Vicarious AI raises another $50M

https://venturebeat.com/2017/07/25/khosla-ventures-leads-50-million-investment-in-vicarious-ai-tech/

52 50 9 years ago

dkural4• 9 years ago

Would be great if they showed some demonstrations / proof of any kind. DeepMind did great in that respect. I understand they're working on a different problem, but it's been a number of years; usually correct ideas don't take that long to show some promising progress.

Barrin92• 9 years ago

I'm honestly surprised how willing investors are to throw money at these ML startups that actually show barely any progress.

qq66• 9 years ago

"Show barely any progress" and "show barely any progress to the public" are very different. I don't have any inside information here, but a lack of public announcements is not proof of anything.

arjo1• 9 years ago

Their 2017 paper: https://www.vicarious.com/img/icml2017-schemas.pdf

eli_gottlieb4• 9 years ago

That's it? That tells me the same thing I saw when I attended a Vicarious researcher's talk at MIT CSAIL: they're basically reinventing stuff the Bayesian generative modelers in cognitive science have been doing for years, throwing some video-game playing at it, and calling it AI instead of cognitive science.

Give me $50mil and I could do a good deal more than that, just because I bother to look stuff up!

fuelfive• 9 years ago

There are a few papers up on the site already and some more coming. As a business, publications aren't really the KPIs we prioritize (vs an academic-style lab like DM or OAI).

dkural4• 9 years ago

Out of curiosity, what's the KPI?

p.s. I don't think papers qualify as a "demo". I'm with you re: not running it like an academic lab and chasing a minimal publishable unit or fashions of the time. Demo = Demonstrably do something with your technology no one could do before, ie beat people at Go in DeepMinds case, in your case you could set up your own benchmark / pass a Turing test for 3 year olds or something.

lumost• 9 years ago

The economics for some of these labs is also pretty different from what you'd normally expect. There are numerous semi-private organizations in the defense space ( such as Boston Dynamics ) and similarly in the pharmaceutical space. Which may only produce commercially viable technology after a decade or more. For many of these entities, the funding dynamics are dependent on the overall scope of the problem along with the quality of the team rather than any direct foreseeable economic outcome.

dkural4• 9 years ago

I understand, I don't mean commercial progress either.

jonmc12• 9 years ago

Interesting the bird analogy again - “airplanes don’t flap their wings” was in a recent article by Jeff Hawkins describing Numenta's approach (http://spectrum.ieee.org/computing/software/what-intelligent...). Vicarious co-founder Dileep George was also a co-founder at Numenta. Perhaps they have similar philosophies to approach but Numenta modeling the Neo-cortex while Vicarious modeling the entire brain?

boltzmannbrain• 9 years ago

No, quite different approaches and goals.

teabee89• 9 years ago

They agree on the bird analogy, except Numenta (IMHO rightfully) thinks you can't come up with airplane designs without having understood how birds actually fly.

arketyp• 9 years ago

Back in the Dileep George days Numenta sought mathematical support for their theory in a Bayesian approach à la Judea Pearl. But I don't think they ever had any illusions that the brain explicitly performs statistical calculations. Hawkins et al. at Numenta are now embracing the theory of sparse distributed representation, developing upon the work of Pentti Kanerva as a formal foundation. It's biologically much more plausible, but technologically less feasible (with the current hardware paradigm). I suspect Dileep George in the end was more eager to put the theory into use.

skummetmaelk4• 9 years ago

Airplanes don't flap their their wings, but they are also orders of magnitude less energy efficient than birds.

If you want computational power on the scale of the brain, power consumption is a real concern.

Houshalter3• 9 years ago

Are planes less energy efficient than birds? Sure they consume more energy, but they can lift many orders of magnitude more weight. To carry that much weight, that much distance, you would need an unfathomable number of birds. Flying for a very long time.

The same with with brains. As far as I can tell, transistors are orders of magnitude more energy efficient than synapses. Synapses use tons of slow chemical reactions to send a signal. Transistors just send a few electrons near light speed.

dragandj• 9 years ago

Yet, no one tries to travel across the continent in an armchair tied to a flock of those energy efficient birds.

hacker_94• 9 years ago

Birds only have to carry themselves though, so it's an unfair comparison.

dkural4• 9 years ago

Except, airplanes kinda do flap their "wings" - each motor blade is a wing. The motors then turn those blades - i.e. flap the blade wings, for the same purpose - push air and generate lift. Airplanes also have an additional set of wings for gliding.

free2rhyme214• 9 years ago

OpenAI & DeepMind > Vicarious

Hyperbolic• 9 years ago

I am always skeptical of people trying to "emulate" the human brain in machine learning. We currently do not have the tools to accurately record and analyze the dynamics of networks of neurons in the brain, and any group that claims to advance ML with knowledge of the cortex is spouting bullshit. Modern advances in ML are driven by great engineering, not biological insight.

shahbaby2• 9 years ago

Even DeepMind says that future advances will come from neuroscience.

https://www.theverge.com/2017/7/19/15998610/ai-neuroscience-...

In the field of AI, our "great engineering" is not even a worthy comparison to what nature has achieved. Maybe there are a few more things to learn from it.

Also we've been studying the neocortex for a long time and have learned a lot more about it than most people realize.

ogrisel2• 9 years ago

While Demis Hassabis and probably some DeepMind researchers use some understandings from the neuroscience literature as an inspiration for their work, I am pretty sure that a majority of DeepMind researchers would rather use a combination of mathematics and trial and error experiments to build an intuition to guide the design of their next iteration of intelligent learning systems.

Houshalter3• 9 years ago

I recall a quote from some neuroscientist. That whenever he hears people say "we know nothing about the human brain", he wants to smack them with a 900 page neuroscience textbook.

I think the biggest issue is these domains are isolated and don't talk to each other. It's not that ML researchers couldn't be inspired by neuroscience research. They just don't know any.

I talked to a researcher outside of the mainstream who was obsessed with biologically plausible models. He got good results, but not SOTA.

However his main argument was that his methods were much faster and more data efficient than standard practice. E.g. they did online learning and didn't suffer from catastrophic forgetting. Didn't require supervision and labelled data.

Standard methods are optimized towards getting the most accuracy on benchmarks, and not necessarily under realistic conditions. Real brains don't get to save huge dataset and iterate over them later. They need to learn in real time and without forgetting previously learned knowledge. Given just a stream of unlabeled data. ANNs can't do this at all. Some biologically inspired models claim to be able to do this well.

eli_gottlieb4• 9 years ago

Real brains don't get to save huge dataset and iterate over them later. They need to learn in real time and without forgetting previously learned knowledge. Given just a stream of unlabeled data.

It's more than just that. Real brains need to optimize their internal data for action. The ultimate test of whether you've represented the world correctly is: but can you do stuff? Can you control an inverted pendulum (to pick a task) while constrained to have your representation updates be Lipschitz functions, with a Lipschitz constant based on your actuators' state (ie: speed, angle, force, etc)?

Note the trick here! Your representation doesn't have to be reconstructive itself (allowing for you to conditionally simulate only Lipschitz transformations), but the updates you perform on that transformation from sensory reafferant signals do need to change only at a bounded rate, because the physics of the thing you're moving actually have that property.

tomp• 9 years ago

I talked to a researcher outside of the mainstream who was obsessed with biologically plausible models. He got good results, but not SOTA.

Can you expand on this? Do you have some resources that describe (something similar to) what he did?

hacker_94• 9 years ago

Hey we fully mapped a worm's ~300 neurons [1], now we just need to keep scaling up until we get to the billions that the human brain is made up of.

[1] http://www.openworm.org/

eli_gottlieb4• 9 years ago

Hey we fully mapped a worm's ~300 neurons [1]

And functionally, what did that tell us?

jgalt2122• 9 years ago

Numenta has been working on this strategy since 2005. Has anyone identified any of their products in wide use?

https://en.wikipedia.org/wiki/Jeff_Hawkins#Numenta

rf15• 9 years ago

Did somebody say "Rocket AI"?

blueyes3• 9 years ago

Never before has an AI startup done so little with so much. That puts their total funding with debt at around $130 million. It's an utter waste. Reminds me of the $15 billion IBM has spent on Watson. The people with the money are very poor judges of technologists in this space.

jorgemf2• 9 years ago

Can you expand what have Vicarious done?

jgalt2122• 9 years ago

given his answer the implication is zero, but I too an curious about any of their accomplishments, proofs of concepts and/or MVPs.

blueyes3• 9 years ago

They've published a few research papers. Few other researchers cite them, so it's hard to argue that their research has been influential, or even significant. In 2013 they announced they had solved/broken CAPTCHA, but refused to share their code, for which they were criticized:

https://www.vicarious.com/news-detail-02.html

More recently, they published a paper on what they call Schema Networks, which are an attempt to blend deep learning with concepts. They're winning Atari games with that, but DeepMind did it years ago.

https://arxiv.org/abs/1706.04317

They're seven years old, and until recently, they'd raised $70-$80 million. I would have expected more than a few research papers in that time, if I were their investors.

Vicarious AI raises another $50M

Related Stories