There are also layers of pattern matching, particularly in creative work. Take for example bluegrass guitar improvisation, something I've been working at intensely lately. In order to play a single note, there are several pattern matches going on at once. First, there's the original song melody and harmonic structure. The notes of the improvisation need to reflect that base - the distance the improvisation extends from the melody/harmony without breaking it is an emotional quality. And this is a moving target! That melody goes by FAST... bluegrass is often played at tempos exceeding 200bpm.
So on one level, I'm following the melody/harmony. On another level, I'm inventing (or remembering?) substitutions that can work over that melody and harmony. On another level, I'm thinking about the overall flow of the improvisation, and how to achieve my artistic goals, express the feelings and ideas I have at the moment. On yet another level, I'm being purely technical - like making my hand slightly roll to play a downstroke on the E string followed by an upstroke on the B string without accidentally hitting the E string twice. On another level, I'm listening to the other musicians, with whom I'm generating a shared tempo, and who are also improvising, generating new ideas I can respond to while playing.
That's expertise. Being able to work on all this patterns simultaneously, some being totally orthogonal to others (like "roll my hand" vs "play a sixth instead of a fifth to imply relative minor").
That's also the importance of consistent patterns. Bluegrass and bop music, both extremely high-speed improvisational forms, have a couple of key things in common. First, time signature... they are strictly 4/4, which makes "chunking" a lot easier. Second, they are built on really common harmonic structures (I/IV/V chords in bluegrass, ii/V/I cycles in bop). Third, they have standard repertoires, and all musicians beyond beginner level are expected to know a lot of standards by heart.
So the complexity of certain parts of the music are tightly constrained, which frees the performers to tackle other forms of complexity (like improvisation) much more aggressively.
> I continue to struggle with the notion that expertise is merely pattern-recognition. It seems too simple — there has to be more.
I think people struggle with that because almost all cognition is at some level pattern recognition. Identifying your mom's face as a baby is pattern recognition. A dog learning to sit is pattern recognition. A mathematician finding the integral of some really difficult function is pattern recognition. A chess grandmaster playing is pattern recognition.
So I think this statement is correct, but it is pretty close to a tautology and hence not that helpful.
I have a ton of trouble recognizing peoples faces or names, especially out of the usual context. But I can remember programs I wrote in detail 10 years ago, network protocols, RFCs/IEEE standards. I think I have decent recall and pattern matching in some areas at a great expense of some other things. It took me until I was in my 20s to know what order the months are in a year. Another anomaly, my hobby besides computers is cars and I can identify every make and model and most years of almost every car on Earth. The name of my neighbors? Not so much. Brains are weird.
ime a lot of the ease of pattern recognition and recall comes down to interest. It's likewise easy for me to recall how and why a 302 differs from a 351 or a 460 but I couldn't tell you which letters proceed/follow 'j' without sounding it out in my head.
My solution has just been to fake it, if I can convince myself e.g. that my neighbors' names and faces are interesting/important than remembering them becomes almost impossible not to.
I think in this meaning “pattern recognition” begins to mean semantic reasoning. Which begins to mean all thought, and so everything can be seen as pattern recognition.
I feel there is a useful difference from what I see as real pattern recognition (identifying a letter) vs understanding and utilising the meaning in the words.
I think it's more complicated than this. I think the article is using two terms interchangeably that have very different meaning:
+ pattern matching - matching a series of events or a system state against an existing inventory of models or patterns to find a good fit and take action based on it
+ pattern recognition - decoding a pattern at work in a series of events or a system that is novel to you (or may be a hybrid of familiar and novel patterns)
As an example of the latter: the fire captain in Klein's "Sources of Power" who finds himself fighting what appears to be a small fire that is very hot and cannot be extinguished even after they douse if with water several times.
He becomes alarmed and orders his crew to pull back just as the floor collapses due to a fire in the basement below. It was not a pattern he recognized (one that matched prior experience) but it was anomalous in a way that he intuited was dangerous.
I think you can argue these are sides of the same coin. Seeing the new pattern emerge happens because you are repeatedly seeing the same event over again by searching prior experiences, it just happens that the experiences were quite recent. It's pattern matching against recent memories of trying water on the fire already and seeing that it failed each time.
I actually believe this is a distinction without a difference. If you look at Klein’s RPD model, the implicit memory operation is pattern matching against a prototype. The situation doesn’t have to be exactly the same as an old one, because the prototype provides expectancies that are really what the firemen is matching against.
I’m not sure when was the last time you read Klein’s work, but he certainly does not distinguish between ‘pattern recognition’ or ‘pattern matching’ — it’s all the same implicit memory operation in his model.
Being an expert is certainly not "just pattern matching" for physical sports, music, real-time video games, and many, many other things. Not only does one need to see the pattern, but they must be able to physically make their body do what they recognize needs to be done. Beyond "just pattern matching", if there's any chance of failure, what separates experts from amateurs is consistency of execution.
I would say there's quite a bit of pattern matching in video games, especially real-time ones like competitive FPSs.
You see a player round a corner and know that they are likely to end up at point A, B, or C within times X, Y, or Z.
You hear a specific gun being used to your left, it is likely being fired from areas D, E, or F because those have good sightlines to where you are.
For both circumstances you know that of the Q number of weapons you have R will be the best option for the range that you will likely encounter the enemy at and grenade S will also help. Weapons T and U will leave you undermatched. You switch to Q, reload it in an area that you know is relatively safe. As your approach the target you know that enabling powerup V will help you get the upper hand.
I didn't say there wasn't pattern matching in video games, I said that pattern matching is only part of being an expert.
To use your FPS example, knowing all of those things still doesn't matter if you miss every shot.
Another example -- 99% of the best Olympic athletes at the 2012 Olympics were less than 40 years old.[1] I don't think that people over 40 are worse at pattern matching, I think being an expert at something requires more than just pattern matching. Depending on the activity, physical strength, eyesight, reaction time, and so many other things matter too.
Here’s a (potentially frightening, depending on your leaning) thought:
If expertise is just pattern matching, and AI is almost entirely optimized for/shooting towards pattern matching, are we to close to AI making experts obsolete?
There's many levels of pattern matching, and computers have to start at the bottom. First you have to convert the real world into a form you can use, like vision or voice recognition. Then you've got to parse that into concepts. Then you've got to analyze those concepts for patterns.
Most of our software is still struggling at stage 1, and our software is so poor at pattern matching it can't detect obvious stage 1 failures at stage 2 or stage 3. We can make a program to recognize species of birds from photos, but it will also tell us that a bunch of static is (with 99.5% certainty!) a robin.
I don't think any of the experts I know are at any risk of being replaced by AI in their lifetimes.
I don't know if I agree with you there. A lot of AI work is around taking data and finding relationships between them. A current-gen AI might not understand what red is, for example, but it has no problem grouping red, blue, and green in to the the same cloud then mapping the relationships to objects that can be those colours.
Personall it feels like the current limit on the way to an AI that can take unknown information and turn it in to a pattern matching system is how much data it can fit in it's matrix at once.
If I remember correctly (and my info isn't out of date, it's hard to google) entity matching in photos is usually done internally with very low resolution files like 320x320 and lower. It's no wonder that it could easily mistake static for a robin. Take the right person's glasses off and you'll be lucky if they can tell you it's a bird.
> I don't know if I agree with you there. A lot of AI work is around taking data and finding relationships between them. A current-gen AI might not understand what red is, for example, but it has no problem grouping red, blue, and green in to the the same cloud then mapping the relationships to objects that can be those colours.
It's also worth pointing out that it doesn't matter what red is. All it matters is that the entity you talk to understand to recognize in the situations you'd expect them to.
In other words, red isn't some absolute concept, for all we know every human may subjectively perceive colors differently but most of us can still recognize the color red when someone asks us to.
I am doing SEO since before there was a word for it. 2004ish (Starting from Dev.)
By now it is all pattern matching for me. I get 1h with a real decision maker within the company and my way forward is clear. I wrote about it here https://t.co/833yGyRDXM
I sometimes get imposter syndrome, but then I recognize it as a pattern and get over it.
If by "pattern" you mean, "complex hierarchical probabilistic predictive models" and by "matching" you mean carefully weighting these models against each other and using them to predict unknown facts and outcomes and to generate well calibrated error bars over their predicted behaviours under different circumstances then, yes.
I think there are two different questions here.
1. Is expertise reducible to pattern matching?
2. Is understanding pattern matching sufficient for building systems that have expertise?
Probably many people here have seen the argument from solid state physicist Philip Anderson that #1 does not necessarily imply #2.
"The main fallacy in this kind of thinking is that the reductionist hypothesis does not by any rneans imply a
"constructionist" one: The ability to reduce everything to simple fundamental laws does not imply the ability to start from those laws and reconstruct
the universe", Anderson, link below.
I wasn't familiar and that was a really great read. Thanks for posting it.
It articulates well the fundamental problem that I struggle to explain when I talk about why I don't think strong AI will be a thing for a very long time, if ever.
"The constructionist hypothesis breaks down when confronted with the twin difficulties of scale and complexity."
By the article's own definition, "expertise" as just pattern matching/recognition is too simplistic... that leaves out the "action" part mentioned in the article. e.g. an expert violin player better be able to play damn well to be called an expert. You have to add practice/work to the mix, not just pattern recognition.
Being an "expert" is great as a goal, but it's not the only be all end all. I'm reminded of a plaque from schoellkopf power station that said:
to know what to do... wisdom
to know how to do it... skill
to do the thing as it should be done... service
In Daniel Kahneman's Thinking, Fast and Slow, there is an interesting section on intuition, and specifically the intuition of experts. It says something very similar, that "intuition is nothing more and nothing less than recognition." With expertise, one has more things that they can recognize, and they can do so quickly and almost effortlessly.
I find posts like this simultaneously heartening and disheartening. Heartening because it means a lot of behaviors that are billed as "genius" or "high IQ" or what have you, might just be an individual unconsciously drawing upon years, maybe even decades, of experience that is layered on top of each other. The most obvious example of this, for our community, are actually codebases. The more you work on it, the easier it becomes to debug and figure out issues, sometimes without even looking at the code. When I was first starting out as a junior software engineer, I was shocked at how quickly some of the seniors could figure out bugs. I was upset because I thought that they might just be smarter than me, from a raw intelligence perspective.
But now, some years later, I've become that person, at least in my current company. I can debug issues pretty quickly, and my brain has a store of literally thousands of bugs and snags that have come up over the years, to the point where if someone says, "This feature isn't returning the right variables", I can usually point them to the fix without even looking up from my laptop. It sounds like magic, but really it's just localized expertise.
That's the heartening part--that all it takes is years to do it. The expertise is available to everyone, perhaps.
The disheartening part is the other side of the coin--what if you've started late? Does this mean that software engineers who start late, in their 40's or 50's, for example, might never reach the heights of software engineers who started early? That isn't true--there are examples to the contrary, I hope--but maybe they have a much steeper hill to climb. Or they have to figure out a different approach to getting up the hill, maybe combining years of experience from a different field.
I love essays about these topics: expertise, pattern-matching, your brain being shaped by what you do, etc. If anyone has any others, I'd love to see them.
As someone that started "programming" very early on (when I was 9 years old) and noticed over the years the huge advantage I had over my peers in terms of "thinking as a computer" I can definitely say that experience matters. But it's most definitely not all that matters, not even the most important thing after a certain point:
* other people may be faster learner and may forget less
* other people's interest in the subject matter may decrease at a slower pace than mine's
* other people are simply younger and their body can be "fully awake" for more time, have more energy available to learn (instead of dealing with family issues), their brain is healthier/younger
There's also a unintuitive negative effect of having expertise: you easily miss out the simple explanations/solutions. This makes total sense, the more things you about a subject matter the more your brain will attempt to explain behavior through those means. As you become an expert in a subject matter you deal with more specialized/detailed aspects which results in your brain trying to think about those to explain situations or find solutions when a "newcomer" will very easily figure out that it's a simple problem/solution that can be done.
So don't feel disheartened, there are pros and cons to being at every level of expertise, it's a tradeoff.
You brought up some really good points. Your second point about interest--that's a great point. I think interest--and, subsequently, focus and attention--are force multipliers, especially in a field like programming, where experience and expertise don't just help you lay more bricks, but actually increase how many bricks you can lay in the same unit of time as before.
The negative effect is interesting. I've heard it said before that non-experts often come up with novel solutions in some fields. That might because of what you pointed out, that the experts are too focused on a certain range of solutions, so they might be missing out on a range that a "layman" can easily identify.
I tutor for time management and study skills at my University, and I tell this to students a lot, especially when they ask about computer science courses in particular. I had someone ask how to get better at debugging (she was taking our C course which is known for giving people a hard time). I struggled to give her an answer besides knowing the right resources, how to properly search and troubleshoot, and raw practice.
It made me realize how much of software development is just natural to me because there's years and years of practice under my belt.
So on one level, I'm following the melody/harmony. On another level, I'm inventing (or remembering?) substitutions that can work over that melody and harmony. On another level, I'm thinking about the overall flow of the improvisation, and how to achieve my artistic goals, express the feelings and ideas I have at the moment. On yet another level, I'm being purely technical - like making my hand slightly roll to play a downstroke on the E string followed by an upstroke on the B string without accidentally hitting the E string twice. On another level, I'm listening to the other musicians, with whom I'm generating a shared tempo, and who are also improvising, generating new ideas I can respond to while playing.
That's expertise. Being able to work on all this patterns simultaneously, some being totally orthogonal to others (like "roll my hand" vs "play a sixth instead of a fifth to imply relative minor").