> It turned out that either visual stimulus—the grating or moving dots—resulted in the same patterns of neural activity in the visual cortex and parietal cortex. The parietal cortex is a part of the brain used in memory processing and storage.
> These two distinct visual memories carrying the same relevant information seemed to have been recoded into a shared abstract memory format. As a result, the pattern of brain activity trained to recall motion direction was indistinguishable from that trained to recall the grating orientation.
Isn't the alternative explanation that our tooling for inspecting the brain at work abstracts too much detail away for us to be able to tell the difference?
Indeed, it's like saying "the CPU used 16 Wh of energy, executed 1 billion MOV instructions and 2 billion ADD instructions for both these two tasks, thus the algorithm it ran must be identical".
That's possible. But I think the reason why this is interesting is because you are seeing the same kind of representation in the brain for two seemingly different phenomenon, motion and orientation.
It's intuitive to see how you could represent motion by an orientation (we do this with vectors in math), but its interesting to actually see it happen.
Right, it's quite obvious that the memory is not being stored bit-for-bit in exactly the same way because if you ask the person what they saw after the experiment, they will be able to recall the difference of "lines" or "dots".
But the paper is explicitly looking at the representation in working memory; so two obvious possibilities are, one that the "orientation" and "dotness vs. lineness" attributes are being decoupled and stored separately in working memory (different "registers" if you will). Or the "dotness / lineness" is getting stored somewhere else (not working memory, some other memory system) because it's not "behaviorally relevant" (i.e. relevant to the task that the participant is attending to while creating the working memory). I'd guess at the first because my impression was that essentially everything that makes it into episodic memory starts in working memory, but I'm not a neuroscientist.
I think the OP is getting way ahead of itself with "The findings suggest that participants weren’t actually remembering the grating or a complex cloud of moving dots at all.". The paper is making a much more modest claim that "direction" is recorded in the same underlying way, specifically during a task where you're being asked to recall direction. It's completely possible that this intermediate/common representation would not be generated if you're just looking at the pattern and not performing a task related to direction.
I couldn't find the full paper on SciHub, just the abstract linked in the OP: https://pubmed.ncbi.nlm.nih.gov/35395195/. I'd hope the full paper talks about all this in more detail.
Exactly. Our tooling doesn't even allow us to make clear inferences about neuronal activity yet - all we can image with fMRI as of today is heamodynamics.
The work proposes that because responses to difference visual stimuli show the same haemodynamic spatiotemporal response in memory areas, the actual memories must share a common representation.
This is debatable since we know that localized spatiotemporal responses reflect the variability in the vascular tree and blood flow and volume, regardless of the experiment at hand [see 1 for a discussion]. To dare to claim that effects could be neuronal one needs to run a bunch of extra control experiments (vascular reactivity mapping / breathhold hypercapnic challenges, resting-state imaging as a control dataset), none of which were conducted in the proposed work.
This is known to MR physicists, but hasn't clicked in yet within departments of cognitive neuroscience.
It all works this way (though a certain female family member would disagree, claiming to remember conversations word-for-word years later).
But my memory works this way. A summary "party at so-and-so's house, weather was nice, overall vibe was ___". The rest is context. You know what the house/backyard is like, you know the general feel of that time of year, you know the crowd that usually comes, you can easily synthesize details like the smell of the BBQ and the taste of the food... build up a complete "memory" from stuff that could be summarized in a paragraph of text plus generic (not specific to one memory episode) context.
I can build up a relatively vivid mental image of my walking route to school (from the bus terminal) over 40 years ago. Is it accurate? Who cares. As long as no detailed record exists to compare it to that would reveal the "lossy compression".
One of my oddest part of my dreams, is that they often tend to be places from my childhood or young adult life and that my brain seems to processing the 3D layout. Like I will walk specifically to school, remembering the route, or through my church and I had re-visited a giant thrift store from many moons ago and my feet just trod the path right where I knew I wanted to go. It's like watching my mind process these locations into mental maps in dreams. Kinda neat
Using routes is a key technique in memory techniques (an the so called 'memory palaces'), presumably because when we went hunting for food we needed to find our way home, so memories attached to routes are a lot stronger.
Interestingly I was able to retrace the walk two decades later (we had emigrated to another country in the meantime) and while the "vibe" matched, the details were quite different from what I thought I remembered (this is an old town in south Germany where things don't change that quickly so it wasn't redevelopment).
But it was possible, with a bit of head scratching, to walk the route just from memory.
Ditto on the "certain female family member who insists that she remembers things word-for-word". When she recounts her meeting with a friend it is needlessly tedious (I try to be a good listener of course). Complains that my recollections are too vague and she wants to know what really happened and is frustrated I won't give her details.
I think a large part of it is just that you store what is important to you. To me the day-to-day politeness is just filler. I don't care if they had black coffee or a latte. If someone was struggling with something and poured out their heart over multiple conversations, I'm going to remember what arguments and concerns they had and the mental model I built up around that situation. The filler is just unimportant and doesn't stick around.
My wife is the opposite. Signs of weakness are an embarrassment to be forgotten. She lives for the day-to-day.
The Myers-Briggs system distinguishes call these two perspectives "Sensing" (detail orientated) and "Intuition" (theory/model based) [not the best names]. And it posits that it's less a matter of importance people place on things and more that people literally notice different things and perceive the world differently (so it's not even just about remembering, it's about what you notice and how your mind represents the world in the first place).
Meyers-Briggs is a fundamentally non-empirical model. I wouldn't recommend it as the basis for any argument or position concerning real world phenomena.
I don't think Myers-briggs is fundamentally non-empirical. Empirical evidence is certainly lacking for it, but I think there's a good argument to be made that this is due to poor experimental design (for example applying the categorisations to persons rather than to mental processes) and a general difficulty in empirically measuring internal mental processes (it's notable that more mainstream competitors to the Myers-Briggs like the five-factor model don't even attempt this).
I would also point out that I was replying to a comment that was an empirical observation. My comment highlighted that their empirical observation corresponds to the pre-existing Myers-Briggs theory (which suggests that other people have previously had similar observations).
No, Meyers-Briggs is fundamentally non-empirical, it has no empirical validity, nor was it derived from any empirical process. If you have a personal faith in its validity then I'm sure I don't mean to disrespect that.
If you're talking about the testing instrument itself, then sure I don't think it's great. But if you're talking about the underlying theory (and that's much more interesting - when we discuss physics we talk about the theory of relativity not the tools we use to measure it) then I would like to point out that the theory (lets call it Jungian Type Theory - JTT) was derived from an empirical process. Specifically it was derived from Jung's observations as a clinician over several decades. That's not a controlled experiment, but it's certainly empirical.
Regarding my personal experience with it: I have personally found it highly useful as a predictive model of behaviour. It's the only thing I've found that allows me to explain the aspects of people's personalities that aren't easily explained by their environment or life experiences. And by combining with JTT with an understanding of someone's life experience and environment I've found that I can pretty much always find a satisfying explanation for someone's behaviour in a way that I am not able to do with either one individually.
Regarding its validity:
- Firstly, there's no need to tiptoe around the subject if you think it's bullshit. I won't be offended, and I can totally understand why might be skeptical given the experimental evidence that exists so far.
- Having said that, I would like to challenge to idea that we ought to expect JTT to be experimentally validated given how limited our current ability to inspect the brain is. Specifically (and unlike other models of personality) the theory's primary claims are that there are certain internal thinking processes, which of course we currently have no way of observing directly, and which will not necessarily correspond straightforwardly or 1:1 with observable behaviours (environmental and life experience factors being huge confounders).
- Given this I find it entirely unsurprising that experimental designs which rely on numerical scorings of observed behaviours fail to find an effect
- I think it no coincidence that this theory came out of clinical psychology, because you need to be able to control for the environmental and life-experience factors in order to be able to see the other pattern that is (well ok, might be) sitting there beneath them. And therapeutic relationships which continue over a number of years are one of the only scenarios where that context is available outside of close personal relationships.
A while back I went on a google maps street view tour of a place I lived until I was 9 but hadn't been to in well over a decade. I wasn't sure what to attribute to the tenuous nature of my ancient memories versus what things had actually changed since I last looked. It was honestly a bit uncomfortable and disorienting having this gaping hole in my perception of reality. Was the swing set always blue in that park? I thought it was yellow. Maybe they repainted it? I will never know.
though a certain female family member would disagree, claiming to remember conversations word-for-word years later
Surely many people do. Otherwise you wouldn't have all these biographies and non-fiction books packed with conversations people have managed to recall in a level of detail enough to not get sued. I can barely remember a line of conversation from this week, let alone important ones from years ago, so I always assumed most/many people can remember conversations to some reasonable level in a way that I cannot.
I think most people remember the basic concepts and then they fill in the details using what they know about the situation and participants. I have remembered events a certain way that in my mind was very clear. But upon reviewing said events in old video, it turns out I got quite a few details wrong. Sometimes two people will recall the same event very differently. Which is why I think our justice system relies far too heavily on witness testimony.
I suspect many (most?) conversations in biographies and non-fiction books are not necessarily quoted verbatim. In most cases, the author may at best have had access to diaries or other notes from the time that recorded a summary of what was said, or they may have interviewed people who, years later, summarised what they recall -- more or less accurately -- being discussed.
The author may then present this in the form of quoted speech in order to make it more vivid and compelling for today's reader, but it rarely corresponds to a precise transcript of the original conversation.
This is more about how what you can remember about an event after five seconds differs from all that you experienced, as opposed to what you can remember a year later. I think most people can give a word for word summary of an utterance after a few seconds so this particular experiment doesn't really have any bearing on your relatives claims, which are more about recall from long term memory rather than working memory.
My accusation is that the conversation memory works the same way as the BBQ party memory. You remember a skeleton. This subject was discussed, and things were said that gave me a feeling of ____. And a few more easily compressed details. The rest is interpolated. Imagine a language model the size of GPT-3 being trained on one particular person's manner of speaking and then given a one-paragraph summary of a conversation to get it started. Barring an audio recording or a transcript, who's to say that these weren't the words that were spoken?
Of course the engineer is tempted to test this by secretly recording a conversation and trying to trip up the perfect rememberer, a year later. But the non-geek life experience accumulated says don't go there.
I should add that as a geek I ought to have a better ability to remember, say, computer code that I've written. But am I the only one who, going back to something I haven't touched for two years, has to re-learn my own code?
>But am I the only one who, going back to something I haven't touched for two years, has to re-learn my own code?
No, that is perfectly normal, and it starts much earlier, weeks sometimes days after leaving the code.
Depending on its complexity and level of its abstraction.
You mentally build something highly abstract without much emotional or bodily bond.
Your brain has not much incentive to rememeber it.
Adding to that, there's a lot of sampling bias as well. If a function fits my mental model of it, then I'm unlikely to revisit it. If a function doesn't fit my mental model, then it is very likely that I'll misuse it, increasing the likelihood of a bug, and increasing the likelihood that I re-read the code.
Not to forget, memories are not only unreliable per se, but also change with each act of their remembrance.
For example, by character peculiarities, new experiences, current circumstances, etc.
Often they are made up on a whim, without the remembering person being aware of it.
So in a sense, memories have a past and a history.
>I can build up a relatively vivid mental image of my walking route to school (from the bus terminal) over 40 years ago. Is it accurate? Who cares.
Not only that, but by recalling and rebuilding memories, how gaps are filled in depends on your current mental state. For example, if I'm feeling depressed and brooding over past social interactions, I'll likely imagine people having meaner expressions or saying harsher things than they did. The big problem is that your memory of the event is "written over" based on the rebuilt memory. Again, only the seemingly important bits, but people are more likely to remember emotionally strong portions. Like those imagined harsh words.
I realized I was doing this when I thought a professor strongly disliked me, avoided his classes for a couple years, but then found him pleasant. My depression and social anxiety had warped my memories over the years. Being aware that this happens really helps. I trust negative parts of memories less, and I consciously stop myself when I start to brood (or at least, have fun with a puzzle while thinking back on things).
Just thought I’d share an example of how this memory issue manifests for me.
I listen to a lot of music and use it (like many) to index various stages of life - i even have playlists by year to help facilitate this.
The problem occurs when I listen back to music i was listening to in, say, 2008. All of a sudden, i’m transported back. But, each time i do this, the effect wears off a little, because their is some - let’s call it - “meta data” being written from the current moment in time im listening from which is adding new color to the initial index.
This effect has been studied and noted somewhere before, but i’ll have to dig up when i’m not on mobile.
Photographic memory is not a real phenomenon though. But eidetic memory is real, some people can remember almost everything they read. But they don’t remember photographic images.
> I can build up a relatively vivid mental image of my walking route to school (from the bus terminal) over 40 years ago. Is it accurate? Who cares.
It's not just decades old memories. Memory of recent events is likely to be suspect. Which is an issue for the legal system because it relies so heavily on eyewitness testimony.
I think some memories are closer to lossless compressions than lossy which I wonder if it's more of a scale where memories can slide between the two modes with varying degrees of fidelity. Like there are memories that I know I shouldn't remember from childhood that I can remember clearly and others I barely remember what year it happened. So I have to wonder if some of this seeming lossless-ness is more fractal-like in nature where one can just reconstruct from the base encoding and expand it outward to fill in sufficient detail to seem like it's perfectly captured when it's really just merely the reconstruction.
I vaguely remember reading something that traumatic or “very important” memories never go through the usual process of becoming memories. Instead, when you recall them, your brain physically “relives” it so it is never forgotten. Probably a evolutionary trait to make sure we learn as much as we can from the experience. This is also why you remember those “times you almost died” in slow motion. Your brain goes into a high resolution mode in those cases, which you remember as slow motion, like speeding up a camera and playing it back at normal speed.
Sorry I don’t have any sources, I’m just a casual reader in this space.
If you are taking a truncated SVD, the math says that it is the best representation of that data for a given truncation size, and will even give you a measure of how good that representation is. But picking how good you need often ends up being a kind of annoying and fuzzy heuristic thing. In addition, some data just gives you better singular values, and so fundamentally compresses better.
I guess the brain probably is dealing (in a hugely non-mathematical way -- it is just an analogy!) with a similar sort of thing. Somehow we pick some memories to keep in great detail -- either because they seem to be very valuable, or because they just seem to compress nicely.
It is a bit funny that one name for this sort of thing is a "singular experience."
My perception is that I've only collected "raw" materials, left unprocessed. Not sure if it's compressed and lossy or lossless. There's a quote that say like "if you can't explain, you don't really understand", I really hate this quote because I don't have to explain! Those necessary memory will come up, load into my processing unit and execute it.
This is nothing new; I have read several books and works on neurology, and this is best described a "a simplified representation of the environment". Thanks to signal noise and neuroplasticity over time the weakest connection points between "remembered" stimuli deteriorate and all what is left is even more simplified version of a "memory". I am surprised that they did not heard of it yet.
That doesn't sound like quite the same thing. This finding seems to suggest that the memories are compressed from the get-go. Where you are describing why memories get more compressed over time, I think.
The compression already begins with the receptors, maybe I should have started there. Each stimuli/pattern gets more simplified with each neuron layer, e.g. if a region of receptors fire a the same time, fire that one neuron, if not at the same time, inhibit that neuron, if nothing happens, do nothing. It's impossible to "capture" stimuli without compression with neurons in the first place. Information is being "reduced" or encoded if you will along the signal path into the brain, and then over time when recalling this information.
As someone suffering from Aphantasia [0] (I don't have mental imagery at all) and I've been telling people for the longest time that this is how I relate to the world. I summarize things. Even my mother's face. A post by a Facebook engineer [1] felt like a good way to understand it.
Same here. Recently tried to explain this to someone who has vivid imagery, but it was challenging. It seems we do have a wildly different experience of life in this aspect.
Imagine you sit at your desk all day answering emails. Emails come in, responses go out. Except when you step back from the desk, it's just a black void. Information from your eyes? That's just an email saying what grandma looks like. Pain in the leg? Re: URGENT. Nothing exists beyond the emails. The emails are reality. The brains representation language is the same as it's actual language. Why have more than one language?
I realized some time ago when I learned of Aphantasia that it is a spectrum. From 0 to 10, 10 being perfect photographic memory and zero being total Aphantasia, I feel like I'm somewhere in the middle. I can recall images, sounds, memes, faces, but in terrible quality, with very little color or focus, more similar to fast paced dreams than photographs.
I've been wondering the same ever since I read that Nikola Tesla invented/designed the AC motor in his mind's eye.
Seems to be along the lines of lucid dreaming, with a vast difference in degree. Sometimes as I'm falling asleep I can see vivid scenes or objects that I can--to minor degrees--play with for a short time before I either fall asleep or wake up, then it's gone.
I would take something written about Tesla on such an intimate level with a grain of salt though. He is very very hyped and often elevated to a God-like level.
also, consider somebody who is an expert already in the problem domain.
most of us here are programmers and do this on a daily basis. somebody describes "A GraphQL API driven by a clojure back end connected to a postgres database" and to a layperson that looks like either a bunch of nonsense words or maybe a few boxes, clouds, and arrows. but to you and me we can visualize the individual lines of code, configurations, functions, and infrastructural requirements behind that simple sentence.
same with an electrical engineer/inventor in their domain.
I wouldn't say that I exactly 'visualize' it. For me it's more of a bunch of formless even nameless things and connections that don't have specific shapes or places, only logical structures. The lines of code also flows freely but it feels more like synthesis from those ideas rather than copy-typing. Eventually the 'picture' or network in my mind gets complicated/unstable and I have to draw it out.
I've been trying for years to visualize. My SO is an artist and my mother a psychologist, so I've been trying to gets tips and tricks from them. I never managed to even get a hint of color.
You sound similar to me, girlfriend is an artist with an incredible visual imagination, mother is a therapist, and I was at a 0 on the scale before I met her. My gf and I have had some deep conversations, sometimes assisted by MDMA, and at times to the point of crying in front of her in a state of completely trusting her which uncovered some past trauma, social fears, and other discomforts I needed to work through. Anyway after each of these times it would get a little easier to visualize; simple colors at first, then colorful shapes, now small snippets of images that come in and out. Maybe a 2-3 on the scale. Also my memory has improved, not so much for technical stuff, but just remembering the details of my life which before had huge spans (in years) that I mostly didn't remember.
Anyway this might just be specific to me but something to think about.
I definitely think the ability is there and can be unlocked. Like the way you can sometimes figure out how to fire some random muscle on your body with practice.
While I was in jail I managed to imagine images in vivid color and detail several times, but never on demand. It would happen randomly when I was just lying, bored, thinking about stuff. Each time was one of the most amazing experiences of my life. It was like going from being a caveman looking at scratchings on a wall, to straight into a 3D IMAX movie. The quality was that spectacular, with bright colors and completely 3-dimensional. Since I left jail it hasn't happened again a single time.
I have to wonder if that's how others see inside their minds? If so, I am in awe.
I am very interested in the topic, and have been looking into it for ages. I think most people vastly exaggerate their ability to visualize anything. Most people can't really hold a square or a sphere in their mind, rotate it, or change colors. The only people who truly can are really good artists. My point - you may be mis-diagnosing yourself, especially since aphantasia doesn't seem to have clear tests or definitions. How could it, if a verifiable test would be to ask a person to draw what they see, and obviously that confuses the whole test with one's art skill.
Could be that I am mis-diagnosing myself. I've never seen a mental image in my mind. I've never been able to conjure one and I've been trying for years before falling asleep to conjure even a sense of color. Nothing. Black.
To be honest, it doesn't feel like such a handicap to my life that I would start submitting myself to clinical trials. If the worst to come out of my mis-diagnosis is this post, I can live with it.
I’ve always thought of myself as being fairly good at visualization.
For example, I can imagine multiple 3D shapes at one time, rotate them, keep track of which direction a face is pointing on each one, etc.
However, I don’t really “see” any image. It’s more like a feeling of seeing it. Now I’m wondering to what extent other people actually see things they imagine…
Now, close your eyes and imagine another thing and describe it
comparison of imagery in reality and visualised. This presupposes people describe things visually even when directly seeing them, and not in other modes (texture, sound, smell , etc)
I am an atrocious artist who absolutely kicks ass at those mental object rotation tests. I can very easily manipulate objects in my head, but draw a picture? It's an ugly mess.
Is it really unique? I can visualize a sphere, rotate it, rotate the "camera", see it in wireframe, apply any kind of texture, reflections, make it bounce, like working with CAD software. I can picture the image through a fishbowl lens, or through telephoto. However, I do not believe, for example, the reflections or the light sources to be realistic. I can "see" the effect of changing the lenses, but I don't think they correspond to reality. I think that's where people exaggerate. The dimensions, light sources and reflections are not based on reality.
I can picture anything that I want. Movie scenes with my friends faces in them. I always thought everybody could do this. If it's somewhat unique, can I use it for something?
This is absolutely exceptional if true. As far as I am concerned you should be a good artists out of the box. The argument that "well, it's drawing the line well" is mostly bs - it's being able to visualize proportions, distance between things, etc. Most artists actually don't - they follow rules. A person starts as a square and gets broken into quarters with each quarter denoting say the middle of the chest, the groin, the knees, etc. Each box is then broken down more (eyes in the middle of the top box, etc). If you can actually visualize things accurately, especially "wiremap it", you should be able to draw a portrait from memory in 3D. Otherwise you are lying to yourself.
As far as use - Tesla was said to have "run" schematics in his head to see what would break. I believe it. This allowed him to rapidly prototype.
Also, your working memory should be off the charts, which should be indicative of a very high IQ
I am aphantast, but I do not suffer from it.
When I am fully conscious, I have no inner vision, but I have vivd and colorfull dreams.
If I remember them, outside that twilight zone shortly before full awakening.
So I have an idea what it probably is like to have inner vision when fully awake.
Allthough there are some disadvantages, of course.
I admire people that are able to draw and paint based on their inner vision.
Much more important for me was the realization that I can evoke images, scenes, etc. in other people that trigger feelings in them. Which in turn can trigger actions or omissions.
Fear, joy, hate, love, disgust, lust.
Which they can't do to me, at least not just by invoking visual images in my mind through words.
Manipulative, but not manipulatable in this regard.
With time, that came in handy.
By the way, I am friends with a handful of people who suffer from schizophrenia.
They say they envy me a little because in their worst phases they wished they didn't have this movie in their head.
It repeats itself, over and over again.
And aphantasia is a spectrum, I have known people who describe rather dull, colorless inner visions and others who can sustain them only for short periods of time.
On the other hand, I met an artist who seemed to live in his own private vibrating Van Gogh painting.
Judging by his descriptions.
And of course, without DMT.
> Much more important for me was the realization that I can evoke images, scenes, etc. in other people that trigger feelings in them. Which in turn can trigger actions or omissions. Fear, joy, hate, love, disgust, lust. Which they can't do to me, at least not just by invoking visual images in my mind through words. Manipulative, but not manipulatable in this regard. With time, that came in handy.
I've also discovered this but I can only admit it to my closest friends else I'd be labeled a psychopath. There are things that trigger these kind of feelings in me, but it's more about situations than images and never in remembering something.
I do have a feeling that we might be more susceptible to do really nasty deeds if push comes to shove (Nazi Germany?) so I think it's something we need to be careful about as we can be manipulated into doing things that other people might find gut wrenching just thinking about.
You can also reverse it and perhaps claim that doctors can benefit from less visceral reaction to seeing blood/internals. Though of course it is a learned behavior anyone can get better at.
I read somewhere that déjà vu is when your two hemispheres get out of sync for a split moment and record the same input one after the other. But I don’t know if that’s true for sure. Or maybe they’re just rearranging the matrix.
Another theory I read was that some signal from your hippocampus (memory storage) fires, so that the rest of your brain believes erroneously that the current sensory input is coming from memory.
I believe this is the case. I have heard the same thing and it matches my experience every time I get déjà vu. I have this strong sense that what is happening has happened before, but I am unable to relate it to anything nor recall what should happen next.
The researchers used a magnetic resonance imaging technique to get their data:
> "It turned out that either visual stimulus—the grating or moving dots—resulted in the same patterns of neural activity in the visual cortex and parietal cortex. The parietal cortex is a part of the brain used in memory processing and storage. These two distinct visual memories carrying the same relevant information seemed to have been recoded into a shared abstract memory format. As a result, the pattern of brain activity trained to recall motion direction was indistinguishable from that trained to recall the grating orientation."
Alternative hypothesis: the technique used wasn't sensitive enough to distinguish between how the brain handled the different information types.
In my experience this plays out on multiple timescales. When you get older you start to have entire decades of life boiled down to the factual knowledge you gained plus a handful of episodic memories.
It's a good reminder to write shit down and take lots of mundane pictures. You don't realize until it's too late though.
Actually this generation will have a very different relationship with the past. Never before there was so many high resolution traces of your daily life.
Indeed. We take photos and videos for our son pretty much everyday. He would need a good chunk of his life to review all these if he wants when he grows up and moves to his own house.
There was a great article on this recently. The author made the point that if there was a day, week, or month where there were no backups or photos then you might put less value on that time of your life in retrospect. Conversely there might be a timespan where you took too many photos and might feel like there was more value to be had in that time.
Interesting. I forgot who said that in your best moments you don't have time for anything else. Indeed if you don't take pics it might just be because it was deeply interesting and not worth taking your smartphone out of your pocket.
A friend of mine from some time back chatted about this once. His take was that as you get older, your "mental models" grow and are able to cover larger parts of your day/week/month and your mind simply keeps the important parts but lets the rest fade.
When you're younger, those models are less complete and larger parts of your waking moments are needed to build the foundations of these models, so you feel like time is slower since so much more of your time is kept "fresh".
I'm probably butchering his take on it, but I blame my own mental models for compressing away the finer details!
> When you're younger, those models are less complete and larger parts of your waking moments are needed to build the foundations of these models, so you feel like time is slower since so much more of your time is kept "fresh".
As a corollary, if you want to keep on feeling young and feeling time pass slowly, you need to keep on incorporating "new" experiences into your life that extend (or change) your mental models.
The weird (scary?) point will be when we figure out how to subtly present adversarial information to the brain that will be coded in a way that collides with some target information to induce false recognition/ memories.
So I remember reading somewhere, probably on HN, that we don't remember real facts but instead we remember our last call of a particular memory. I've hijacked some unpleasant memories that way. I'll add some colors, a round ball bouncing, all kind of stuff that'll alter the memory. It doesn't make it totally disappear but it kinda smoothen it.
I have seen research where false memories were induced into people by photoshopping childhood images of those people into events that did not happen to them — and worse, in 16% of cases just by showing adverts of things that could not happen such as meeting Bugs Bunny at Disney World (wrong franchise): http://people.uncw.edu/tothj/PSY510/Loftus-Memory%20for%20Th...
Fascinating - thanks for the article. It's strange to think that however much we think that we're completely rational and can trust our own memories, we're more like malleable rationalizing machines.
Compression is a component of general intelligence. A few years ago I was very sceptical of machine learning ever leading to general intelligence. I've since changed my mind. There are a lot of parallels to this work and the concept of "embeddings" in machine learning.
Intelligence requires the ability to generalize. A prerequisite for generalization is the ability to take something high-dimensional and reduce it to a lower-dimensional representation to allow comparison and grouping of concepts.
We're doing this all the time. Take a pen for example: we're able to combine information from sight, touch, and sound. Through some mechanism, our brains reduce the multi-sensory information and create a consistent representation that is able to invoke past memories and knowledge about pens.
Our brains encode the embeddings in a very different way to deep learning neural networks, but the commonality is that both are able to compress data into a _useful_ representation. Note that as a result of this, the quality of the compression is important. Some forms of compression might be very efficient but they also tangle concepts together, resulting in loss of composability. The ideal compression (from an intelligence point of view) is both information efficient and maximally composable.
This also ties in to the cybernetic concept of the law of requisite variety, where adaptable entities need to be able to compress their sense-data about their environment into an internal model that corresponds in complexity to their need to act - this necessarily involves compression as the totality of reality is effectively infinite and can't fit between your ears.
There's also the Hutter prize that ties data compression directly to intelligence through Kolmagorov complexity.
Information and cybernetic theories cut pretty close to a general theory of intelligence in my opinion!
A nice definition of intelligence I've heard is exactly the ability to form models of the world with predictive power. And a model is essentially a compression of real-world data. Physical laws are a great example of this.
That's actually one of the flaws of Austrian economics. Mises just postulates that certain of his personality traits apply to all humans and ignores the personality traits of other humans. He is also confusing the informational theoretical world inside his mind with the harsh reality of the real world.
The word "capital depreciation" kind of summarizes all sorts of flaws. Gold is considered an ideal currency yet it suffers from no capital depreciation at all, while all other forms of physical capital depreciate. Once you have built all the houses, roads and schools, etc there is nothing left to build. If you happen to have built too many schools, then you are a fool and should convert the school back to gold by demolishing it and selling the scrap or in other words, idle capital, that is left to depreciate, must be eliminated.
When you think about this, this is secretly rewarding monopolies and all sorts of other nonsense that capitalism suffers from. For 10 people there are exactly 10 jobs. If a workaholic hoards a job by working 80 hours per week to retire early, then somehow there won't be enough "jobs" for everyone. This gives employers the upper hand during negotiations as employees start competing with each other instead of employers competing over employees. The current system is also dumb because it tries to represent capital depreciation through an increase in the price level i.e. 2% inflation targeting.
Slight tweak to this imo: models that can predict which new reframings/samples of current scientific-community-consensus SOTAs/benchmarks/datasets will disprove contemporary consensus is science :)
Can you recommend any philosophy of science (or life) treatises about this?
I long considered myself a Popperian. A few years ago, I decided that I'm a "Predictionist" (a placeholder made up word until I learn better). I'm struggling to figure out what that even means.
I still agree with Popper. Empathetically.
I'm just tired of arguing. I forfeit. I give up. I no longer believe that discourse is helpful, that people are persuadable, that we can share Truth.
Instead, I just want to know the predictive strength of someone's Truth.
For example:
The Earth is flat? Oh? Cool. Please, tell me, how does that Truth help me?
My research involve applying Popper's epistemology to natural language processing.
So I am quite involved in this.
As far as I can tell, almost all of what Popper tried to do with quantification measures of information are exactly what you are talking about.
In particular Conjectures and Refutations covers this really extensively so I'd recommend reading or re-reading that. Though Logic of Scientific Probability covers an early form. David Miller's Critical Rationalism covers it well too and some of it's problems.
I.e:
His notion (shared with positivists like Carnap and others) that science is a set of logical statements. A collection of statements is a theory, a theory entails a set of predictions which is called the information content of the theory (sometimes I(c) or C(I) in his notation).
If the I(c) > I(c') where c' is a competing theory then it is said to have more explanitory power. I.e. it makes more predictions.
This is part of his defnition of what makes a good explanation and what david desutch calls "hard to vary".
The other main part of the definitition is about whether these statements reflect Truth in anyway.. that is covered by his notion of verisimiltude or truthlikeness which is quantified as the degree to which the information content of a theory I(c) can be corroborated.
Both of these are essentailly "The predictive strength of someone's Truth"
The problem you and many other have probably encountered is the information content of an explanation is *intractable* it's an open set of statements which cannot full by fleshed out. So instead we can never have a perfect quantification of whether my theories or your theories are better... there may indeed be statements entailed by flat earth theory that have yet to be discovered and could indeed be more corroborated and provide better information content than a non-flat earth theory! Popper revels in this fact and fully embraces it.
Beyond Popper though, we need to understand more of the dynamics of "predictive strength" - I am finding causality a great source of literature that for which I would recommend Judea Perl and the Book of Why among other things.
For Philosophy of Science in particular there are ton's of great articles on stanford encyclopedia of philosophy about Explanation that go into this in depth - in fact the positivists like Carnap wrote amazing things about this which I would recommend.
I like to define intelligence as knowing data, but knowing data only creates idiot savants. What is lacking in AI today is artificial comprehension. What we're calling "artificial intelligence" lacks comprehension. Until the concepts handled by AI are composable, forming virtual operating logical mechanisms, and an AI comprehends by trial combinations of concepts (virtual operating logical mechanisms) we are only creating idiot savants incapable of comprehending what they do.
In the incredible story "Funes the Memorious" the eponymous Funes has an absolutely perfect memory, but is functionally mentally handicapped.
He can't even abstract to the existence of "trees" because he can recall and diff all of the details of every tree he's ever seen.
He can't even identify that he's seen a particular tree before, because he can diff how different it looked in a particular configuration of leaves and shadows because of different wind and cloud cover
Not capable of functioning independently or surviving for any long period unassisted but having a brain and cognition setup that allowed for amazing feats of mental wizardry. If you could have that ability and function normally in society you could do some astounding things.
I thought that the idea of Mentats (human computers from the Dune novel) were kind of ridiculous, but yeah, when you look at savants like Peek, makes you kind of wonder if such a thing would be possible.
Yeah, if you had the appropriate brain surgery as an infant and took highly specialized drugs to maintain or accelerate your mental functioning, you too could be like Kim Peek.
I like that Frank Herbert imagined that it would be possible and went with it before we had any real proof that it could be, and then he turned out to be right.
The human brain also forgets, something that may be a feature instead of a bug. Also, beyond compression––brains are simulation machines: imagining new scenarios. Curios to understand if ML provides anything analogous to simulation that isn't rote interpolation.
I don't think it's true. I can imagine a lot of aspects of systems around me I cannot possibly experience in any way, except maybe them leading to some outcome that I might experience as well. I sometimes do verify this experimentally, but that comes later.
GP said something about compression being a component of intelligence, parent said also brains simulate, then I said yeah i agree, and believe that the content of the simulation is the experience itself, whereas people often think there are 2 things, themselves and the world. I don't believe that at all. There is only one thing happening, the experience you are having now.
Ah ok, that doesn't work for me because I want to call dreams and hallucinations experiences as well. And those are disconnected from some idea of reality.
The primary aspect is first person perspective, not congruence with the external world.
>In fact, she was not very good at memorizing anything at all, according to the study published in Neurocase.[1] Hyperthymestic individuals appear to have poorer than average memory for arbitrary information.
I am quite a novice in ML topics, but isn’t this concept of simultaneously training a generator and validator sort of this?
I don’t know the exact term but I think of deep fake generators with an accompanying deep fake recognizer working in tandem bettering each other constantly?
Absolutely. Generative methods are all the rage now. Those methods work on learning information-rich representation spaces. You could argue it's still "interpolation" but instead of interpolating in data-space per se you are interpolating in representation-space.
I have always thought that the best measure of intelligence is compression of information. If you can create a smaller, abstract model that is still accurate despite a loss in details, then you are intelligent.
Interesting counterargument from AI researcher François Chollet (creator of Keras and one of the main contributors to TensorFlow): https://www.youtube.com/watch?v=-V-vOXLyKGw
How do theories such as "The 100th Monkey" as well as transferred information via DNA to offspring translate to ML|AI at all?
For example, couldn't a sufficiently developed AI modify some code/libraries it utilizes/learns from/creates, to ensure any new spawns of said AI/ML/Bot has the learned previous behaviors?
I doubt 100th Monkey will ever hit AI.
So that's an interesting aspect to the limits to AI 'evolving'
>able to compress data into a _useful_ representation. Note that as a result of this, the quality of the compression is important. Some forms of compression might be very efficient but they also tangle concepts together, resulting in loss of composability
---
I wonder if various factors inform how/what compression is used on a memory...
For example, a memory of putting the object back where it belongs/got it from vs the memory of a violent attack is through the lens of emotional (trauma) and thus the memories will be stored differently.
Its interesting in that I have been wanting to post an ASK HN on memory and dreams...
Now with this post, and your comment, I will post that.
---
The idea is that the surunding meta-information of a memory is important.
Lenses of senses that colour a memory are many, and individualistic.
i.e.
A person who is a psychopath, has an emotional block on the lens that they would see their actions through (remorse, guilt, empathy, etc) - thus they may not recall or RE-MIND themselves of an action/situation.
A memory that is laid with a sensuous experience, such as sex with someone you love/lust deeply may last a lifetime.
Certain things that one does/says can also lead to a lifetime of regret ; a cringe-worthy action/comment from decades ago can still haunt your thoughts.
---
I think the mystique btwn ML and biological memories is a really interesting space, as an ML|AI based system will never achieve the 100th monkey or DNA|biological transfer of information, but an approximation/facsimile based on evolved|updated libraries/files/code which are maintained exclusively by the AI entity will/does exist
Speculating here: if the brain really uses embeddings similar (in concept) to neural network embeddings, the mechanism could explain a lot of the peculiarities of the brain. Embeddings are naturally entangled, so are memories. For example, a specific smell can evoke a previous memory.
I saw a presentation once (It was not recorded, so I can't link it) that said the difference between real intelligence and artificial intelligence is the ability to quickly learn.
As an example, he said imagine if he invented a word. Poditon. And he told us that a poditon is any object that can fit on a podium. Instantly, you know whether or not any object can be classified as a poditon. A laptop is a poditon, but a car is not.
We are not at the stage where we can just tell a program "Any object that can fit on a podium is a poditon" and then ask "Is X a poditon?" and get a correct answer. And we probably won't be there for another couple decades.
So, I tried this out with GPT-3 in the OpenAI Playground. (The model called text-davinci-002.) My prompt looked like this:
Definition: A blorple is anything that is red and more or less round.
Question: Is a tennis ball a blorple?
Answer: No, because although tennis balls are round they aren't red.
Question: Is a cherry a blorple?
Answer: Yes, because cherries are red and approximately round.
Definition: A poditon is anything that can fit on top of a podium.
Question: Is a laptop computer a poditon?
Answer:
GPT-3 says: "Yes, because laptop computers are small enough to fit on top of a podium."
Is a normal-sized automobile a poditon? "No, an automobile is too large to fit on top of a podium."
Is the sun a poditon? "No, because the sun is too large to fit on top of a podium."
Is a human being a poditon? "Yes, because human beings are small enough to fit on top of a podium."
Is a house a poditon? "No, because a house is too large to fit on top of a podium."
While generating those answers it also spontaneously answered the question for tennis balls (yes) and books (yes).
> We are not at the stage where we can just tell a program "Any object that can fit on a podium is a poditon" and then ask "Is X a poditon?" and get a correct answer. And we probably won't be there for another couple decades.
If that presenter actually said that, they need to take a look at "Few shot learning in language models" (just Google the term and start reading the papers).
Already plugged this book elsewhere in the thread, you might be interested in "The Mind is Flat". One chapter of the book explores the concept you're describing. Our brain creates the illusion of a "full picture" when often our imagination and internal representation is quite sparse. I think that's one of the key impressive qualities of our brains and general intelligence. We only do the minimum necessary imagination and computation. As we explore a particular concept or scene, our brains augment the scene with more details. In other words, our mind is making it up as we go along.
Can you expand on this? Can you give an example of a kind of image it might work well with? I’ve always assumed the apparent detail of mental images was a kind of illusion, a bit like the illusion of detail outside the centre of the visual field.
I found out the separation when I can read Chinese perfectly fine but can't write it. When I'm writing, or trying to write it, I employ the technique you described. Though more often than not, I'd have tiny parts missing here and there.
I still think the approach outlined in the paper (using embeddings to map the physical world) is sound especially for the field of self-driving which is in dire need of generalization, but I've since changed my mind and currently do not believe we can achieve AGI (ever).
While embeddings are a great tool for compressing information, they do not provide inherent mechanisms for manipulating the information stored in order to generalize and infer outcomes in new, unseen situations.
And even if we would start producing embeddings in a way where they would have some basic understanding of the physical world, we could never achieve it to the level of detail necessary - because physical world is not a discrete function. Otherwise we would be creating a perfect simulation (within a simulation?). And the last time I was playing God, was in "Populous".
> I've since changed my mind and currently do not believe we can achieve AGI (ever).
Considering we (as in humans) developed general intelligence, isn't that already in contradiction with your statement? If it happened for us and is "easily" replicated through our DNA, it certainly can be developed again in an artificial medium. But the solution might not have anything to do with what we call machine learning today and sure we might go extinct before (but I didn't have the feeling that's what you were implying).
It's semantics at this point but we did not create ourselves, it was a complex process that took billions of years to create each one of us. Something being conceivable isn't the same as it being practically possible. I can imagine what you propose, but the same goes for traveling to distant stars or a time machine for going to the future. All perfectly possible in theory.
Intelligence is an abstract concept, it depends on what exactly one means by that. I have watched a rockets take off from earth. I have never seen a self-aware machine.
It is not a contradiction as I meant "achieving" in the context of creating it (through software).
The fact it happened to us is undeniable (from our perspective), but the how/why of it is still one of the many mysteries of the universe - one we will likely never solve.
I’m curious why you think that. Do you think it’s a fundamental problem with the discrete nature of traditional computers? Or a problem with scale and computational limits? If it’s the latter, if a hypothetical computer has unlimited time and memory capacity, why do you think AGI would remain impossible?
Machines are good at computation, which is not equal to reasoning, but rather a subset of reasoning.
And not only they are good at computation, but they are exceptionally good at it - I have no illusion of trying to compete with a machine doing square roots or playing chess. And increasingly harder problems are being expressed as computation problems, with more or less success - most famously probably self-driving.
But at the end of the day it feels like using an increasingly longer ladder to reach the surface of the Moon.
While imaginable, and every time we extend the ladder the Moon does get closer, it is fundamentaly impossible.
Ever since Gödel we’ve had a pretty convincing proof that there is nothing that you can do in terms of reasoning that can’t be expressed using computation. And since Turing we’ve got a framework that shows there’s nothing computable that you can’t compute using a universal computer.
So unless there’s something mystical beyond the realm of mathematics to ‘reasoning’ it can’t be a superset of computing.
If a finite amount of matter in a brain with a finite amount of energy can do it, then a universal computing machine with a finite amount of storage and a finite amount of time can do it.
There are actually a lot of well-defined things beyond the power of a Turing machine (for example a Turing machine plus a halting oracle that only works on Turing machines without a halting oracle) but in terms of finite amounts of electrons doing normal low-energy electronic stuff you are quite likely correct. Humanity may go beyond computability if as some papers have suggested quantum gravity requires solving uncomputable problems.
Even if our brains reason based on quirks of quantum mechanics (seems unlikely given the scale at which neurons operate), what stops us from creating non-biological machines that interact with QM in the same way to produce artificial reasoning?
I am not saying that anything more than a really big computer is necessary for reasoning, only that one day physics knowledge may reach beyond the Turing machine (quantum computing does not).
Do you believe human brains contain a halting oracle? Or the moral equivalent of one - something that enables our brains to accomplish some non computable reasoning task?
FWIW this is the same argument once made against human flight. In the late 19th century, there were a lot of debates in the form
> Clearly flight is possible, birds do it
> Sure but how/why is one of the many mysteries of the universe, one we will likely never solve.
"Man won't fly for a million years – to build a flying machine would require the combined and continuous efforts of mathematicians and mechanics for 1-10 million years." - NYT 1903
The real answer to how birds fly is that they're extremely light weight so that wing muscles can lift them. Common pigeons or seagulls only weigh about 2 or 3 pounds. The largest birds of prey top out around 18. Anything heavier is flightless. A 150-pound human isn't getting anywhere on wing muscle power.
The largest Pterosaur are estimated to have had wingspans of more than 9m and weigh up to 250kg (550 pounds) and we believe they were able to fly. [1]
But that's not the most relevant point here. The fact that humans did achieve to fly, but through a different method than birds is exactly a supporting argument that we might achieve AGI with a different approach than how our brains do it.
There are countless similar examples. We see a natural phenomenon, we know it's possible and we find a way to replicate the desired effect (not the whole phenomenon) artificially. I haven't heard anything here that it will be any different for intelligence, except that we don't know how yet.
The chain of reasoning that everything observable in nature is replicable by humans would also imply us being able to replicate creation of a living cell from non living material and then endow that organism with consciousness.
Further more it would also imply us being able to replicate birth of stars, black holes, and "the big bang" itself.
I am not qualified enough to speak if there is anything fundamentally impossible with all of these, but that would basically make human race "God".
> us being able to replicate creation of a living cell from non living material and then endow that organism with consciousness.
Afaik we are very close to artificially creating living cells. This is one recent example [1]. The consciousness part is similar to AGI.
> Further more it would also imply us being able to replicate birth of stars, black holes, and "the big bang" itself.
Some things might be a logistical challenge rather than one of knowledge. Fusion energy attempts to replicate the way stars produce energy and we already managed to replicate the effect, we are just (many years) shy of maintaining it to produce positive energy.
But you might be right and some things are impossible to replicate. I'm much more inclined to believe we can't replicate the big bang than general intelligence as mother nature replicates general intelligence millions of times each day. And by now we started to have a discussion about believes rather than knowledge, which is a much more healthy way to put it, as we indeed don't know.
> Afaik we are very close to artificially creating living cells. This is one recent example [1].
I beg to differ. It may look impressive on the surface, just like GPT-3 looks impressive on the surface, but is far from the real thing. It is just another extension of the ladder to the Moon.
The effort described in the article is nowhere near a living cell. It lacks protein building and DNA/RNA mechanisms. They basically describe a group of nanomotors.
I can recommend watching James Tour on this very topic [1] and Stephen Meyer on the related topic of intelligent design [2]. Those two lectures were eye-opening for me learning more about this field. Note: both of them are self-confessed theist scientists which to me did not represent a problem (my viewes are agnostic, and it only made it more interesting as you rarely get to hear different views about these matters than pop-sci).
> The consciousness part is similar to AGI.
It is not clear what you mean by that. One thing is to build computer code and then have it manifest 'intelligence'. Whole another thing is doing same with organic matter that can not be 'programmed', even if we knew how to do it (let alone there is no evidence that 'programming' is responsible for consciousness at all to begin with).
This is also known as 'hard problem of consciousness' and David Chalmers is considered one of the leading experts in the field [3]. Basically smartest scientists in the world are clueless about this and do not know even where to begin, in many ways similar to AGI.
> Some things might be a logistical challenge rather than one of knowledge. Fusion energy attempts to replicate the way stars produce energy and we already managed to replicate the effect, we are just (many years) shy of maintaining it to produce positive energy.
I can see why one can have this position where it seems like we are making progress in everything we talked about, but that is the main punchline of the ladder to the Moon analogy. Indeed it is imaginable, and indeed every step makes us closer. But it does not mean we will ever reach it.
I agree with you that the discussion ultimately boils down to direction and strength of one's beliefs.
I think AGI will remain out of reach. Even a simpler thing like level 5 self-driving, which is only like 0.3 AGI or something, will remain forever out of reach no matter how much compute we throw at it (though I also think that if we ever reach 0.3 AGI we will also reach 100%).
The reason is that the mundane world keeps surprising us everywhere we look and constantly keeps creating more questions than answers. Just look at the questions field of quantum mechanics is trying to tackle, but also every other field of research science - astronomy, genetics, biology, antropology even mathematics... Now imagine trying to keep up with all that - by writing code.
Thanks for your perspective. We’re still in disagreement but I wouldn’t bet on either side of the AGI debate with any significant conviction.
Embeddings are very good at a few things: combining concepts (addition), untangling commonalities (subtraction) and determining similarity between concepts (distance).
> While embeddings are a great tool for compressing information, they do not provide inherent mechanisms for manipulating the information stored
What are the manipulations you’re referring to? I would love to learn more. From my understanding, embeddings actually provide great generalisation. If you have a well conditioned embedding space then you can interpolate into previously unseen parts of that space and still get sensible results. That is generalisation to me. Many current ML methods do _not_ result in a fully meaningful embedding space but my hunch is that we will get there with future insights and advances.
> We’re still in disagreement but I wouldn’t bet on either side of the AGI debate with any significant conviction.
That is probably a superior position to hold. I am agnostic by nature, and interestingly this is one of the rare topics I've taken a hard position on. It could be a result of the years spent in the field but also some kind of bias.
> What are the manipulations you’re referring to?
Need to take a step back and mention that in the field of AI there is a great debate between symbolic and non-symbolic approaches. (and after decades spent with AI under symbolic approaches domination we are now in the golden age of non-symbolic AI; with symbolic starting to have a comeback. this podcast can be a good starting point to learn more https://lexfridman.com/gary-marcus/ - although I disagree with GM on many things - and this tweet for learning about symbolic making a comeback https://twitter.com/hardmaru/status/1470847417193209856)
Basically embeddings are "non-symbolic AI" (which is great and this is where their huge potential stems from), but the very way they are generated and then later utlized is completely "symbolic". Which means the the limits of embeddings is defined by the limits of (in this case human written) symbols used to define them. Hope that makes sense.
Could you please stop posting unsubstantive and/or flamebait comments to HN? You've been doing it a ton, unfortunately, and we have ban that sort of account. It's not what this site is for, and it destroys what it is for.
What we want here is thoughtful, curious conversation, not people bashing each other's comments or inflicting snark on each other or ideological talking points.
> A prerequisite for generalization is the ability to take something high-dimensional and reduce it to a lower-dimensional representation to allow comparison and grouping of concepts.
I've been thinking that it might actually the other way around - intelligence is taking lower dimensional data and being able to infer higher level representation in terms of context, meanings and other abstractions. I.e. understanding when a stop sign isn't a stop sign.
It's actually the compression that forces it to learn higher level concepts.
In your stop sign example, say we are trying to teach a visual model the difference between toy stop signs and real stop signs.
To train it you feed it a 3D model of the world and the actions a person takes in response (ie, ignoring toy stop signs but stopping for real ones). Once the embedding is well trained (with lots of data) if you then run it through something like UMAP to reduce the number of dimensions in the embedding from hundreds to 2 or 3 you'll see it has "discovered" the concept of "scale" - all the small toy stop signs will be clustered together and the real ones clustered elsewhere.
That generalisation forced by compression is where the abstraction of "scale" comes from.
(Of course in real life you'd use a more complex model than just an embedding for this, but in principle this is the idea).
>>> Intelligence requires the ability to generalize
Counterexample: I saw a TED talk video [1] by Dr. Grandin (autistic) telling that she is unable to generalize a "church" (starting at 02:50). I would say that she is extremely intelligent though
I think compression is a bad word or description. Another definition of intelligence is sometimes to differentiate essential from superficial information. Of course that often aligns with the application of compression of information.
Announcements like this seems so out in front of what we actually understand. It's not like we can take someone's brain and read memories from it, right?
That makes a lot of sense. One big result from a lot of the subliminal stimuli research scientists do is that nothing that doesn't enter your consciousness and get combined with your other sensory input streams get preserved by the brain for more than a second or so. As best we can tell conscious awareness has a far narrower bandwidth than your visual cortex so of course its dropping details.
The most impressive part here is the "decompression" imo. Computers are already being used to do stuff that's more or less similar (creating apps, 3D models, pictures, videos from code) but the speed at which a human brain does it is incredible.
It can be pretty inaccurate, though, adding extra objects/words/feelings/circumstances that literally were not there :D
> It turned out that either visual stimulus—the grating or moving dots—resulted in the same patterns of neural activity in the visual cortex and parietal cortex. The parietal cortex is a part of the brain used in memory processing and storage.
>These two distinct visual memories carrying the same relevant information seemed to have been recoded into a shared abstract memory format. As a result, the pattern of brain activity trained to recall motion direction was indistinguishable from that trained to recall the grating orientation.
>This result indicated that only the task-relevant features of the visual stimuli had been extracted and recoded into a shared memory format. But Curtis and Kwak wondered whether there might be more to this finding.
——
That is outrageously bad logic and is basically assuming your conclusion. This is not good science.
The subject is given two tasks requiring working memory. The researchers observe activity in the parietal and visual cortices via fMRI, and find the neural activity between the two tasks is indistinguishable. And conclude
> ... distinct visual stimuli (oriented gratings and moving dots) are flexibly recoded into the same WM format in visual and parietal cortices when that representation is useful for memory-guided behavior.
Seems a pretty big leap to me.
I'm not a neuroscientist, and fMRI is amazing. But I think there's more handwaving about how 'thoughts' and 'memories' are 'encoded' as if the brain were a piece of electronics we fully understood.
There's no magic -- everything we think has to happen at some physical level, but I think there is a generation of neuroscientists who are fooling themselves by projecting a reductionist mental(?!) model of how the brain works that is as yet unjustified, and interpreting all of their results in the light of that model.
I read it as saying they actually got topographical line-like structures in the MRI, similarly to the well established result that there are typographically-arranged visual neurons that essentially light up like pixels in response to a scene.
So if that’s right then they are actually measuring a spatial encoding. Anyway, it’s just the abstract so I could be completely wrong.
Isn't this easy to visualize? Think of driving down the highway. There'll be certain features that you remember in more detail than others. Trees, for example, will generally just be trees with the exception of a few "interesting" ones.
Most useful thing I've ever learned about memory: every time you recall a memory, you change it. Memory is not a fixed or static 'historical record'; ultimately, it's unreliable.
346 comments
[ 3.7 ms ] story [ 278 ms ] thread> These two distinct visual memories carrying the same relevant information seemed to have been recoded into a shared abstract memory format. As a result, the pattern of brain activity trained to recall motion direction was indistinguishable from that trained to recall the grating orientation.
Isn't the alternative explanation that our tooling for inspecting the brain at work abstracts too much detail away for us to be able to tell the difference?
But the paper is explicitly looking at the representation in working memory; so two obvious possibilities are, one that the "orientation" and "dotness vs. lineness" attributes are being decoupled and stored separately in working memory (different "registers" if you will). Or the "dotness / lineness" is getting stored somewhere else (not working memory, some other memory system) because it's not "behaviorally relevant" (i.e. relevant to the task that the participant is attending to while creating the working memory). I'd guess at the first because my impression was that essentially everything that makes it into episodic memory starts in working memory, but I'm not a neuroscientist.
I think the OP is getting way ahead of itself with "The findings suggest that participants weren’t actually remembering the grating or a complex cloud of moving dots at all.". The paper is making a much more modest claim that "direction" is recorded in the same underlying way, specifically during a task where you're being asked to recall direction. It's completely possible that this intermediate/common representation would not be generated if you're just looking at the pattern and not performing a task related to direction.
I couldn't find the full paper on SciHub, just the abstract linked in the OP: https://pubmed.ncbi.nlm.nih.gov/35395195/. I'd hope the full paper talks about all this in more detail.
The work proposes that because responses to difference visual stimuli show the same haemodynamic spatiotemporal response in memory areas, the actual memories must share a common representation. This is debatable since we know that localized spatiotemporal responses reflect the variability in the vascular tree and blood flow and volume, regardless of the experiment at hand [see 1 for a discussion]. To dare to claim that effects could be neuronal one needs to run a bunch of extra control experiments (vascular reactivity mapping / breathhold hypercapnic challenges, resting-state imaging as a control dataset), none of which were conducted in the proposed work.
This is known to MR physicists, but hasn't clicked in yet within departments of cognitive neuroscience.
[1] https://www.frontiersin.org/articles/10.3389/fnins.2020.5960...
But my memory works this way. A summary "party at so-and-so's house, weather was nice, overall vibe was ___". The rest is context. You know what the house/backyard is like, you know the general feel of that time of year, you know the crowd that usually comes, you can easily synthesize details like the smell of the BBQ and the taste of the food... build up a complete "memory" from stuff that could be summarized in a paragraph of text plus generic (not specific to one memory episode) context.
I can build up a relatively vivid mental image of my walking route to school (from the bus terminal) over 40 years ago. Is it accurate? Who cares. As long as no detailed record exists to compare it to that would reveal the "lossy compression".
But it was possible, with a bit of head scratching, to walk the route just from memory.
I think a large part of it is just that you store what is important to you. To me the day-to-day politeness is just filler. I don't care if they had black coffee or a latte. If someone was struggling with something and poured out their heart over multiple conversations, I'm going to remember what arguments and concerns they had and the mental model I built up around that situation. The filler is just unimportant and doesn't stick around.
My wife is the opposite. Signs of weakness are an embarrassment to be forgotten. She lives for the day-to-day.
I remember what computers people own, what the bread-winners of the family do for a living, and an insane amount of TV and movie trivia.
It's all about what captures your interest.
I would also point out that I was replying to a comment that was an empirical observation. My comment highlighted that their empirical observation corresponds to the pre-existing Myers-Briggs theory (which suggests that other people have previously had similar observations).
Test have shown that MBTI results do correlate with Big Five personality traits[1] which are generally regarded as valid.
See https://web.archive.org/web/20121011195955/http://leadu-libr...
[1] https://en.wikipedia.org/wiki/Big_Five_personality_traits
Regarding my personal experience with it: I have personally found it highly useful as a predictive model of behaviour. It's the only thing I've found that allows me to explain the aspects of people's personalities that aren't easily explained by their environment or life experiences. And by combining with JTT with an understanding of someone's life experience and environment I've found that I can pretty much always find a satisfying explanation for someone's behaviour in a way that I am not able to do with either one individually.
Regarding its validity:
- Firstly, there's no need to tiptoe around the subject if you think it's bullshit. I won't be offended, and I can totally understand why might be skeptical given the experimental evidence that exists so far.
- Having said that, I would like to challenge to idea that we ought to expect JTT to be experimentally validated given how limited our current ability to inspect the brain is. Specifically (and unlike other models of personality) the theory's primary claims are that there are certain internal thinking processes, which of course we currently have no way of observing directly, and which will not necessarily correspond straightforwardly or 1:1 with observable behaviours (environmental and life experience factors being huge confounders).
- Given this I find it entirely unsurprising that experimental designs which rely on numerical scorings of observed behaviours fail to find an effect
- I think it no coincidence that this theory came out of clinical psychology, because you need to be able to control for the environmental and life-experience factors in order to be able to see the other pattern that is (well ok, might be) sitting there beneath them. And therapeutic relationships which continue over a number of years are one of the only scenarios where that context is available outside of close personal relationships.
As we all know "all models are wrong but some are useful"
Surely many people do. Otherwise you wouldn't have all these biographies and non-fiction books packed with conversations people have managed to recall in a level of detail enough to not get sued. I can barely remember a line of conversation from this week, let alone important ones from years ago, so I always assumed most/many people can remember conversations to some reasonable level in a way that I cannot.
The author may then present this in the form of quoted speech in order to make it more vivid and compelling for today's reader, but it rarely corresponds to a precise transcript of the original conversation.
And, is she right or does she likes to be right?
Of course the engineer is tempted to test this by secretly recording a conversation and trying to trip up the perfect rememberer, a year later. But the non-geek life experience accumulated says don't go there.
No, that is perfectly normal, and it starts much earlier, weeks sometimes days after leaving the code. Depending on its complexity and level of its abstraction.
You mentally build something highly abstract without much emotional or bodily bond. Your brain has not much incentive to rememeber it.
For example, by character peculiarities, new experiences, current circumstances, etc. Often they are made up on a whim, without the remembering person being aware of it.
So in a sense, memories have a past and a history.
Not only that, but by recalling and rebuilding memories, how gaps are filled in depends on your current mental state. For example, if I'm feeling depressed and brooding over past social interactions, I'll likely imagine people having meaner expressions or saying harsher things than they did. The big problem is that your memory of the event is "written over" based on the rebuilt memory. Again, only the seemingly important bits, but people are more likely to remember emotionally strong portions. Like those imagined harsh words.
I realized I was doing this when I thought a professor strongly disliked me, avoided his classes for a couple years, but then found him pleasant. My depression and social anxiety had warped my memories over the years. Being aware that this happens really helps. I trust negative parts of memories less, and I consciously stop myself when I start to brood (or at least, have fun with a puzzle while thinking back on things).
I listen to a lot of music and use it (like many) to index various stages of life - i even have playlists by year to help facilitate this.
The problem occurs when I listen back to music i was listening to in, say, 2008. All of a sudden, i’m transported back. But, each time i do this, the effect wears off a little, because their is some - let’s call it - “meta data” being written from the current moment in time im listening from which is adding new color to the initial index.
This effect has been studied and noted somewhere before, but i’ll have to dig up when i’m not on mobile.
https://www.wnycstudios.org/podcasts/radiolab/episodes/91569...
One of the researchers interviewed on that episode was Karim Nader, who focuses on what he calls memory reconsolidation:
https://www.mcgill.ca/psychology/karim-nader
Or at least they think they do.
It's not just decades old memories. Memory of recent events is likely to be suspect. Which is an issue for the legal system because it relies so heavily on eyewitness testimony.
https://theconversation.com/new-research-reveals-how-little-...
Not only is human memory unreliable, it is also malleable. And if we are just a collection of our memories, then who are we really?
Sorry I don’t have any sources, I’m just a casual reader in this space.
I guess the brain probably is dealing (in a hugely non-mathematical way -- it is just an analogy!) with a similar sort of thing. Somehow we pick some memories to keep in great detail -- either because they seem to be very valuable, or because they just seem to compress nicely.
It is a bit funny that one name for this sort of thing is a "singular experience."
[0] https://en.wikipedia.org/wiki/Aphantasia
[1] https://medicine.exeter.ac.uk/media/universityofexeter/medic...
Imagine you sit at your desk all day answering emails. Emails come in, responses go out. Except when you step back from the desk, it's just a black void. Information from your eyes? That's just an email saying what grandma looks like. Pain in the leg? Re: URGENT. Nothing exists beyond the emails. The emails are reality. The brains representation language is the same as it's actual language. Why have more than one language?
Seems to be along the lines of lucid dreaming, with a vast difference in degree. Sometimes as I'm falling asleep I can see vivid scenes or objects that I can--to minor degrees--play with for a short time before I either fall asleep or wake up, then it's gone.
His feats are perhaps even more remarkable if he didn't have any superpower abilities such as this.
most of us here are programmers and do this on a daily basis. somebody describes "A GraphQL API driven by a clojure back end connected to a postgres database" and to a layperson that looks like either a bunch of nonsense words or maybe a few boxes, clouds, and arrows. but to you and me we can visualize the individual lines of code, configurations, functions, and infrastructural requirements behind that simple sentence.
same with an electrical engineer/inventor in their domain.
Anyway this might just be specific to me but something to think about.
While I was in jail I managed to imagine images in vivid color and detail several times, but never on demand. It would happen randomly when I was just lying, bored, thinking about stuff. Each time was one of the most amazing experiences of my life. It was like going from being a caveman looking at scratchings on a wall, to straight into a 3D IMAX movie. The quality was that spectacular, with bright colors and completely 3-dimensional. Since I left jail it hasn't happened again a single time.
I have to wonder if that's how others see inside their minds? If so, I am in awe.
To be honest, it doesn't feel like such a handicap to my life that I would start submitting myself to clinical trials. If the worst to come out of my mis-diagnosis is this post, I can live with it.
For example, I can imagine multiple 3D shapes at one time, rotate them, keep track of which direction a face is pointing on each one, etc.
However, I don’t really “see” any image. It’s more like a feeling of seeing it. Now I’m wondering to what extent other people actually see things they imagine…
Look at this thing, and describe what you see
Now, close your eyes and imagine another thing and describe it
comparison of imagery in reality and visualised. This presupposes people describe things visually even when directly seeing them, and not in other modes (texture, sound, smell , etc)
I can picture anything that I want. Movie scenes with my friends faces in them. I always thought everybody could do this. If it's somewhat unique, can I use it for something?
As far as use - Tesla was said to have "run" schematics in his head to see what would break. I believe it. This allowed him to rapidly prototype.
Also, your working memory should be off the charts, which should be indicative of a very high IQ
Allthough there are some disadvantages, of course. I admire people that are able to draw and paint based on their inner vision.
Much more important for me was the realization that I can evoke images, scenes, etc. in other people that trigger feelings in them. Which in turn can trigger actions or omissions. Fear, joy, hate, love, disgust, lust. Which they can't do to me, at least not just by invoking visual images in my mind through words. Manipulative, but not manipulatable in this regard. With time, that came in handy.
By the way, I am friends with a handful of people who suffer from schizophrenia. They say they envy me a little because in their worst phases they wished they didn't have this movie in their head. It repeats itself, over and over again.
And aphantasia is a spectrum, I have known people who describe rather dull, colorless inner visions and others who can sustain them only for short periods of time. On the other hand, I met an artist who seemed to live in his own private vibrating Van Gogh painting. Judging by his descriptions. And of course, without DMT.
I've also discovered this but I can only admit it to my closest friends else I'd be labeled a psychopath. There are things that trigger these kind of feelings in me, but it's more about situations than images and never in remembering something.
I do have a feeling that we might be more susceptible to do really nasty deeds if push comes to shove (Nazi Germany?) so I think it's something we need to be careful about as we can be manipulated into doing things that other people might find gut wrenching just thinking about.
Of course, for hell on earth, you need both.
The brain is nothing like a computer, so the hash table analogy is almost certainly inaccurate, but it's a funny idea.
Short term memory operates radically different than long term memory
Co processors doing specialized processing, with a limited ability of other processors to do it
Each having their own currently unknown instruction set
Bus between the processor with varying bandwidth constraints
Modules for processing certain kinds of input, maybe some completely vestigial after deprecation
> "It turned out that either visual stimulus—the grating or moving dots—resulted in the same patterns of neural activity in the visual cortex and parietal cortex. The parietal cortex is a part of the brain used in memory processing and storage. These two distinct visual memories carrying the same relevant information seemed to have been recoded into a shared abstract memory format. As a result, the pattern of brain activity trained to recall motion direction was indistinguishable from that trained to recall the grating orientation."
Alternative hypothesis: the technique used wasn't sensitive enough to distinguish between how the brain handled the different information types.
It's a good reminder to write shit down and take lots of mundane pictures. You don't realize until it's too late though.
i'm sorry we don't have the technology for that yet
It is really nice to have a ritual where you review your day before bed; and after doing this for 5 years, I have an awesome photo journal.
When you're younger, those models are less complete and larger parts of your waking moments are needed to build the foundations of these models, so you feel like time is slower since so much more of your time is kept "fresh".
I'm probably butchering his take on it, but I blame my own mental models for compressing away the finer details!
As a corollary, if you want to keep on feeling young and feeling time pass slowly, you need to keep on incorporating "new" experiences into your life that extend (or change) your mental models.
We already have a few examples of adversarial information for humans: optical illusions being the most widely discussed.
I have seen research where false memories were induced into people by photoshopping childhood images of those people into events that did not happen to them — and worse, in 16% of cases just by showing adverts of things that could not happen such as meeting Bugs Bunny at Disney World (wrong franchise): http://people.uncw.edu/tothj/PSY510/Loftus-Memory%20for%20Th...
1. you sense an experience,
2. retain it (percepts),
3. when repeated, you extract the common denominators to form a concept (something you can recall and communicate).
Intelligence requires the ability to generalize. A prerequisite for generalization is the ability to take something high-dimensional and reduce it to a lower-dimensional representation to allow comparison and grouping of concepts.
We're doing this all the time. Take a pen for example: we're able to combine information from sight, touch, and sound. Through some mechanism, our brains reduce the multi-sensory information and create a consistent representation that is able to invoke past memories and knowledge about pens.
Our brains encode the embeddings in a very different way to deep learning neural networks, but the commonality is that both are able to compress data into a _useful_ representation. Note that as a result of this, the quality of the compression is important. Some forms of compression might be very efficient but they also tangle concepts together, resulting in loss of composability. The ideal compression (from an intelligence point of view) is both information efficient and maximally composable.
There's also the Hutter prize that ties data compression directly to intelligence through Kolmagorov complexity.
Information and cybernetic theories cut pretty close to a general theory of intelligence in my opinion!
The word "capital depreciation" kind of summarizes all sorts of flaws. Gold is considered an ideal currency yet it suffers from no capital depreciation at all, while all other forms of physical capital depreciate. Once you have built all the houses, roads and schools, etc there is nothing left to build. If you happen to have built too many schools, then you are a fool and should convert the school back to gold by demolishing it and selling the scrap or in other words, idle capital, that is left to depreciate, must be eliminated.
When you think about this, this is secretly rewarding monopolies and all sorts of other nonsense that capitalism suffers from. For 10 people there are exactly 10 jobs. If a workaholic hoards a job by working 80 hours per week to retire early, then somehow there won't be enough "jobs" for everyone. This gives employers the upper hand during negotiations as employees start competing with each other instead of employers competing over employees. The current system is also dumb because it tries to represent capital depreciation through an increase in the price level i.e. 2% inflation targeting.
I long considered myself a Popperian. A few years ago, I decided that I'm a "Predictionist" (a placeholder made up word until I learn better). I'm struggling to figure out what that even means.
I still agree with Popper. Empathetically.
I'm just tired of arguing. I forfeit. I give up. I no longer believe that discourse is helpful, that people are persuadable, that we can share Truth.
Instead, I just want to know the predictive strength of someone's Truth.
For example:
The Earth is flat? Oh? Cool. Please, tell me, how does that Truth help me?
As far as I can tell, almost all of what Popper tried to do with quantification measures of information are exactly what you are talking about.
In particular Conjectures and Refutations covers this really extensively so I'd recommend reading or re-reading that. Though Logic of Scientific Probability covers an early form. David Miller's Critical Rationalism covers it well too and some of it's problems.
I.e: His notion (shared with positivists like Carnap and others) that science is a set of logical statements. A collection of statements is a theory, a theory entails a set of predictions which is called the information content of the theory (sometimes I(c) or C(I) in his notation).
If the I(c) > I(c') where c' is a competing theory then it is said to have more explanitory power. I.e. it makes more predictions.
This is part of his defnition of what makes a good explanation and what david desutch calls "hard to vary".
The other main part of the definitition is about whether these statements reflect Truth in anyway.. that is covered by his notion of verisimiltude or truthlikeness which is quantified as the degree to which the information content of a theory I(c) can be corroborated.
Both of these are essentailly "The predictive strength of someone's Truth"
The problem you and many other have probably encountered is the information content of an explanation is *intractable* it's an open set of statements which cannot full by fleshed out. So instead we can never have a perfect quantification of whether my theories or your theories are better... there may indeed be statements entailed by flat earth theory that have yet to be discovered and could indeed be more corroborated and provide better information content than a non-flat earth theory! Popper revels in this fact and fully embraces it.
Beyond Popper though, we need to understand more of the dynamics of "predictive strength" - I am finding causality a great source of literature that for which I would recommend Judea Perl and the Book of Why among other things.
For Philosophy of Science in particular there are ton's of great articles on stanford encyclopedia of philosophy about Explanation that go into this in depth - in fact the positivists like Carnap wrote amazing things about this which I would recommend.
He can't even abstract to the existence of "trees" because he can recall and diff all of the details of every tree he's ever seen.
He can't even identify that he's seen a particular tree before, because he can diff how different it looked in a particular configuration of leaves and shadows because of different wind and cloud cover
Not capable of functioning independently or surviving for any long period unassisted but having a brain and cognition setup that allowed for amazing feats of mental wizardry. If you could have that ability and function normally in society you could do some astounding things.
A great story about a mega-savant who can function is "Understand" by Ted Chiang, if you're interested
I like that Frank Herbert imagined that it would be possible and went with it before we had any real proof that it could be, and then he turned out to be right.
GP said something about compression being a component of intelligence, parent said also brains simulate, then I said yeah i agree, and believe that the content of the simulation is the experience itself, whereas people often think there are 2 things, themselves and the world. I don't believe that at all. There is only one thing happening, the experience you are having now.
I can't follow what you are trying to say.
The primary aspect is first person perspective, not congruence with the external world.
Having it is a premium feature.
https://en.m.wikipedia.org/wiki/Hyperthymesia
So no, not premium. A trade off.
I don’t know the exact term but I think of deep fake generators with an accompanying deep fake recognizer working in tandem bettering each other constantly?
For example, couldn't a sufficiently developed AI modify some code/libraries it utilizes/learns from/creates, to ensure any new spawns of said AI/ML/Bot has the learned previous behaviors?
I doubt 100th Monkey will ever hit AI.
So that's an interesting aspect to the limits to AI 'evolving'
---
I wonder if various factors inform how/what compression is used on a memory...
For example, a memory of putting the object back where it belongs/got it from vs the memory of a violent attack is through the lens of emotional (trauma) and thus the memories will be stored differently.
Its interesting in that I have been wanting to post an ASK HN on memory and dreams...
Now with this post, and your comment, I will post that.
---
The idea is that the surunding meta-information of a memory is important.
Lenses of senses that colour a memory are many, and individualistic.
i.e.
A person who is a psychopath, has an emotional block on the lens that they would see their actions through (remorse, guilt, empathy, etc) - thus they may not recall or RE-MIND themselves of an action/situation.
A memory that is laid with a sensuous experience, such as sex with someone you love/lust deeply may last a lifetime.
Certain things that one does/says can also lead to a lifetime of regret ; a cringe-worthy action/comment from decades ago can still haunt your thoughts.
---
I think the mystique btwn ML and biological memories is a really interesting space, as an ML|AI based system will never achieve the 100th monkey or DNA|biological transfer of information, but an approximation/facsimile based on evolved|updated libraries/files/code which are maintained exclusively by the AI entity will/does exist
As an example, he said imagine if he invented a word. Poditon. And he told us that a poditon is any object that can fit on a podium. Instantly, you know whether or not any object can be classified as a poditon. A laptop is a poditon, but a car is not.
We are not at the stage where we can just tell a program "Any object that can fit on a podium is a poditon" and then ask "Is X a poditon?" and get a correct answer. And we probably won't be there for another couple decades.
Is a normal-sized automobile a poditon? "No, an automobile is too large to fit on top of a podium."
Is the sun a poditon? "No, because the sun is too large to fit on top of a podium."
Is a human being a poditon? "Yes, because human beings are small enough to fit on top of a podium."
Is a house a poditon? "No, because a house is too large to fit on top of a podium."
While generating those answers it also spontaneously answered the question for tennis balls (yes) and books (yes).
Decades sure do go by quickly, these days.
But yeah, that's pretty amazing.
If that presenter actually said that, they need to take a look at "Few shot learning in language models" (just Google the term and start reading the papers).
Clearly, much less information is stored than the whole image... yet my mind DALL-E style fills in the gaps and "sees" a map.
"Towards conceptual generalization in the embedding space" https://arxiv.org/abs/1906.01873
I still think the approach outlined in the paper (using embeddings to map the physical world) is sound especially for the field of self-driving which is in dire need of generalization, but I've since changed my mind and currently do not believe we can achieve AGI (ever).
While embeddings are a great tool for compressing information, they do not provide inherent mechanisms for manipulating the information stored in order to generalize and infer outcomes in new, unseen situations.
And even if we would start producing embeddings in a way where they would have some basic understanding of the physical world, we could never achieve it to the level of detail necessary - because physical world is not a discrete function. Otherwise we would be creating a perfect simulation (within a simulation?). And the last time I was playing God, was in "Populous".
Considering we (as in humans) developed general intelligence, isn't that already in contradiction with your statement? If it happened for us and is "easily" replicated through our DNA, it certainly can be developed again in an artificial medium. But the solution might not have anything to do with what we call machine learning today and sure we might go extinct before (but I didn't have the feeling that's what you were implying).
Flight had. Intelligence has.
We have interstellar objects intrude and fly by our solar system all the time, on several occasions.
The fact it happened to us is undeniable (from our perspective), but the how/why of it is still one of the many mysteries of the universe - one we will likely never solve.
And not only they are good at computation, but they are exceptionally good at it - I have no illusion of trying to compete with a machine doing square roots or playing chess. And increasingly harder problems are being expressed as computation problems, with more or less success - most famously probably self-driving.
But at the end of the day it feels like using an increasingly longer ladder to reach the surface of the Moon.
While imaginable, and every time we extend the ladder the Moon does get closer, it is fundamentaly impossible.
So unless there’s something mystical beyond the realm of mathematics to ‘reasoning’ it can’t be a superset of computing.
If a finite amount of matter in a brain with a finite amount of energy can do it, then a universal computing machine with a finite amount of storage and a finite amount of time can do it.
> Clearly flight is possible, birds do it
> Sure but how/why is one of the many mysteries of the universe, one we will likely never solve.
"Man won't fly for a million years – to build a flying machine would require the combined and continuous efforts of mathematicians and mechanics for 1-10 million years." - NYT 1903
NYT is just the epitome of the dunning-kruger in the west.
But that's not the most relevant point here. The fact that humans did achieve to fly, but through a different method than birds is exactly a supporting argument that we might achieve AGI with a different approach than how our brains do it.
There are countless similar examples. We see a natural phenomenon, we know it's possible and we find a way to replicate the desired effect (not the whole phenomenon) artificially. I haven't heard anything here that it will be any different for intelligence, except that we don't know how yet.
[1] https://en.m.wikipedia.org/wiki/Pterosaur_size
Further more it would also imply us being able to replicate birth of stars, black holes, and "the big bang" itself.
I am not qualified enough to speak if there is anything fundamentally impossible with all of these, but that would basically make human race "God".
Afaik we are very close to artificially creating living cells. This is one recent example [1]. The consciousness part is similar to AGI.
> Further more it would also imply us being able to replicate birth of stars, black holes, and "the big bang" itself.
Some things might be a logistical challenge rather than one of knowledge. Fusion energy attempts to replicate the way stars produce energy and we already managed to replicate the effect, we are just (many years) shy of maintaining it to produce positive energy.
But you might be right and some things are impossible to replicate. I'm much more inclined to believe we can't replicate the big bang than general intelligence as mother nature replicates general intelligence millions of times each day. And by now we started to have a discussion about believes rather than knowledge, which is a much more healthy way to put it, as we indeed don't know.
[1] https://www.nyu.edu/about/news-publications/news/2021/septem...
I beg to differ. It may look impressive on the surface, just like GPT-3 looks impressive on the surface, but is far from the real thing. It is just another extension of the ladder to the Moon.
The effort described in the article is nowhere near a living cell. It lacks protein building and DNA/RNA mechanisms. They basically describe a group of nanomotors.
I can recommend watching James Tour on this very topic [1] and Stephen Meyer on the related topic of intelligent design [2]. Those two lectures were eye-opening for me learning more about this field. Note: both of them are self-confessed theist scientists which to me did not represent a problem (my viewes are agnostic, and it only made it more interesting as you rarely get to hear different views about these matters than pop-sci).
> The consciousness part is similar to AGI.
It is not clear what you mean by that. One thing is to build computer code and then have it manifest 'intelligence'. Whole another thing is doing same with organic matter that can not be 'programmed', even if we knew how to do it (let alone there is no evidence that 'programming' is responsible for consciousness at all to begin with).
This is also known as 'hard problem of consciousness' and David Chalmers is considered one of the leading experts in the field [3]. Basically smartest scientists in the world are clueless about this and do not know even where to begin, in many ways similar to AGI.
> Some things might be a logistical challenge rather than one of knowledge. Fusion energy attempts to replicate the way stars produce energy and we already managed to replicate the effect, we are just (many years) shy of maintaining it to produce positive energy.
I can see why one can have this position where it seems like we are making progress in everything we talked about, but that is the main punchline of the ladder to the Moon analogy. Indeed it is imaginable, and indeed every step makes us closer. But it does not mean we will ever reach it.
I agree with you that the discussion ultimately boils down to direction and strength of one's beliefs.
[1] https://www.youtube.com/watch?v=zU7Lww-sBPg
[2] https://www.youtube.com/watch?v=7pk9oDrpf6k
[3] https://www.youtube.com/watch?v=C5DfnIjZPGw
Do you mean with embeddings as the approach, or in general?
The reason is that the mundane world keeps surprising us everywhere we look and constantly keeps creating more questions than answers. Just look at the questions field of quantum mechanics is trying to tackle, but also every other field of research science - astronomy, genetics, biology, antropology even mathematics... Now imagine trying to keep up with all that - by writing code.
Also, mastering these things would make us 'God'.
Embeddings are very good at a few things: combining concepts (addition), untangling commonalities (subtraction) and determining similarity between concepts (distance).
> While embeddings are a great tool for compressing information, they do not provide inherent mechanisms for manipulating the information stored
What are the manipulations you’re referring to? I would love to learn more. From my understanding, embeddings actually provide great generalisation. If you have a well conditioned embedding space then you can interpolate into previously unseen parts of that space and still get sensible results. That is generalisation to me. Many current ML methods do _not_ result in a fully meaningful embedding space but my hunch is that we will get there with future insights and advances.
That is probably a superior position to hold. I am agnostic by nature, and interestingly this is one of the rare topics I've taken a hard position on. It could be a result of the years spent in the field but also some kind of bias.
> What are the manipulations you’re referring to?
Need to take a step back and mention that in the field of AI there is a great debate between symbolic and non-symbolic approaches. (and after decades spent with AI under symbolic approaches domination we are now in the golden age of non-symbolic AI; with symbolic starting to have a comeback. this podcast can be a good starting point to learn more https://lexfridman.com/gary-marcus/ - although I disagree with GM on many things - and this tweet for learning about symbolic making a comeback https://twitter.com/hardmaru/status/1470847417193209856)
Basically embeddings are "non-symbolic AI" (which is great and this is where their huge potential stems from), but the very way they are generated and then later utlized is completely "symbolic". Which means the the limits of embeddings is defined by the limits of (in this case human written) symbols used to define them. Hope that makes sense.
if you say so
“A few years ago I was very sceptical of machine learning ever leading to general intelligence. I've since changed my mind.”
thanks for sharing
“…both [human brains and artificial neural networks] are able to compress data into a _useful_ representation.”
thanks for pointing this out
What we want here is thoughtful, curious conversation, not people bashing each other's comments or inflicting snark on each other or ideological talking points.
If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.
I've been thinking that it might actually the other way around - intelligence is taking lower dimensional data and being able to infer higher level representation in terms of context, meanings and other abstractions. I.e. understanding when a stop sign isn't a stop sign.
In your stop sign example, say we are trying to teach a visual model the difference between toy stop signs and real stop signs.
To train it you feed it a 3D model of the world and the actions a person takes in response (ie, ignoring toy stop signs but stopping for real ones). Once the embedding is well trained (with lots of data) if you then run it through something like UMAP to reduce the number of dimensions in the embedding from hundreds to 2 or 3 you'll see it has "discovered" the concept of "scale" - all the small toy stop signs will be clustered together and the real ones clustered elsewhere.
That generalisation forced by compression is where the abstraction of "scale" comes from.
(Of course in real life you'd use a more complex model than just an embedding for this, but in principle this is the idea).
This leads the compression to be overfitted to the learning environment. Not that our intelligence is entirely immune to that either
Counterexample: I saw a TED talk video [1] by Dr. Grandin (autistic) telling that she is unable to generalize a "church" (starting at 02:50). I would say that she is extremely intelligent though
[1] https://www.ted.com/talks/temple_grandin_the_world_needs_all...
It can be pretty inaccurate, though, adding extra objects/words/feelings/circumstances that literally were not there :D
>These two distinct visual memories carrying the same relevant information seemed to have been recoded into a shared abstract memory format. As a result, the pattern of brain activity trained to recall motion direction was indistinguishable from that trained to recall the grating orientation.
>This result indicated that only the task-relevant features of the visual stimuli had been extracted and recoded into a shared memory format. But Curtis and Kwak wondered whether there might be more to this finding.
——
That is outrageously bad logic and is basically assuming your conclusion. This is not good science.
https://pubmed.ncbi.nlm.nih.gov/35395195/
The subject is given two tasks requiring working memory. The researchers observe activity in the parietal and visual cortices via fMRI, and find the neural activity between the two tasks is indistinguishable. And conclude
> ... distinct visual stimuli (oriented gratings and moving dots) are flexibly recoded into the same WM format in visual and parietal cortices when that representation is useful for memory-guided behavior.
Seems a pretty big leap to me.
I'm not a neuroscientist, and fMRI is amazing. But I think there's more handwaving about how 'thoughts' and 'memories' are 'encoded' as if the brain were a piece of electronics we fully understood.
There's no magic -- everything we think has to happen at some physical level, but I think there is a generation of neuroscientists who are fooling themselves by projecting a reductionist mental(?!) model of how the brain works that is as yet unjustified, and interpreting all of their results in the light of that model.
So if that’s right then they are actually measuring a spatial encoding. Anyway, it’s just the abstract so I could be completely wrong.
Reminds me of another paper which has impacted me deeply:
https://josh.works/driven-by-compression-progress-novelty-hu...