A little thing I realize upon reading about all the brain study is how bad we are at reverse engineering a totally alien "computer".
In popular media, human is often depicted as some super scavenger that can exploit and adapt alien technology to their own in a matter of 1-2 lifetimes. And yet here we are in reality, after thousands of years with abundant and intimate access to one of the most efficient and most capable information processor known to us, and we are still absolutely clueless about how it work.
The closest thing we built to mimic it cost maybe a million times more energy and perform a fraction of the functions of the original one. So much for "reverse engineering".
If you are comparing the energy cost of LLMs, consider that the "training cost" of the human brain is the accumulated evolutionary activity of the last 4 billion years.
Human inference is much more efficient though - we run at about 100 Watts[0], probably a lot less for the thinking alone.
Remember, though that LLMs also need the accumulated evolutionary activity. They're not that interesting if you don't feed them human created content. Contrast that with the human computer that builds itself and then infers language and complicated physical and mental tasks.
Given I can do token generation on a 30 watt laptop, the language/Joule part at least seems to be the computer winning. (Separate to any discussion about quality).
Not so for images, as I can imagine visual scenes in what feels like real time, while even the fastest image generator I've seen was 200-300 ms at 512x512.
You're not really imagining visual scenes, the resolution just isn't there if you think about it. It's more like you're imagining low resolution encodings, which is definitely the smart and efficient way of doing it that evolution would choose over a wasteful full sim.
It sounds like you're subjective experience of visualization is wildly different than the average, then. I think I have an excellent imagination and visualization ability, but it doesn't compare to just looking out my eyeballs in terms of resolution or cognitive load.
Look up aphantasia on https://en.wikipedia.org/wiki/Aphantasia they have a sidebar describing the spectrum of the ability to visualize images. The highest level is considered photographic. The other side is the inability to visualize anything.
I mean, a bunch of rocks falling on a keyboard can also generate tokens. So can an autocomplete software running on a 10W phone. But the quality is so far apart to the point you would find most people do not consider them on the same categories of "inference" as an LLM, not to mention a real human's mental capacity.
By the way, rocks falling is potential energy turned to kinetic energy so it is quite a lot of energy used. Didn't fact check this but if you drop 10 100g rocks per second from 2m height it is 30w according to GPT.
Although I guess you could use one rock fall to generate A LOT of tokens if you don't use keyboard, but just use the variance data from falling rock to generate many tokens at once.
From the 100W consumed by the brain most of it is to keep the autonomic functions (respiratory, cardiac, digestion, processing senses [eyesight, hearing, touch, etc.]), actual thinking, executive function, language processing, etc. is a small percentage of the 100W in use.
But those 30 watts on the laptop also cover the screen, hard drive, wifi, etc.
If you want to count the brain alone, then it must be compared with just the silicon alone; if you want to count the whole body, then it must be compared with the whole machine.
If the question is "I want to get this much mental stuff done", then you need the brain and the cost of keeping the brain alive and connected to the outside world — which for a human brain is the body, and for a silicon "brain" the computer/network.
If you want a humanoid robot to do be a drop-in replacement for human labourers in all professions, then sure (modulo the open question of how much brain power all our jobs actually require) a whole robot body has to be ready and you need to compare that body to our body.
If you want a special purpose machine that fully automates one specific product line but doesn't need to move, the "brain" might just be punched cards: https://en.wikipedia.org/wiki/Jacquard_machine
For me, the only visual parts of a dream that have ever been wrong like that, has been text — but the last time I recall dreaming of text was the late 90s, so I don't know if that's still the case.
Odours are a thing I can't "visualise", and those are usually entirely absent from my dreams.
Imagination has been found to activate the same V1 cortex that vision does. The V1 cortex is retinotopic, so individual neurons (or at most small groups) act like pixels. There are approximately 280 million V1 neurons, so the neuroanatomical resolution of the visual cortex is much greater than 500x500.
This may not be the resolution we perceive as we imagine, but perception involves many additional systems and bottlenecks (notably, one can train oneself to have a more acute imagination, and there are people who report their imagination to be as acute as their vision).
I have no idea at which level of processing in which of my brain's visual areas my visual imagination is generated, however I can say that the subjective experience is "retina resolution".
Based on my own experience, it's only high resolution because when I focus on an area my mind generates the details. Until then, it's all broad strokes. JIT image generation in other words.
If a CPU has been developed so much within a few decades, then with the capabilities provided on ML/LLM/AI/and other-such-acronyms, in a million years we should at least have a "Lucy" (https://www.imdb.com/title/tt2872732/) evolved. In the meantime, I expect that in a few hundred years we will have a Borg Queen (perhaps without the interconnectivity).
We don't know how brains of our lab animals work either. We literally throw millions of chickens into meat shredders alive, daily. I don't think it's an ethics issue.
Biological systems are not "engineered" in the same way a computer is. Computer chips, despite all their complexity, are designed to be understood by humans - they are modular, and abstract-able to facilitate design. Biological systems do not have these properties unless it is advantageous for the task at hand. You often have systems where everything interacts with everything else in meaningful ways. So it isn't really accurate to directly compare reverse engineering technology to reverse "engineering" nature
I’ve always considered dreaming to be a daily disk defragmentation process where they bolted extreme Winamp visualizations on top to make it more interesting.
There is some recent research that some chemicals are flushed through the brain during sleep to remove byproducts. Why that can't happen without sleep is another interesting question.
> Why that can't happen without sleep is another interesting question.
I would guess that since sleep has been optimized by evolution for a long time, and evolution is so creative, that there are now countless dependencies on that time/phase separation.
But umbrella reasons probably include all of these:
1. To recognize unhealthy build of material for removal, you have to stall normal production. Otherwise production and removal would waste each other's efforts and pernicious build up would be undetectable.
2. Some cellular machinery may be used during normal and clean up operations, in different things. Evolution is great at reusing whatever is available.
3. Sleep involves coordinating many systems. Some subsystems might technically be capable of operating more flexibly, but coordinating all the systems together makes overall coordination simple, efficient and reliable.
4. Sleep isn't low energy behavior. Segmenting energy use between waking and sleeping maintains a lower metabolism envelope.
5. Dreaming is high level behavior. Can't high level think and dream at the same time.
6. Dreaming sensory free, motor free, operation is fundamentally at odds with normal active sensory and motor activity.
7. A billion random dependences on phase separation have accumulated over millennia, at the psychological, gene, epigenetic and chemical level, with no pre-ordaned rationale other than the separation was dependable for the other reasons given.
Intersting that octopuses, whose brains are an entirely separate "alien" evolutionary invention, also appear to dream. [0]
I wonder if this is heavily based on the fact that during sleep, all inputs are shut down. Like, you can't hear, you can't see, you can't feel, no smell, no taste. All senses to the outside world are off.
What is it about external inputs that makes the brain not be able to do it's sleep "duties"?
Is it that external inputs take up processing power that needs to be used during sleep instead?
But that's not true. The senses are not off as you wake up when light increases, when there is a loud sound or when there is a stron smell. In addition there are numerous examples that the sound of the alarm clock gets integrated into the dream, maybe as a telephone sound for example.
Because when we are awake our brains run "hot" producing chemical waste - evolving on a world experiencing predictable day and night cycles led to sleep - it is not like we could do anything without sunlight back then.
It was also dangerous to do anything i.r.o nocturnal predators.
> Why devote this kind of energy to the creation of wildly imaginative and highly emotional nocturnal experiences for an audience of one?
I think that this is a mistake. The audience size is greater than one.
The author also writes that dreams “[give] us outrageous scenarios so we can better understand the everyday, serving as an overnight therapist”. I believe this is a multiplayer game.
I am a psychotherapist (in training). When people report their dreams, they normally do not mention their own behaviour. However, when their behaviour is investigated it often looks odd given the “outrageous scenario” they are in. More specifically, the behaviour often reveals clear displays of their unconscious interpersonal anxieties that their waking symptomology only opaquely reveals. An example, might be someone who is “shy” when awake behaving in an outrageously passive way, when in a confrontation; a confrontation that their waking shyness - which might include symptoms which mean they avoid contact with others - would have meant that they would not have had. It is no surprise that people do not typically report their own behaviour within dreams, as that is the nature of unconscious beliefs - we think they are unremarkable and true. Often an outsider is required to notice that the behaviour is unusual given the “outrageous situation” - often the role of a therapist (who shares this theory on dreams).
In the paper linked below, I outline many examples of this, and discuss the surprisingly specific supporting neurology:
I also agree with those who say that sleep has multiple purposes. REM sleep typically starts four hours into a sleep cycle, after the “deep cleaning” work seems to be done. This would suggest that the brain is prioritising its basic functioning over the discovery of unconscious patterns, which is, I believe, a sensible trade-off. It is not imperative that we spot our unconscious patterns, though in the long run, it can make a big difference to our lives.
REM sleep takes place much sooner than 4 hours. It takes place at about 90 minutes of sleep. And, hopefully, the sleeper goes through several cycles throughout the night. I would be screwed if it took that long to get to REM as I wake up 4-5 times a night.
Sorry, you are correct. However when it first takes place, when there is more “deep sleep” work to be done, it is very brief, and becomes the more dominant aspect of sleep after that work has been done:
What I liked about the whole lucid dreaming scene is that they already figured this out, but are using different terminology to describe it.
You are exhausted, not relaxed, after a lucid dream. And you always need a second sleep session with REM sleep after that.
But you make a lot of progress on internal reflections, because your subconciousness fills in the emotional blind spots your alter ego cannot grasp (yet). I think that's why a lot of people in lucid dreams can relate so much with the 2 systems theory of Daniel Kaneman, and why there's so much debate about it.
> The rational, executive network in the brain is switched off, and the imaginative, visual and emotional parts are dialled way up. As a result, the dreaming mind is given free rein in a way that has no parallel in our waking lives. We couldn’t think this way when we are awake even if we tried.
I read this and felt compelled to say this reminds me of how psychedelics like LSD and psilocybin work :)
There's a decent hypothesis about the reason for dreaming, but it's a bit absurd to call it the whole point of sleep.
As a person with severe insomnia, often going several days without sleep, you absolutely start breaking down without sleep. Brain zaps, hearing things, seeing things, etc. I've read people have even died in severe cases.
So it feels like there's a medical necessity that's the whole point. Whether dreams are a neat side effect of said process or a purposeful side project of sorts is worthy of debate, however.
Are you sure about people dying from lack of sleep ? I read the opposite but I lack proof. To me sleeping is an essential "maintenance" of the body & the mind but that lack of sleep wouldn’t kill you in itself.
However there is no doubt sleeping is essential to psychological and physical health.
I also sometimes struggle with insomnia and I recommend reading "Why we sleep". It helped me understand sleep better than I would have thought and it actually helped me manage insomnia. The book is also actually really captivating which is a good thing given I thought I knew enough about sleep hygiene. I was so wrong.
Yes, creatures definitely die with sleep deprivation. Obviously you won't try that on humans, but for mices it's a known phenomena, they die quick enough. So sleep is necessary for basic body chemistry at least.
"...measurement of EE, based upon caloric value of food, weight, and wastes, indicated that all TSD rats increased EE, with mean levels reaching more than twice baseline values."
The causality of this condition is not the same. Prion issues cause both sleeplessness and death, which is not the same thing as sleep deprivation leading to death. Thanks for the link though, this is interesting.
Well, in many authoritarians and totalitarian police states, state security organs have practiced extreme sleep deprivation as a way of breaking people, and sometimes killing them too. (Also something practiced by more than a few democracies in their uglier moments and aspects).
One particular anecdote that I've read described in detail was how during the worst days of the Stalinist Soviet state, the NKVD (predecessor to the KGB but much more brutal and lethal) in its obsession to get signed confessions from arrested prisoners, would put them on what was commonly called the "conveyer belt"
Solzhenitsyn described it graphically in his book as a process by which a prisoner would be interrogated in turns by an unbroken chain of well-rested NKVD interrogators whose added-together job was to never once let that victim sleep.
This was described as killing many of these people within a month or less even though they were fed and given water (sometimes even by force with tubes down their throats, since the point was breaking them, not killing them directly through starvation or thirst). In some cases there were of course also severe beatings and other tortures that could have easily been partial causes of death, but I remember the author describing cases where victims' hearts would finally stop as they descended into slow madness even if they weren't beaten at all over the course of those sleepless weeks.
Extreme sleep deprivation can kill, i'm sure of it.
There is a genetic disease from an Italian family (Fatal Insominia) - they cannot sleep even with the help of medication and die a few months later due to heart failure.
If I wasn't sleeping, that would be an emergency. Not sleeping will end a life.
Whatever is going on, whatever I was doing or not doing. I would (try to) sleep in a random field without any "need xyz for 3+ days (other than water and a multivitamin, fasting)" even if it's prescription or life assumption whatever. Just to see. Just to sleep.
Unexpected and unnoticed things can dramatically effect sleep.
To me dreams feel more like a side effect of sleep. The fact that one is asleep is the restful part. Dreams do give some insight, but often times I don't dream at all (as in I don't remember them) and feel really well rested as well.
Sometimes I only get by on power naps and those power naps are pretty good too.
Same with my cat. Every time I go to sleep it will always pick me to snuggle with usually in between my head and elbow. Within a few mins it’s purring and its limbs twitch and I get poked by claws. It’s in deep sleep and seems happy.
I have some pretty wild dreams myself, and what’s wilder is that I have those dejavue (sp?) moments regularly irl.
Yes, but each language splits reality up differently, and with different implications.
I've noticed several native German speakers, when using English, would say things like "I spilled coffee on my table and now he is wet" — I don't think they were asserting that the table was objectively male, just because it was grammatically male.
You don't even know, if the dog exists. Might all be a dream or simulation after all.
But going for the most likely explanation, surely dogs can dream. They are also mammals, have quite similar brains and act similar in sleeping from the outside, so why shouldn't it be dreaming?
We only know that humans dream because we tell each other about our dreams. There's no evidence that any other species dreams, as they don't tell us about it and we don't have access to their qualia.
There's plenty of evidence, just not conclusive evidence perhaps. Birds and mammals both have REM sleep, the stage where we experience most dreams. This feels like slothful induction because we have almost as much evidence for animals dreaming as we do for other people dreaming, just lacking verbal reports.
They may act like humans do when we dream, and their brain activity while they sleep looks similar to ours when we dream. But neither of these facts allows us to conclude that they're actually dreaming.
The verbal reports are key to our shared understanding that most humans experience dreams when sleeping. We do not have this for dogs or any other species.
When it comes down to it, we can't say for sure that dream reports aren't the unconscious actions of philosophical zombies either. If we're willing to accept dream reports on the assumption that other people have conscious experiences, then it's not unreasonable to allow closely related animals the same.
> (...) given the many potential benefits of dreaming for our waking life, maybe it’s not the sleep we really need, but the dreams.
Waiting for the pot brigade to use that sentence as a conversation starter...
We need sleep to recover physically. Like really need it. Anyone who has done physical exercise or just work and tried to rest in other ways will learn that sleep is the best way to rest.
If subjectively experiencing a dream were "the whole point" of sleep, that would make it pretty hard to explain why total sleep deprivation is fatal. Clearly there are far more important things going on physiologically during sleep.
Can dreams be therapeutic? I'd wager that happens just about as often as they're psychologically harmful. And both of those outcomes are far less frequent than completely forgetting your dreams.
I've read we rehearse things in our dreams. If we played tennis one day, we simulate playing tennis in our dreams that night, and the next day we are better at tennis.
I also kind of wonder - does sleep have an evolutionary function to not move and not make noise during the night to patiently pass the time of night predators?
I can't remember where but I read somewhere a theory that says that most of life started sleeping, in the water, this was the default state.
At one point in time stuff started being awake and developed wake consciousness, and the evolutionary pressure was on the wakefulness of things, because if one thing is awake everything has to be eventually.
It made me think - why are we thinking our wake life is real life? When we're asleep - we're sure the dream world is the real world.
It's like one life is not enough from an evolutionary standpoint. You have to have 2 lives, interconnected on 2 different dimensions, alternating day and night.
> When we're asleep - we're sure the dream world is the real world
I don’t think this is true, or at least is not universally true. People often describe dreamlike states as being surreal or just off that we suspect we are in a dream. This is a common trope in movies.
Link is broken (has an extra p at the end, sleepp rather than sleep) and ends up in a page saying "Sorry – we haven’t been able to serve the page you asked for.", here is it fixed
Looks like The Guardian may have had a typo (-sleepp) in their url, fixed it, and didn't redirect traffic from the typo to the new url. Either way here is the working url:
One thing I believe is that the less you're involved in fantasy, the more realistic your dreams are. So if you're a simple, practical man without beliefs in supernatural, your dreams will be about daily life; but if you read fantasy and have superstitious beliefs and watch a lot of movies, you will have more complex and bizarre dreams. I mean, it's hard to dream about something you haven't encountered.
Yes, I know what you mean, but you can say no to movies, social media, tv, and choose to live in the countryside for example (of course that depends on your financial situation and where you live). That alone makes a big difference.
Dreaming stopped for me after being put on Trazodone. In some the medication induces nightmares or vivid dreams, not for me. I can testify that dreams contribute a lot to the perceived quality of sleep. In recent months I've had a few very short dreams. They're dull, bland, nothing much happens. I feel robbed of a part of my life. I took Trazodone for about a year, now some 8 years ago.
Interesting. I take Trazadone to reduce the number of nightly awakenings. I still get up 3-4 times a night but at least that gives me the opportunity to make it through a sleep cycle without waking up.
It's a pretty uncommon antidepressant, you usually only get it because you have some other stuff going on that prevents using the more mundane ones, or your depression is so bad you need a second one you can take over top of the SSRI. It also gets a lot of use in PTSD treatment.
Basically no one wants to be on this drug. By the time they're offering it to you it's probably in your best interest to consider it.
Since the late 80s (when neural nets were a thing) I've figured that dreams are a way to "reweight" brain encodings: throw noise in, and the pathways to whatever strong signals come out probably ought to be reduced.
That’s not how the brain works. Neurons will generate and connect dendrites while awake or asleep, creating new pathways that simply didn’t exist moments before.
Further, the longer a dendrite stays connected, the more optimized it becomes (myelin sheathing aka “white matter”) and then it doesn’t go away even if it becomes unused. It probably (helps) explains why we fall back on old habits so easily, even if we’ve avoided them for years.
There isn't such a thing as inhibiting a pathway in the brain. Not directly, anyway.
There are inhibitory neurons that attach to other neurons and prevent them from firing. These neurons literally hug other neurons to prevent them from firing, which has the effect you're wanting but may not have the desired macro effect as there may be ways to still fire the neuron (not enough inhibition, or other brain chemicals like cortisol -- stress -- might prevent it from being fully inhibited). Further, the brain mostly works on creating more efficient pathways vs. inhibiting. If a behavior previously existed, life tends to suggest that it was important at some point and may be important at another point.
For example, my muscle memory for some activities still exists from being in the military. I will still annihilate entire teams in a game of paintball (which is pretty much the only place these skills are still useful, short of returning to a warzone).
Another example is that parts of your brain are dedicated to recognizing everyone you ever knew. Short of them dying or moving away, eventually, you may run into them again.
From some strange experience I believe REM sleep does "reprocess" things and change weights. Deeper more ingrained stuff requires this more disruptive process to change.
In popular media, human is often depicted as some super scavenger that can exploit and adapt alien technology to their own in a matter of 1-2 lifetimes. And yet here we are in reality, after thousands of years with abundant and intimate access to one of the most efficient and most capable information processor known to us, and we are still absolutely clueless about how it work.
The closest thing we built to mimic it cost maybe a million times more energy and perform a fraction of the functions of the original one. So much for "reverse engineering".