Ask HN: What excites you most about Neuromorphic Hardware? | Hacker News Reader

Ask HN: What excites you most about Neuromorphic Hardware?

69 pointshsikka44 comments

I just finished https://arxiv.org/abs/1705.06963 and was pleasantly surprised at state of neurotrophic computing. Specifically, the use of organic materials to build robust, low power networks that get past the von Neumann bottleneck and allow us to incorporate new levels of sensing into our environment and lives seems extraordinary. TPUs and chips are obviously interesting for orgs like google, but what about synthetic clusters of neurons integrated into our environment? How significant could that be?

View on HN

Neuromorphic hardware is barking up the wrong tree if you ask me. What excites me is not hardware that works like brains work, but software that does what brains can do. If you want to fly, don't try to build a bird. Build an airplane.

I would agree if the goal is to fly. For some researchers, however, the goal is actually to apply neuromorphic hardware to humans, i.e. to build a bird rather than just to fly.

In a sense, aren't we trying to take inspiration from a model evolved in many million years? We are trying to build an airplane taking inspiration from a bird. Also, emulating the way brain work can be seen as one of the many paths towards AGI, can't it? Even if, AGI speaking, to my understanding the computation speed is not the problem.

Neural networks are already loosely inspired by the brain. Neuromorphic hardware is taking things too far. Just as airplanes were a better path to flight than ornithopters, iterating on AI techniques that work is a better path to AGI than brain imitation.

Remember when William Ditto actually made wetware computing using live neuron cells (yep living cells), but he saw the potential of such computer and stopped research. Instead he focused in chaos computing, and founded Chaologix, where he made analog circuit that can transform to different logic gates very fast. But recently they are acquired by ARM.

The only company I know that is doing wetware computing today is koniku.io though they are super secretive about their progress. And they only made sensors for now.

Now we are trying to emulate plasticity of growing brain with rigid, solid state electronics. I'm not that positive, so stay to deep learning and backprop for now...

Whoah this is super interesting, do you have any papers or research you can point me to? I'd love to learn more about this because i'm a neuroscience and AI student.

Do you have any more details on koniku.io or similar processes? I find the subject interesting, while other people on the thread might be right and is actually barking at the wrong tree that doesn't make it less interesting.

In case anyone is interested, Carver Mead's 1990 seminal paper on the subject [0]. It's a fairly accessible read, and covers the power/computational efficiency trade-offs on the spectrum from natural, biological systems to manufactured analog vs. digital electronic systems.

[0] https://web.stanford.edu/group/brainsinsilicon/documents/Mea...

[1] https://en.wikipedia.org/wiki/Carver_Mead

PS: I did not read the paper again.

Most of such analysis reporting brain to be more power efficient than computers talk about energy it would require to emulate brain operations in silicon. That does not sound like a fair comparison. How about the energy a brain would need to emulate a computer chip, say multiplying a billion floating point numbers?

For a fair comparison, we must do a comparison for the same neutral task, one that both machines and brains can do. It's would need discussions to define what would this be since capabilites of each still show wide differences.

Likewise, some texts assume each synapse to carry a memory of say a byte, and then claim our brain has a memory of about 10^15 bytes. A human brain cannot actually recall all that information, the latter is estimated to be at about 10-30 MB only (per an old book I read).

> brain cannot actually recall all that information

it’s not random byte addressable memory. Neither are the big DL models. But that doesn’t mean it’s not making use of the 10^15 bytes when doing things it’s good at.

You should probably read the paper again. The comparison Mead proposed is for the energy spent on a simplest fundamental computing event. In the case of a brain, it's a single neural pulse going through a synapse. In the case of a digital computer, it's a switching of a single transistor. In both cases, we can calculate how many events happen per second, and we know the power consumption. A brain simply does more operations per second than the fastest computer chip, while spending less energy.

You might disagree on what constitutes a fundamental computing event, and we can discuss that, but the idea seems valid to me.

An apples-to-apples comparison should not care for what the fundamental unit is, unless it can be argued to be the equal in terms of what it does to the problem being solved. The common factor needs to be the same problem given to both to see how much power is consumed. Or alternatively, we could compare what they could do with a given amount of power with a common output performance metric.

Don't you realize that brain and computer are optimized for different tasks? Should we compare them at matrix multiply, or playing chess, or writing a novel, or ...?

Which is why I am claiming that a comparison of energy efficiency of the two, as seen in most literature, is invalid.

I think this comparison assumes that we want to do whatever brain does - the way brain does it. Because we don't know any alternative way to do it. So we try to emulate the brain (replicate neuronal operations) using transistors (or memristors, etc).

This does not make sense for some tasks, such as doing matrix multiply in FP64 precision, but it does make sense for the tasks we care about the most - whatever it is that makes us intelligent (AGI). At least until we can abstract the details of the brain operation which are not important from AGI standpoint.

We should wait for the energy comparisons to be made till we understand. :-) Emulating brain with machines or vice versa is not an interim solution.

Alternatively, perhaps the common task could today be defined based on something that Deep Learning can handle, like say visual object recognition using a pretrained model.

Energy efficiency! Our brains vastly outperform existing hardware while using less energy and producing less heat. Once we understand more about what makes them capable of that we will be able to push hardware well beyond its current thermal-related problems.

Can you share some references. See my comment on this page here: https://news.ycombinator.com/item?id=18425019 Thanks.

I've seen use of rat brains to control devices. Is that considered neuromorphic? Has there been an attempt to use human brain matter in such devices?

Don't you think this would constitute slavery? Shouldn't this be extremely illegal?

Have you ever heard of dairy farms, or horses? If not, I'd advise against searching.

Cows and horses are not people. The use of human brains or human brain matter brings up slavery in a way that rat brains don't because owning animals is not considered morally equivalent to owning people.

Ah, sorry, I skipped over the 'human brain in...' part of the post, and thought GP was responding to 'rat brains'. Please ignore.

Yes, I think it would represent a new kind of slavery, and should be extremely illegal.

Professor Kevin Warwick is at least one person to work on 'rat brain robots'.

I recall he also implanted something in himself to allow controlling other things with normal arm movements, which is sort of using 'human brain matter in such devices', though in situ.

At a high level, I think the top level advantages potentially are in power saving and flexibility when working on machine learning problems.

The brain is very power efficient compared to modern computers. Incorporating more “organic” structured can lead to much greater power efficiency.

As more is learned about efficient neural network models, implementing them in hardware will lead to much faster and cheaper learning models.

Really the only thing that interests me is being able to control a computer with thought, if I can look at a screen and 'click' in my head, to me that's basically Star Trek utopia. Having a thought-controlled keyboard would be awesome too, but not nearly as cool as a mouse.

The rest of it can go in the bin, but I'm sure some of it might be neat.

The name. Branding for this is great. The technical side also has huge potential.

(No this isn't sarcasm or some kind of hipster in-joke)

can someone explain it to the uninitiated?

Interesting that in Fig. 10, half of the HW implementations are analog or partially analog.

If we learn how to program them reliably it’s huge.

Wow bit of ‘light reading’ needed before answering this question. :-p

Note that the linked article is only about 22 pages, followed by 2682 (!) references.

I for one welcome our new Typhon overlords.

Binding them to Neanderthal organiods. Yes, we have plans to do exactly this.

If you'd like to be part of the wildness of that idea and have neuromorphic and/or spiking neural network and/or genetics experience, send an e-mail to patrick.ryan@emblem21.com

Patrick, don't you think you should reflect on what you are doing before taking sentient neurons and forcing them to live their entire lives confined inside a machine, with no possibility of escape? How would you enjoy being born into such captivity?

What would you do if you were experiencing an endless state of pain but lacked the language ability to communicate your pain to your owner?

I have written to the Vatican already suggesting that they push to make neural slavery illegal, as it violates the principle of the dignity of sentient life. I hope you will please reflect on what you are doing and stop doing it voluntarily instead of waiting for the laws to change and make what you are doing illegal. Whether legal or not, any form of slavery is immoral.

Is this satire or do you actually believe that this constitutes sentience being digitally enslaved?

This is not satire. If it helps explain my position, I also agree with the scientific consensus that animals, not just humans, are sentient, and that sentience comes in varying degrees rather than just a binary yes or no. https://en.wikipedia.org/wiki/Animal_consciousness#Cambridge...

For what it's worth, I am vegan and have invested 10 years worth of my salary into synthetic meat companies to help save animal lives.

Curious to know how the Vatican comes into this. Not much of science pays any attention to that regressive medieval institution.

The Vatican these days is no longer regressive, at least on scientific issues. Here is a letter from the Pope in 2015 urging action on climate change: http://w2.vatican.va/content/francesco/en/encyclicals/docume...

Here is a video produced by the Vatican along with the letter: https://www.youtube.com/watch?v=KXA5_juFgDg

The Big Bang theory of the origin of the universe was invented by a Catholic priest. https://en.wikipedia.org/wiki/Georges_Lema%C3%AEtre

I've also spoken with a retired Bishop who mentioned that in his personal view, it is possible that a multiverse with more than one universe could exist. All in all, Catholic leadership today is impressively open-minded.

Neanderthals aren't legally human and organoids do not have awareness in the manner you are attempting to establish. Interestingly, a network of organiods, connected via neuromorphic arbiters, could provide some interesting intepretations of awareness, especially when the organoids cross species lines.

From a national security perspective, the West has to get over its crippling squeamishness on all things genomics or else the Chinese will completely dominate the neuromorphic space.

From a moral perspective, if you are concerned about stopping slavery, I recommend tackling actual instances of it, such as the open air slave markets in Libya or addressing the valid concerns of any one of the hundreds of millions of rural Chinese.

From an ethical standpoint, there are a couple federally funded institutions that traffic stem cells who could use some spotlight in them.

I'm unsure if a religious institution mired in worldwide pedophilia accusations is the right party you want to associate with to stop this. I recommend stepping off the soapbox for a moment and try explaining, precisely, how many neurons are required for the sentience you are describing. Is it 1? 10? 1,345?

Quantifying sentience is your burden. Blanket policies of "all neurons are sacred" are prematurely alarmist.

> Neanderthals aren't legally human

Them being extinct for 40K years leave a paucity of cases on the topic, but I don't think there is any firm grounding for that claim.

Please understand that I have no problem with genetic engineering in general, and am probably one of the few people on this thread with genetic engineering experience. I have customized a DNA sequence and had my custom version synthesized into a plasmid, which worked as designed. I am a huge proponent of genetic engineering technology when applied to extend human life and augment human capabilities.

I think that an arms race mentality is a dangerous approach to engineering ethics. I would rather that the arms race of the Cold War have been avoided if it were possible, even if it meant one or both sides being willing to lag behind when it comes to ethically dangerous technology.

Do you have a citation for the supposition that Neanderthals are not legally human? I believe that the legal personhood of Neanderthals has not yet been established, and would likely have to go the Supreme Court. If legal personhood was denied to Neanderthals (homo sapiens neanderthalensis), would you say that it should be denied to mentally challenged homo sapiens sapiens as well?

I am Catholic and have no choice but to align with the Vatican. I am just trying to do the right thing no matter whether it's with the help of a national government, a religious authority, or simply individuals who want to do the right thing.

Please don't accuse me of not doing enough. I have invested 10 years of my salary into animal welfare companies and outrighted donated 2 years of my salary to other philanthropic organizations including medical and educational organizations. I'm just trying to do the best I can and there's not much more I can do.

Minor modifications to your genome miss the mark. We didn't master metallurgy to gently carve our names into tree bark. We plow entire forests.

Arms races exist regardless of your preference for narrative. The Chinese have zero concerns for forging chimera for any purpose they choose. That's the reality right now. Meanwhile, the Western response is to empower moral supremacists who all believe we're one random CRISPR event from an accidental Holocaust. It's neurotic paranoia rooted exclusively in cultural instability and it's for children.

Neanderthals aren't human for the same reason celery isn't human. Just because we may have a shared genetic branch (sharing in genomic expression is a ton more complicated than linear composition comparison) does not grant the entire mass of biology with human rights. Fighting for the human rights of non-humans is misguided bourgeoisie neurosis at best when you consider the entirety of human suffering that exists within a 500 foot radius of whereever you may be. Your desperate conflation to associate Neanderthals with, wow, the mentally disabled precisely proves my point.

Some science on the matter, which I suspect will have little influence on the subject (Never doubt the ability of moral supremacists to anthropomorphize): https://cosmosmagazine.com/biology/no-human-dna-found-in-nea...

Stuck backing the Catholic horse, eh? How many Neanderthals have been baptized and why has the Church clearly denied the noble Neanderthal such divine access to God's love? Perhaps a retcon is in order and we can throw in the Austropithicine for good measure. This, of course, presupposes an agreement that there were "humans" before Adam or that homosapien evolution is valid. Genesis is going to take a hit here, one way or another...

My question still remains: Precisely how many neurons count as sentience, or are we Overton Window dancing to see if we can start with a "all neurons are sacred" policy?

Reply on news.ycombinator.com