Emergent dynamics of neuromorphic nanowire networks | Hacker News Reader

Emergent dynamics of neuromorphic nanowire networks

phys.org

74 pointsknown9 comments

This nature paper is spawning a bunch of articles with rather sensational titles. There was a post about one such article a few days ago. The actual paper is quite readable and less sensational: https://www.nature.com/articles/s41598-019-51330-6

https://news.ycombinator.com/item?id=21879444 was the previous discussion, but this one already seems better, so we can leave it up for now.

We'll change the title to that of the paper, which is less sensational.

"After current flowed through the network, the connections between nanowires persisted for as much as one minute in some cases, which resembled the process of learning and memorization in the brain. Other times, the connections shut down abruptly after the charge ended, mimicking the brain's process of forgetting."

Memories are of something. There is no indication that anything like that is going on in this device.

Sure, this might be "brain-like" on the physical level, but that's a far cry from being brain-like on a functional level, much less being brain-like on a subjective, experiential level (ie. we actually experience memories)

Still interesting and promising research, though.

I'm always wary of comparisons and interpretations that work regardless of what happens. One one thing happens, that's like a brain. When the complete opposite thing happens, that's also like the brain.

It reminds me of bad Freudian analysis: drinking too much or talking too much indicate an oral fixation. So does not drinking enough or talking enough. Also, not drinking or talking excessively is not an indication that you don't have an oral fixation, since any other vague metaphor about the mouth can also indicate an oral fixation.

The behaviors this chip exhibited couldn't be not exhibited by the microchip. The comparisons are suitably vague and can be applied regardless of the outcome of the study. Especially when you consider we don't even know precisely how memory, sleep, or wakefulness work in humans. So you're vaguely comparing a device to something you don't even understand.

The actual paper might be more rigorous and level-headed, but the write-up is pure pseudoscience.

I mean, woah. Hard to decide if these scientists created a chaotic system and are simply drawing the conclusion that "our system of silver nanowires is chaotic and the brain works kinda chaotically, so, this is a proto-brain" or truly "we spawned a chaotic system that self-organizes into meaningful, manipulable thought-like behaviors".

It seems that a similar experiment was used to implement a ML algorithm to predict traffic patterns--but they didn't provide enough detail for us to ascertain if those silver nanowires were more-or-less "a tiny, classical, programmable cpu built out of nanowires".

The biggest questions in my mind are:

1) How much can this system be manipulated/programmed to solve or "focus" on specific (or fuzzy) tasks?

2) What does the I/O look like for this? Can varied and non-trivial datasets be fed into the system? What does getting data out look like?

Super interesting if this all legit.

More like what conclusion are the reporters drawing, because usually it's a million miles from what the researchers are concluding, but much more exciting.

> 1) How much can this system be manipulated/programmed to solve or "focus" on specific (or fuzzy) tasks?

There's a small subfield of ML called Reservoir Computing

https://en.wikipedia.org/wiki/Reservoir_computing

In RC the input is processed by a random network (the so called reservoir) followed by a small output layer.

reminding that scientific reports publishes articles that are reviewed to be technically correct but regardless of their importance.

Are they saying a Current is spontaneously emerging and disappearing?

Reply on news.ycombinator.com