fields and capitalizing on what they can teach each other, we seek to answer the deepest questions about intelligence — and to deliver transformative new gifts for humankind.<p>Ugh, this kind of corporate-speak is nauseating. Can anyone understand what this "quest" actually entails?
Understanding the recipie and salient ingredients by which nature generates intelligence so that we can build a theory of intelligence is long overdue but they’re starting at too high of a level. We don’t have a good model for artificial neurogenesis that allows us to create complex AI from a simpler set of building blocks, we don’t have a genotype-to-phenotype mapping or a generic representation to encode complex phenotypes in a mathematical genetic abstraction, we don’t have an abstraction by which mutation can create open ended phenotypic variation, we don’t have a model for artificial evolution to drive the evolution of novelty.<p>If we want to solve this problem we’re going to have to reverse engineer intelligence. Otherwise we’re just going to continue to run into walls by trying to either brute force our way from the ground up and by ignore lessons from biological intelligence or philosophize from the top down.
I don't think connected neural ensembles made from a deep learning architecture can scale to what we would call general artificial intelligence.<p>At least not with decades of manual supervision.
I respect what they are doing but I disagree with the methods. Intelligence could be one of those kind of things that is very hard to engineer directly. And we'd have more success if we inspected the underlying processes and then simulated those.<p>I've posted this before but here is my proposal: <a href="https://scrollto.com/life-a-universe-simulation/" rel="nofollow">https://scrollto.com/life-a-universe-simulation/</a><p>What I propose is a minimum viable digital environment that can support the creation of self-organized turing machines that feed off their environment. What this really means is, coming up with a digital environment that can support the evolutionary process. Evolution requires vast space, vast time (in this case -- clock cycles), and principles that allow for both storage and movement of information. The storage and movement of information is accomplished most simply by roughly emulating mass/energy conservation/conversion laws that we have in our universe. With just collisions that form stationary quasiparticles, and can also annihilate to reform the moving fundamental particles, universal computation is enabled.<p>Toffoli and Fredkin discovered the power of collision-based computing decades ago. There is a lot of literature and good results they derived on the power of these types of systems.<p>Let's create life the only way we know it formed -- evolution. It's far more elegant and less engineered than trying to unravel how chaos formed competitive results on a million-deep evolutionary ancestor tree.