Hacker News new | comments | ask | show | jobs | submit login
How the Brain Creates a Timeline of the Past (quantamagazine.org)
188 points by chmaynard 8 days ago | hide | past | web | favorite | 23 comments





"Just a few months after Howard and Shankar started to flesh out their theory, other scientists independently uncovered neurons, dubbed “time cells,” that were “as close as we can possibly get to having that explicit record of the past,” Howard said. These cells were each tuned to certain points in a span of time, with some firing, say, one second after a stimulus and others after five seconds, essentially bridging time gaps between experiences."

Most relevant idea from the article for me


We know that entering a new room in your house impacts your thinking -- like a new "window" of thought opens. There is, in fact, a concrete reason why you will forget what you were looking for as soon as you enter the room you intended to search for said object.

It's interesting, as mentioned, how you can recall other things that happened in temporal or spatial proximity to another memory you're recalling.

My mother has an uncanny ability to unwind a series of connected memories -- "I know that was in 1973 because that's when so-and-so got married, and we had traveled to such-and-such location that year, which is why I had the new suitcase that..."

Fascinating research.


As a programmer I think I forget a lot because programming keeps my brain busy thinking about things. I forget features and solutions I've done before. I think it's because the space inside the brain is limited. I'm thinking hard about a new problem every day. Whereas my boss can remember things I did a year ago I can't easily.

I find that my brain seems to be very selective about what it copies from "cache" into long-term memory. I think this is the result of decades of splitting my time between two 'modes' of thinking - one where I try to permanently commit some important fact to memory (like how a system works, or what functions it needs, or whatever), and the other where I'm churning through hundreds of pages of documentation trying to find the one particular fact that I need, where everything apart from that fact gets explicitly forgotten as soon as possible to make space for stuff that I actually need to keep.

Maybe its just you being in the business longer than me, but I've had pretty much the opposite thought: I'm learning new stuff every day, wondering when I will start to actually forget chunks of it. Currently, it seems like most technical knowledge tends to stick in my brain, with no apparent limit yet. I can still recall most of the Redis protocol commands despite not having used it in more than six years.

After almost 20 years as a commercial developer I forget entire projects, entire jobs even. Someone will mention a company name and it takes me a few seconds to realize that the reason the name is familiar is because I briefly worked there 15 years ago. Jobs I held for multiple years are fine but any short stints are a blur.

I've experienced this as well and the phenomenon is more pronounced if the role/company was unremarkable at the time, and it was as if I overwrote the memory with the excitement of the next job, then rinse and repeat with every new job.

I guess something like that is going on in my brain too. The code I come up with often feels exciting because it is a solution to a problem, but then the next several 100 code-solutions overwrite the memory of what once felt important. It is no longer important, so I think it's good my brain let's go of it. But often I lament having not much documented why I wrote some code the way I did

I suspected that would probably happen someday (I'm 24). Do you notice this effect with your private life too? Forgotten relationships, appartments, vacations, etc?

Yes. Some stick in your memory but others kind of get swept under the carpet.

They are all still in your memory somewhere. A photo, something someone says, or even a smell can cause them to come rushing back. "Holy crap, twenty years ago I went to blah and this happened... I'd forgotten about that."

Think twice before verbalizing similar moments of sudden recall about old relationships with your current partner. "Man, that made me thing of this chick I was seeing for like a month a few years before I met you..." In terms of their appeal as a sexual partner very few 45 year old long term partners like to feel compared to a 22 year old you had a brief fling with.


> wondering when I will start to actually forget chunks of it.

When you start forgetting it, you don't notice that. There's no bell ringing indicating now something's gone from your (easily) accessible memory. We are not aware of what we forget :-)


I actually think it's much simpler than all of this. I may be extrapolating from my own brain a bit, I'm fairly certain that the brain creates a timeline of the past by running a garbage collection function every night that goes something like:

    if (memory.timestamp < YESTERDAY) delete(memory);

I think its a bit more complex than that, maybe something like

  if (memory.timestamp < YESTERDAY && memory.significance < 0.1) {
    delete(memory);
  } else {
    dream_playback(memory).then((score) => {
      memory.significance = score;
    })
  }

So if I understand correctly, you are suggesting that the dream playback is used as a way to determine the significance of a memory?

Hopefully this leads to better temporal priors for machine learning models. Sequence encoding frequently uses position coding[0] but the typical approaches (e.g. low frequency sin waves) are fairly naive. In reinforcement learning we have (somewhat obvious) works that show that providing explicit time representations help[1]. Simply throwing a recurrent model at a sequence may pick up simple ordering but improving the temporal representation would improve data efficiency.

[0] https://arxiv.org/abs/1706.03762

[1] https://arxiv.org/abs/1712.00378


Lots of people have suggested time distortion effects from psychedelics have to do with altering how this system works -- if time input decays more slowly, that would explain things like visual trails and so on.

Are there any research groups who are actively investigating how insights into the human mind may be applied to artificial intelligence methods? If that's even applicable?

Edit 0: I should clarify that I am well aware that neuron architecture already has influenced modern AI methods, I was leaning more towards non-neuron based methodologies.


This is what everybody has been working on for the past ten years.

Numenta is probably the most direct / well-known example of what you're asking. They get mixed reactions from the community (some fair, some not). DeepMind occasionally publishes crossover papers from neuroscience (e.g. grid cells; metalearning). Lots of other groups are looking at neuroscience for applications to ML too like University of Washington and Columbia.

"Biologically Inspired Cognitive Architectures" and they have some overlap with the AGI (AI at human level and beyond) folks as well.

http://www.bicasociety.org/


Neural networks has greatly been influenced by insights into the human mind. Hence, also why it's called "neural".

So yeah there are research groups actively into this :D Usually, researchers at universities.


The journalist failed to explain "Laplace transform" part unfortunately. How something like complex variable operation happens in the brain. Most likely the original researchers meant something different.

Maybe a bit unrelated but I thought of this as I was reading through the article: I think it'd be very interesting to invest in a dreams research centre and I'll definitely do so as well eventually



Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: