10 ms·
Brains speed up perception by guessing what's next (2019)
- gaudat 4y agoThis is sort of like pipelining in computer processors. As humans have quite a lot of latency, we make up for it by speculating what will happen in the short future with our experiences. This may also explain why people act irrationally for stressful situations or avoiding dangers and risks.
- hef19898 4y agoAnd why training for those situations, from disaster relief over surgeons to the military and professional fighters, is the way it is.
- dang 4y agoDiscussed at the time: Brains Speed Up Perception by Guessing What’s Next - https://news.ycombinator.com/item?id=19858375 https://news.ycombinator.com/item?id=19858375 - May 2019 (86 comments)
- asicsp 4y agoRelated: Our brain is a prediction machine that is always active (https://www.mpi.nl/news/our-brain-prediction-machine-always-active https://www.mpi.nl/news/our-brain-prediction-machine-always-...) https://news.ycombinator.com/item?id=32395840 https://news.ycombinator.com/item?id=32395840 (270 points | 7 months ago | 205 comments)
- shagie 4y agoFrom 2014, two articles (same study, one from a science journalism source, the other from the university's publication) Your Brain Is a ‘Prediction Machine’: It Predicts What the Other Person Is Going to Say https://www.learning-mind.com/your-brain-is-a-prediction-machine-it-predicts-what-your-interlocutor-is-going-to-say/ https://www.learning-mind.com/your-brain-is-a-prediction-mac... You Took the Words Right Out of My Brain - https://www.nyu.edu/about/news-publications/news/2014/april/you-took-the-words-right-out-of-my-brain.html https://www.nyu.edu/about/news-publications/news/2014/april/...
- ChaitanyaSai 4y agoAnticipation and prediction are so important in getting a leg-up in the real world. Real-world physical processes take time. Cannot just wait for food to enter the mouth to start revving up the digestive system. What is Health? by Peter Sterling is a lovely book that goes into this. Peter Sterling is a neuroscientist and the "co-inventor" of the term Allostasis (from Wikipedia: "Allostasis is the efficient regulation required to prepare the body to satisfy its needs before they arise by budgeting those needed resources such as oxygen, insulin etc., as opposed to homeostasis, in which the goal is a steady state." ) https://www.goodreads.com/book/show/44512576-what-is-health https://www.goodreads.com/book/show/44512576-what-is-health
- phkahler 4y ago>> Anticipation and prediction are so important in getting a leg-up in the real world. As a kid sometimes I would look around the room and pick an object to touch or pick up. Then I'd close my eyes and see if I could do it. This might involve taking several steps or even turning at the end before reaching out. There is some feedback from the feet but it's mostly open loop without vision. I could visualize my position along the way - a form of prediction. This skill is still useful when I need to turn off a light before crossing a room or opening a door :-)
- lsaferite 4y agoSpatial awareness and the ability to internally visualize actors in that space along with predicting their vectors is a very useful skill.
- davidkuennen 4y agoFor me it's most noticeable when driving. Especially on the Autobahn and fast. You're guessing what other drivers will do constantly and can react way faster that way.
- scotty79 4y agoIt's also way subtler effect. You see faster what you expect to see. You hear faster what you expect to hear.
- dreilide 4y agoWonder if this brain predicting thing partially explains 'priming'. e.g. with the brainstorm/green needle audio https://www.youtube.com/watch?v=qXxV2C1ri2k https://www.youtube.com/watch?v=qXxV2C1ri2k
- frereubu 4y agoI was studying something similar in my MSc about the neurobiology of language. It's well known that reading circuits in the brain make predictions about which words come next - a sentence like "The jam on the motorway made traffic slow" doesn't elicit much activity, whereas "The jam on the motorway was sticky" does, because "motorway" makes it much more likely that "jam" was about slow traffic rather than the condiment. (There are perhaps some interesting parallels to LLMs there...) But we were looking in the visual cortex, way back before the brain is even perceiving words, and we were finding pre-activation of circuits even there depending on whether people were presented with real words (e.g. sheep) or non-words (e.g. brast). I think there's a lot to be said for conceptualising the brain as a statistical machine - it makes a lot of sense in evolutionary terms. There's a great video introduction to that idea with Karl Friston from UCL here: https://www.youtube.com/watch?v=NIu_dJGyIQI https://www.youtube.com/watch?v=NIu_dJGyIQI
- atorodius 4y agoI was recently wondering if this changed with age: when talking with older ppl, I sometimes think they prefer hearing what they want to hear - or maybe rather, what they predict I would say? And it can be challenging to get theough to them with novel concepts.
- frereubu 4y agoIf the concept of the Bayesian brain - https://en.wikipedia.org/wiki/Bayesian_approaches_to_brain_function https://en.wikipedia.org/wiki/Bayesian_approaches_to_brain_f... - is roughly right, it's plausible that as you get older you might build up a pretty unassailable statistical model of the world which would need very significant evidence to the contrary to change in a short timeframe. But I think there are probably other, more human, effects dominating like the one you suggest - only listening to evidence that backs up your existing points of view, and that happens from a pretty young age! That would actually be against the idea of the Bayesian brain, which should accept new evidence and update the statistical model appropriately. Much as I find the general approach of the Bayesian brain useful, in many domains humans aren't really particualarly optimal in statistical terms...
- Agingcoder 4y agoThis might be tangentially related - I'm not sure. I've recently started playing jazz piano, after years of classical piano. Now, I'm supposed to be a reasonably good sight reader, or at least that's what I thought. The improv part of jazz I'm not good at but I get, but I've also tried playing from transcriptions (Bill Evans in particular), and have found I'm extremely slow at reading the music. After thinking about it, I suspect it's because the music makes no sense to me so I can't guess what's coming ( which is what I do with classical music). So while I anticipate with classical music and just confirm after the fact/right when the note actually shows up, with Bill Evans I have to think hard and actually confirm everything. My ears help but not much, since 'confirming' the sound is much harder with jazz than with classical music (it's unexpected , so again, there's some kind of delay between 'I hear the note' and 'ok that's right' while there's essentially no delay with classical music where my brain is kind of 'primed for acceptance'
- mjburgess 4y agoA little illustration: https://www.youtube.com/watch?v=LObPaCloY8E#t=1m https://www.youtube.com/watch?v=LObPaCloY8E#t=1m
- eru 4y agoOver time with more exposure and practice, Jazz should make more 'sense' to your fingers and ears?
- canadianfella 4y ago[dead]
- Agingcoder 4y agoYes, you're right and it's slowly happening. But it's a very slow process, and very surprising to me : I expected to be able to read fully written jazz easily, and, well, I can't, because reading fluently seems to be as much about predicting what's next as about knowing the alphabet and syllables!
- ghosty141 4y ago
- nsbk 4y agoApparently predicting is the most efficient way of running a brain, as opposed to reacting. Your brain is a predicting machine. Lisa Feldman, a top neuroscientist and writer, goes over this topic in a podcast episode with Lex Fridman. This is the chapter on the predicting brain, but the whole of it is worth a listen, mind blowing stuff: https://youtu.be/NbdRIVCBqNI?t=1443 https://youtu.be/NbdRIVCBqNI?t=1443
- psiops 4y agoAlmost like a, say, fancy autocomplete.
- nsbk 4y agoIndeed! maybe LLMs work in a more similar to our brains than we think
- psiops 4y agoIt could be. Maybe "understanding" is not so much more than predicting what's to come next with reasonable accuracy. Maybe LLMs understand better than we think.
- nsbk 4y agoThis idea is both fascinating and terrifying in equal parts
- gremlinsinc 4y agoThink back to cavemen days, or before actual spoken languages. You pointed and said oooh oooh... Still communication, but you couldn't wonder about the stars in the skies because you didn't even have words for what those were, or a clue about much, because prior generations of info wasn't passed down as easily as it is today. Entire generations of info is now instantly at our fingertips whenever we want it, even w/ creative spins thanks to GPT models. Language is the glue that holds up consciousness/sentience, it is intelligence. It's very hard to be intelligent without language, or the ability to understand language. A lot of learning deficits are because of ones inability to understand the other person (Teacher, etc, or focus on them long enough to understand what they're saying), I think language itself could turn most things into AI or just I, perhaps anything that can communicate and build language could be intelligent regardless if it's silcon or carbon. There's a lot of 'understanding' we glean from other senses, from what we see, etc... but we only know a tree is a tree because we've been told that (reinforcement learning), and because language exists to tell us that over and over again. So vision, audio, etc are all still language-based.
- qikInNdOutReply 4y agoAnd editing the history if it guessed wrong. You were frozzen in fear, stunned, unable to contemplate, the hardware abstraction is fine, its the pilots failure. https://en.wikipedia.org/wiki/Saccadic_masking https://en.wikipedia.org/wiki/Saccadic_masking https://en.wikipedia.org/wiki/Change_blindness https://en.wikipedia.org/wiki/Change_blindness https://en.wikipedia.org/wiki/Blindsight https://en.wikipedia.org/wiki/Blindsight
- thunderbong 4y agoFascinating! How much of this world I perceive is real, I wonder!
- mo_42 4y agoIt sounded like the brain has a prior on what it perceives. So something like Bayesian information processing. This seems not to be the case. At least in this preprint: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6516078/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6516078/
- pyrolistical 4y agoMakes sense. Those who predict well are going to avoid that tiger and live another day
- INTPenis 4y agoI wonder if this is related to prejudices. A survival instinct.
- epgui 4y agoYes.
- martopix 4y agoIf you've ever tried machine learning, it's easy to understand what a prejudice is: inference based on learning on biased data. Or sometimes the data doesn't even need to be biased: your stereotyping may be factually true on average. It's just that we should not make assumptions based on averages, especially when they involve people.
- INTPenis 4y agoWell... I mean if I'm going through an area known for violence and robberies and someone is following me, my prejudices might keep me alive. That's an extreme example of course but I think humans use their instincts every day. Calling it prejudices make it sound negative but I think it's just part of our survival instincts. And just like anything else it can be used negatively, or positively. As long as we don't use our prejudices premeditated to harm others I think it's a survival tool. But when we use them to defame groups of people who are nowhere near us then it can become negative.
- throwaway50607 4y agoAsk anybody who had psychosis - that's their prediction machine going crazy.
- edgyquant 4y agoI think it depends. Part of it is definitely your “social predictor” turning on without being around others. But I often wonder if part of it is just thoughts - they’re things made up of cells (neurons) so they are organisms in their own right (and some theories suggest they compete with one another for your attention) Most people are desensitized to this, some psychosis is the desensity breaking. Perceiving your thoughts as beings independent of yourself.
- saghm 4y agoNow I'm worried some enterprising security researcher will find a Spectre vulnerability in my thought process
- operatingthetan 4y agoTons of vulnerabilities already discovered https://en.wikipedia.org/wiki/Cognitive_bias https://en.wikipedia.org/wiki/Cognitive_bias
- jprete 4y agoAlso optical illusions.
- shagie 4y agoI'd suggest the BLIT series by David Langford http://www.infinityplus.co.uk/stories/blit.htm http://www.infinityplus.co.uk/stories/blit.htm (the BLIT story itself) https://www.nature.com/articles/44964 https://www.nature.com/articles/44964 (yes, its a work of fiction) https://www.lightspeedmagazine.com/fiction/different-kinds-of-darkness/ https://www.lightspeedmagazine.com/fiction/different-kinds-o... https://en.wikipedia.org/wiki/BLIT_(short_story) https://en.wikipedia.org/wiki/BLIT_(short_story)
- gremlinsinc 4y agoThey already have that, it's called a cult, or some might say religion (I would, but that'll offend others).. I used to be religious (Mormon), now agnostic but still somewhat spiritual, I don't believe in a God, just that the universe could itself be one big consciousness, or even part of a bigger organism, etc... and I'm still inspired by people, my kids, the universe, technological accomplishments, etc... Same feels I got in religion, I get from watching a meteor shower now.
- nathias 4y agoall perception is already anticipation of the future, this is how the brain compensates for input lag
- richrichardsson 4y agoI've always thought that the feeling of déjà vu is the prediction "algorithm" being comprehended ahead of the real world input, thus it feels like you already experienced that real world input.
- jamal-kumar 4y agoWhat kind of like this? [1] I know this is a silly thing to compare but haven't we all had huge issues with security vulnerabilities involving processors guessing what's next since like 2018? [2] I wonder if there's any general parallels between any potential information processing system and the concept that to make it go faster you can do this. I didn't read through the whole neuroscience article yet but that's definitely the first thing that came to mind. [1] https://en.wikipedia.org/wiki/Branch_predictor https://en.wikipedia.org/wiki/Branch_predictor [2] https://en.wikipedia.org/wiki/Spectre_(security_vulnerability) https://en.wikipedia.org/wiki/Spectre_(security_vulnerabilit...
- taskforcegemini 4y agoDon't we all base our actions by guesstimating what the outcome will be?
- jamal-kumar 4y agoMy point is that this exact guessing of what might come next is something that is universally exploitable in systems which do this
- xjay 4y agoThere is something similar in Daniel Kahneman's book Thinking, Fast and Slow. In my words, 1) The sender asks "how many X are there" (assuming that the recipient also knows that the answer is a large quantity). The recipient's mind indirectly loads and caches the concept of a large quantity. 2) The sender changes the "discussion point" to something different that also involves a quantity, however, because the original impression of the first quantity lingers, their perception may now be skewed. Link to one such example: Studying Machine Intelligence with Been Kim - #571 https://www.youtube.com/watch?v=C8SuqeH_Mg0&t=755s https://www.youtube.com/watch?v=C8SuqeH_Mg0&t=755s
- jamal-kumar 4y agoYeah it screams established psychological marketing/behavioral economics techniques to me
- svrtknst 4y agoWithout prediction, we wouldn't have goalies. A puck can pretty easily travel something like 30-40 m/s, and shots are regularly taken 4-5 meters from the net. Goalies have, what, 1/10th of a second to track and evaluate the shot and make a qualified save. If that was pure evaluation and reaction we would be far too slow. I also think most people who've played goalie in any sport know that you get a _sense_ for where the shot is going. High, low, left, right etc so it's not just guesswork and statistics either.
- herval 4y agoWithout prediction, we wouldn’t be able to _walk_!
- jameshart 4y agoThere is a lot built into the biomechanics of your legs too. The relevant term is “passive dynamics”.
- shagie 4y agoAnd a lot of unconscious thought for the biomechanics of the body. https://radiolab.org/episodes/91524-where-am-i https://radiolab.org/episodes/91524-where-am-i The section on Ian Waterman. Another article about him - https://mcneilllab.uchicago.edu/pdfs/IW_lost_body.pdf https://mcneilllab.uchicago.edu/pdfs/IW_lost_body.pdf > Mr. Ian Waterman, sometimes referred to as ‘IW’, suffered at age 19 a sudden, total deafferentation of his body from the neck down—the near total loss of all the touch, proprioception, and limb spatial position senses that tell you, without looking, where your body is and what it is doing. The loss followed a never-diagnosed fever that is believed to have set off an auto-immune reaction. The immediate behavioral effect was immobility, even though IW’s motor system was unaffected and there was no paralysis. The problem was not lack of movement per se but lack of control. https://topdocumentaryfilms.com/the-man-who-lost-his-body/ https://topdocumentaryfilms.com/the-man-who-lost-his-body/
- herval 4y ago
- dankai 4y agoSounds pretty autoregressive to me.
- rcarmo 4y agoNVIDIA DLSS for real life.
- jedberg 4y agoHere is an interesting tidbit I learned: the median response time in a human conversation is 0 milliseconds. In other words, more than half the time you begin responding to your conversation partner before they finish speaking. And in most cases you respond nearly instantly after they finish speaking. Anyone who has worked on voice assistants knows that computers take much longer than that to determine that someone has finished speaking. And anyone who has studied brain functions knows that it takes 100's of milliseconds for the brain to process information like spoken words. So basically that just proves that the brain must be predicting what the other person is going to say.
- thom 4y agoI don’t actually disagree, but ‘must’ seems too strong here. It could mean that they’ve already reached a threshold of comprehension earlier in the current speech, or that they’ve latched on to a pertinent point and don’t care to hear the rest.
- spyke112 4y agoMy wife hates when I do exactly this. It’s extremely hard to change, because it’s so subconcious.
- thom 4y agoI partly blame growing up on IRC for the behaviour - my brain seems to treat all communication as asynchronous. And while I often find myself apologising for the it, I also try and reassure people it's entirely because they said something I found interesting and was responding to it. It is a struggle though!
- jedberg 4y ago> It could mean that they’ve already reached a threshold of comprehension earlier in the current speech I feel like that is a distinction without a difference. If they already comprehend, then haven't they basically predicted the rest? > or that they’ve latched on to a pertinent point and don’t care to hear the rest. This I suppose is a little different, in that they are ignoring the rest, so yeah you're right there, it may not be predictive in that case.
- ptaffs 4y agoi think this helps explain why "you can't tell people anything" problem https://news.ycombinator.com/item?id=35282293 https://news.ycombinator.com/item?id=35282293 in that "most people aren’t very good at visualizing hypothetical, at imagining what something they haven’t experienced might be like, or even what something they have experienced might be like if it were somewhat different." Because the new information isn't aligned with their prediction. I think this is leveraged in propaganda to move a population to believe something untrue by leveraging a context and amplifying it. Literally enforcing <i>prejudice</i>.
- petilon 4y agoIn the last few decades, there has been an increased interest in the role of prediction in language comprehension. The idea that people predict (i.e., context-based pre-activation of upcoming linguistic input) was deemed controversial at first. However, present-day theories of language comprehension have embraced linguistic prediction as the main reason why language processing tends to be so effortless, accurate, and efficient. https://www.psycholinguistics.com/gerry_altmann/research/papers/files/Kamide_et_al_03.pdf https://www.psycholinguistics.com/gerry_altmann/research/pap... https://www.tandfonline.com/doi/pdf/10.1080/23273798.2020.1859568 https://www.tandfonline.com/doi/pdf/10.1080/23273798.2020.18... https://onlinelibrary.wiley.com/doi/10.1111/j.1551-6709.2009.01022.x https://onlinelibrary.wiley.com/doi/10.1111/j.1551-6709.2009... https://www.earth.com/news/our-brains-are-constantly-working-to-predict-what-comes-next/ https://www.earth.com/news/our-brains-are-constantly-working... People say ChatGPT is just a next-word prediction machine. But these articles say human brains are doing the same, at some level.
- nottorp 4y agoJust like ChatGPT right? :)
- turing_complete 4y agoAutoregressive sequence modeling is all you need
- jokowueu 4y agoBranch prediction?
- xpuente 4y agoToo bad. Not a single mention of CTLoop in the whole paper.
- tanseydavid 4y agoThis seems to fit nicely with Jeff Hawkins' "Memory-prediction framework" theory of brain function. https://en.wikipedia.org/wiki/Memory-prediction_framework https://en.wikipedia.org/wiki/Memory-prediction_framework
- eachro 4y agoI think people who have played sports can attest to how much easier (and less physically demanding it is) to play defense in sports like basketball/soccer when you are able to see where the play is developing.