Ever since selling Handspring to Palm in the early 2000s, Jeff Hawkins, creator of the Palm Pilot and founder of Palm, has been working on his true passion: neuroscience and trying to understand how the brain works. Teaming up with several neuroscientists and some former Palm people, his company Numenta, entirely funded by Hawkins himself, is now ready to show its research to the world.
Mr. Hawkins says that before the world can build artificial intelligence, it must explain human intelligence so it can create machines that genuinely work like the brain. “You do not have to emulate the entire brain,” he said. “But you do have to understand how the brain works and emulate the important parts.”
[…]
Now, after more than a decade of quiet work at Numenta, he thinks he and a handful of researchers working with him are well on their way to cracking the problem. On Monday, at a conference in the Netherlands, he is expected to unveil their latest research, which he says explains the inner workings of cortical columns, a basic building block of brain function.
Numenta’s research is apparently so complex that Alphabet’s artificial intelligence research company, DeepMind, told him they simply didn’t understand it. If this work, which I think is detailed in this scientific paper published over the weekend (but don’t quote me on it – it might be another paper altogether), is indeed the breakthrough neuroscience has been waiting for, it could have enormous consequences, not just for neuroscience and biology, but also for artificial intelligence and its applications in the world of computing.
I’m very curious to see if this research holds up to scientific scrutiny and peer review, because even the smallest of steps towards understanding how the brain works would be a massive scientific breakthrough.
I’m very curious to see if this research holds up to scientific scrutiny and peer review, because even the smallest of steps towards understanding how the brain works would be a massive scientific breakthrough.
You can bet your life that it is complete garbage.
Everything Numenta has published has appeared in (worthless) pay to publish open access journals. The ‘Frontiers’ journals are generally considered to be predatory.
https://forbetterscience.com/2017/09/18/frontiers-vanquishers-of-bea…
Edited 2018-10-15 02:55 UTC
I find that bunk in itself. The dismissal of the publisher and the idea that publishing something there somehow make it not proper science – crap.
But then I’m all for open science where people may publish things however they want and their research judged by the methods, results, and successful replications. Opening up peer reviews would remove the only reason for the expensive journals limiting information distribution with no advantage for those doing the actual work for free – scientists and peer reviewers.
The journal they used is a well known predatory (vanity) publisher.
The entire academic publishing process has degenerated into a complete farce. Junior academics can now ‘buy’ promotions by paying to publish, self-citing and using ‘citation cartels’ to boost their output.
I know of two third rate junior Australian academics who ‘bought’ themselves full Professorships by paying to publish more than 100 papers in third rate predatory journals.
Edited 2018-10-16 03:31 UTC
Airplanes do not fly like birds or insects. Why should we expect artificial intelligence to be similar to human intelligence?
When I checked the last time they all operated on the concept of creating lift. Rockets on the other hand…
And dammit, you just reminded me how much I want an ornithopter
The keyword is: “Artificial” and that’s it.
Will be funny when some “AI” declares you as not really intelligent… (go through a list of cognitive biases – this is our primary mode of operation)
Edited 2018-10-17 13:17 UTC
And rockets also have a biological precedence, though not in air but underwater – squids and octopuses. Kites have a precedence with spiders ~flying tethered by their threads on the wind. But the first way we humans achieved untethered flight, balloons, seems unique…
Why speculate, the proof will be in the pudding so to speak
Jeff pops out about every 7 years with fantastic claims to get funding for another 7 years. What they’re doing is advanced pattern matching systems, but has nothing to do with human-level generalized AI.
Or going “Ghost in the shell” style : https://arxiv.org/abs/1809.08632
Mental works at neo-cortex are of the representative (thinking of Shakespeare), holographic kind. Alphabet many years on a different path, not a wrong one.
On this perspective, easier to understand why we prefer paths, workflows, to models. We learn workflows, not models.
Edited 2018-10-15 21:04 UTC
…I’ve only skim read the linked preview article so far, so yet to fully digest
but it seems Numenta and Jeff hawkins’ thinking and models haven’t progressed overly far since On Intelligence was published.
I think the overall loose model for brain thinking of it continually polling multiple inputs and and decision making being a hierarchical based on action potential strength in the decision making unit is fair enough.
But I really don’t like the suggestion of the multiplicity of objection definitions scattered multiply times amongst many cortical columns (1mm^3 neural clusters plus-minus) as intuitive it doesn’t make much sense.
Why I see a post-box in the distance and before I know what it is, the identification phase – I first discern the rough shape (rod-shaped, domed on top)….in the US, could be a fire hydrant. But it’s 3 times the length of the labrador beside it…..and Green! (or UK, Red! or..)…Postbox!!! ta-daaa. But it’s in the reverse direction that the connections and the word association that is more telling and instructive.
Say in your head the word “postbox” – and you automatically and instinctively think of the key ID charactheristics – Green, Pill-shaped with the bottom scalped off, Slit at the top for the letter entrance.
There’s no way, “postbox” or stored at multiple memory or processing locations in the brain any more than “green” or “red” are. They are certainly associative connections each-way between all the “object-ID cells/cell-clusters” but they have to be one-offs. That’s what I don’t like about their current Cortical Column descriptions. Object of course being any one of “object(noun)”,”characteristic”,”feeling(word)”,”grammar-word” ,”grammar-rule”, etc.
When you “see” a post-box (with a dent in it) – you automatically (may) think of all the ID characteristics that define it, and even the actual word “post-box”…pretty much at the same time. But if you have to have earlier in your life crashed off your motorbike just after having been dumped by your partner – you may
instead (re)”feel” or even be re-traumatised by the memory of that event before even getting to some of the ID-associative words (within your head)…..and this will be dependant on your mood that day and the colour of the sky at the time etc etc- ie which thought/association and or feeling surfaces first. But they’re certainly all “object-memory” or “event-memory” clusters/circuits that connected in multiway directions. I’m just not convinced they have the wiring model any way right yet.
Willing volunteers for invasive brain studies using isopope-labelled metabolite/neurotransmitters/gliotransmitters with micro-scale PET scanners yet to be invented might help. (slightly sarcastic)
It’s all interesting theories though. But the expounding proofs and wet-lab / Electrophys work to back em up is still lacking
Edited 2018-10-15 22:52 UTC
[quote]I’m very curious to see if this research holds up to scientific scrutiny and peer review, because even the smallest of steps towards understanding how the brain works would be a massive scientific breakthrough.[/quote]
As with any work in science, you open one, you open up another endless questions. This is how nature was designed from the ground up. It’s like opening a can of worms.
Ehh, it’s messy because it wasn’t designed…