8 ms·
Expert programmers have fine-tuned cortical representations of source code
- cliffy 7y agoI'm going to go out on a limb and guess that experts in `Y` have fine-tuned cortical representations of `domain(Y)`.
- cheez 7y agoHypothesis -> Study -> Confirmation/rejection.
- rriepe 7y agoHypothesis -> Study -> Newspaper article that rephrases it into whatever it thinks is interesting
- contingencies 7y ago-> HN article that hand waves five levels of meta -> driveby rewrite in five characters of APL
- mattrp 7y agoThat's true - when HFT first became a thing, there were a few studies that compared reaction times between the top traders and the machines and found that expert traders that composed the market making function could make better trades in less time (on the order of fractions of a second) with fewer errors. Of course, that's a distant memory at this point -- the machines have obviously won out on the part of the market's structure but kind of interesting that the humans were quite fast (faster than I would have guessed).
- akira2501 7y agoFair point.. but equally interesting is pondering the neural structures of those who are domain experts but were born blind. Do they have a sympathetic structure that aids them? What does that look like? Is there a way to develop both this and "cortical representations" equally?
- bsder 7y agoProbably. It has been pretty much confirmed everywhere it has been tested. For example, chess masters memorize chess positions faster than normal people--but only for chess positions that are "valid". If you can't get to the position with play, they don't memorize them any better than normal people.
- hackinthebochs 7y agoI'd love to see work like this done on chess super GMs.
- kleer001 7y agoand London Cabbies
- codingdave 7y agohttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC18253/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC18253/
- iguy 7y agoIt's been done! An interestingly (IIRC) they had amazing memory for boards which were actual states of play. But if the pieces were arranged randomly, they were little better than beginners.
- cheez 7y ago> For the first factor, we adopted programmers’ ratings in competitive programming contests (AtCoder), which are objectively determined by the relative positions of their actual performances among thousands of programmers Ah perfect.
- flafla2 7y agoThis makes me extremely skeptical of the study. At best, it says something about people who are whizzes at programming competitions, and certainly not "Expert programmers" in general.
- cheez 7y agoEver met a normal person... They have no idea what an expert is like. For them, easily ranked = source of expertise.
- whyaduck 7y agoComing soon - a service to screen prospective software developer hires via fMRI.
- annoyingnoob 7y agoThis was my first thought too, yikes.
- femto113 7y agofollowed by 14 books in the vein of "How to Crack the Coding fMRI"
- lima 7y agohttps://en.wikipedia.org/wiki/Goodhart%27s_law https://en.wikipedia.org/wiki/Goodhart%27s_law
- exdsq 7y agoMy girlfriend is an fMRI physicist with access to a scanner - I’ll recommend this as a study if she comes to a loose end with her research for a while :P
- sickygnar 7y agoclaustrophobic? find a xanax prescription. medical implant? sorry
- moron4hire 7y agoWell this explains why, when I first started to learn to program, it took me hours to find one missing semicolon in 15 lines of C code, but after 20 years I can basically glance at a screen full of code and find them almost immediately.
- oscargrouch 7y agoI am a solo in a multi-million line of code codebase, and im always amazed about how many detail are still in my memory, even when i dont see those particular chunks of code for a year. It amazes me mostly because this is not the sort of tasks our brains were specialized in doing, by estimating our development path in nature as mammals. We discover and/or create new models and yet, we are able to adapt to them with some effort. Our new environments will probably affect the DNA that have the brain design, and we will pass all of this as a legacy to our kind. The brain is such a masterpiece of nature.
- cat199 7y ago> It amazes me mostly because this is not the sort of tasks our brains were specialized in doing, by estimating our development path in nature as mammals. concievably this might be similar to spatial navigation, which would be something we'd evolve for as hunter/gatherers
- vlovich123 7y agoAnd that we build artificial structures that align to how our brain operates. I suspect the source code written & the tools built would look a bit different if they're tuned to a brain with a totally different operational principle.
- oscargrouch 7y agoThis is a great observation. Our tools would be designed according to our experience of the world. But theres another catch, the tools we use change the shape of our mindset, leading us to design other tools that now map to this new perception of reality we have with the new mindset we built by using the former tools. The hunter-gatherer example given in the first comment for instance, i bet with enough evolutionary steps, we will not be able to trace back to that mindset, because we have "patched" our own mindset so many times with new tools(over tools) that we will have very little "mindset identity" to share with our former "hunters-gatherers" primordial cousins.
- jwkane 7y agoProgramming is somewhat unique in the size and complexity of the structures expert practitioners have to hold and maintain. But only somewhat. It would be certainly be interesting to compare with other professionals with similar years of experience.
- thedirt0115 7y agoHow might one quantify "size and complexity" for different domains? Physical architecture seems to me like maybe it's somewhat similar to programming, but even that seems hard to me to compare. Fiction author also seems like an interesting comparison, but again, I 'm unsure of how to begin to quantify size/complexity in a way that's apples-to-apples with programming.
- uberdru 7y agoSo do expert carpenters. And expert musicians. Etc.
- 40acres 7y agoNot surprising but good to see some research. Just off the top of my head I assume working memory capacity may be slightly increased, whatever portions of the brain are used to decode / encode information, and perhaps even spatial intelligence.
- BigHatLogan 7y ago"Tell me what you pay attention to and I will tell you who you are." - Jose Ortega This is fantastic. I've always had a hunch that experts are "seeing" things differently than novices, and seeing things that people who aren't in the same domain aren't even aware of. As people develop more expertise, I imagine the gap becomes larger, to the point where they're aware of things that "laymen" aren't even cognizant of. Two examples: In one of Malcolm Gladwell's books--I forget which one--he writes about a firefighter who, almost unconsciously, ordered his men out of a building right before it collapsed. He didn't even know why. After going through the day's events with a researcher, the researcher realized that the fireman was subconsciously (unconsciously?) attuned to what was happening around him. His brain was operating unbeknownst to him due to his years of expertise. A friend of mine started a small business late last year. He started it on a lark, after a trip to the local autobody shop. Usually his spouse handled all of the car-related affairs, but for some reason he ended up having to do it himself this time. When he went to the store, he noticed that the software they were using to manage their inventory was dated and quite awful. He took down the manager's information and messaged him a week later with a proposal for revamping their software system. The autobody became his first client. The manager said that hundreds of people come in and out of that store every day, but my friend was the first person to point out that their software wasn't very good. My friend isn't an expert programmer by any stretch, but he was more of an expert than anybody in that office in regards to developing software. His brain basically zeroed in on it without him noticing. Related to a thread from last week about our brains physically changing depending on what we do and pay attention to: https://news.ycombinator.com/item?id=22201771 https://news.ycombinator.com/item?id=22201771 I've become somewhat "obsessed" with this topic recently and would love to know if others know of any resources I can check out. I read "The World Beyond Your Head" by Matthew Crawford recently--he spells out this phenomenon quite well. EDIT: Jose Ortega's full name is Jose Ortega y Gasset--I forgot to add the last part.
- low_common 7y agoHow's his business doing? Is he doing consulting?
- BigHatLogan 7y ago
- ible 7y agoI don't know what 'cortical representations' are exactly, but it seems generally true that experts in a domain build up gradually higher level pattern recognition in their area of expertise. Whether it is driving a car, playing a sport or game, or writing software. As a beginning programmer I had to consciously think about fundamental concepts all the time, or grapple with my limited knowledge of a programming language, instead of thinking about the problem. As an experienced programmer I think in higher level concepts and abstractions, and the fine code details happen without me consciously thinking about them particularly. This actually makes learning a new programming language or IDE more painful now than it was when I was new. It probably takes me less time to get to an equal level of skill in language X than a beginner, but getting to the level of fluency where the low level details don't require conscious thought takes time and practice. Being slowed down so much while getting to that point is deeply frustrating.
- courtf 7y agoTook the words right out of my mouth! I feel a little guilty when I'm reminded that "the language is just a tool," meanwhile I am loathe to switch unless truly necessary. Honestly, I don't think I have the memory space to be deeply fluent in more than a few languages, and so when presented with the option of learning/using a new one, I weigh that against proficiency lost in other areas. I know it's not zero-sum, but context switches have a cost. Especially in industry, languages are often so similar that it's hard to justify. I used to feel differently, and be more enthusiastically polyglot (I've got production code currently running somewhere in the world in about 8 languages). It wasn't until I found a language that clicked with me that I even had the motivation to become really fluent.
- croh 7y agoWhat is that language clicked with you ? Can you please share ?
- croh 7y agoI called this building consciousness. If you are pouring lot of code daily and moving between many projects, you will start writing comments automatically. Once you build enough consciousness about anything, learning become very second nature. And this is the reason 10000 hours rule work imo. On side note, the biggest mistake people do is they can't think straight. If you want to learn coding, start coding first. Don't read theory or watch youtube videos. If you want to learn public speaking, start giving speeches. Don't read motivation books. This is how child learn. Fail and improve. Sadly many kids want to become hacker by watching hollywood movies and youtube videos.
- zelly 7y agoIn other words obsession. When you are writing code in your dream that's when you know you love what you're doing.
- quotemstr 7y agoThe brain is a computation device. Like any computer, it has to represent data in some kind of structure in order to operate on it. But what are the exact data structures that brain uses? Nobody really knows, but there are some fascinating clues. Related paper: https://elifesciences.org/articles/21397 https://elifesciences.org/articles/21397 "What the success of brain imaging implies about the neural code" fMRI is a really crappy imaging technology: it has low spatial and temporal resolution. Yet when we use it to read activity in the brain, we're able to gain real insights into how the brain actually works. This paper goes into what fMRI's success tells us how about the brain represents concepts and what fMRI's success rules out: "the neural code must be somewhat spatially and temporally smooth with respect to neural activity (which is several orders of magnitude smaller than voxels) for fMRI to be successful. Whatever is happening in the roughly one million neurons within a voxel (Huettel et al., 2009) through time is being partially reflected by the BOLD summation, which would not be the case if each neuron was computing something dramatically different for in-depth discussion"
- lmilcin 7y agoWhen I was young kid and teenager I would spend a lot of time playing checkers with my father. My father wasn't master or grandmaster level but he played well. We would sometimes spend entire evening playing dozens of fast paced games. When friends asked me how it is possible to calculate so many moves in advance in such short time (I would typically have response prepared for whatever move they made) I realized I don't really know. Only occasionally I would spend time really analyzing the future state of the board and I was typically able to imagine how the board would look like in many moves in advanced and was even able to predict which moves would look tempting in half a dozen moves in the future. It did not feel as if I was playing the game. Retrospectively, it felt more as choosing from menu of options but I did not know or cared to know where the options came from. I came to conclusion after so many repetitions the brain gets somehow wired and runs all those simulations "behind the scenes" so that the conscious part of the mind is relatively free to make very high level decisions.
- virgil_disgr4ce 7y ago> grandmaster level ...of checkers?
- symmitchry 7y agohttps://en.wikipedia.org/wiki/Marion_Tinsley https://en.wikipedia.org/wiki/Marion_Tinsley
- trombonechamp 7y agoI was pretty excited to see this paper but unfortunately it seems like they have a pretty big multiple comparisons problem. The spatial structure of their brain maps and the really strong correlations scream multiple comparisons. Glancing through their methods briefly to try to figure out why, it looks like they are performing their decoding and leave-one-out cross validation independently for each small group of voxels, which should generate a ton of false positives. This is pretty easy to control for, but they didn't do the key control analysis: scramble the trial identity (category) to create null datasets and then recompute the measures. Unfortunately I bet that these null models would have had similar correlations to what they observed.