0:00:06.160,0:00:12.160
I'm going to talk to you today about how your mind learns to program. So learning is an essential

0:00:12.160,0:00:16.640
part of being a software developer, both when we initially learn to program before we begin our

0:00:16.640,0:00:21.760
career but also throughout our career as we learn new technologies, new frameworks, new programming

0:00:21.760,0:00:26.800
languages. But how does our learning actually work? That's what I'm going to talk about today.

0:00:28.480,0:00:35.120
So learning involves storing and thinking concrete facts. So for example when you're very young you

0:00:35.120,0:00:40.000
learn numbers - you learn how to count - so you learn one two three four. Then later when you

0:00:40.000,0:00:45.120
come to learn programming you're probably going to learn that whole numbers are also called integers.

0:00:45.120,0:00:50.480
So your mind stores these facts and it also links together facts that it thinks are related so that

0:00:51.200,0:00:56.480
you can associate these two things together. So that's concrete facts, but a lot of what we

0:00:56.480,0:01:02.000
do in programming is actually learning abstract concepts. So think for example about trying to

0:01:02.000,0:01:06.240
learn about the concept of values or the concept of types when you're initially learning how to

0:01:06.240,0:01:12.720
program. So how can we learn this? One possibility is for example to go to Wikipedia - the start of

0:01:12.720,0:01:16.720
the article "Values" there says that the value is the representation of some entity that can

0:01:16.720,0:01:21.840
be manipulated by a program. The members of a type are the values of that type. But this isn't

0:01:21.840,0:01:26.160
very useful for our learning: it's kind of very abstract and fuzzy and it kind of slips through

0:01:26.160,0:01:31.440
your fingers as you try and reread it to actually understand it. So it's quite difficult to learn

0:01:31.440,0:01:38.000
abstract concepts concepts by directly reading just a description of them. What works better

0:01:38.000,0:01:44.960
is if we use examples. So if you learn instead that "hello" and "bye" are string values, 0,

0:01:44.960,0:01:49.760
12, -5 are integer values, and true and false are Boolean values,

0:01:49.760,0:01:52.960
and then once you've learned these facts your brain can then start linking them

0:01:52.960,0:01:56.720
together to say, all these are different kinds of values - they have different types.

0:01:56.720,0:02:00.320
And this way you can actually start to get a handle on the abstract concepts.

0:02:01.680,0:02:05.760
So abstraction is learned by linking these examples together:

0:02:05.760,0:02:11.440
that's the best way for your brain to learn. So our first lesson is that you should study varied

0:02:11.440,0:02:16.320
examples if you want to learn abstract concepts. And also, if you're having to explain this to

0:02:16.320,0:02:19.920
somebody else, perhaps to a junior developer, you should use examples when explaining,

0:02:19.920,0:02:23.600
because that's going to be easier for them to learn the idea of the abstraction.

0:02:26.000,0:02:30.560
So let's talk now about how memory is organized. So memory is roughly split

0:02:30.560,0:02:35.600
into sort of two different systems. So we have long-term memory, which is what we sort

0:02:35.600,0:02:39.040
of classically think of as memory, where we store all these facts and link them together.

0:02:39.600,0:02:44.480
But we also have something called working memory. So long-term memory is essentially

0:02:44.480,0:02:49.920
unlimited and can store all these different types of facts, whereas working memory actually has a

0:02:49.920,0:02:54.240
very small capacity. You might have heard of this before: the idea that you can only keep,

0:02:54.240,0:03:00.640
sort of, around five things, four to six numbers - estimates vary - in your working memory at any

0:03:00.640,0:03:06.560
one time. So that's a very small sort of limited fixed capacity. And it's not something that you

0:03:06.560,0:03:11.280
can train to improve: it seems to just be sort of biologically inherent. So you can never increase

0:03:11.280,0:03:18.160
your working memory capacity across your lifetime.
So what things are actually held in the working

0:03:18.160,0:03:23.680
memory? Well, it's essentially the concepts pulled in from long-term memory. So each of the

0:03:23.680,0:03:28.400
items held in working memory is something that you've learned and stored previously into your

0:03:28.400,0:03:34.240
long-term memory. But you might then think, okay, well if working memory is kind of so limited,

0:03:34.240,0:03:39.760
how can we ever sort of get smarter? How can we improve our capacity? How can we get better at

0:03:39.760,0:03:44.320
solving sort of increasingly complex problems? And the answer to that lies in the long-term

0:03:44.320,0:03:49.200
memory. So we're pulling in sort of one item, something called a "chunk", from long-term memory,

0:03:50.240,0:03:54.480
but they don't all have to be necessarily the same size. You can learn more about something

0:03:54.480,0:03:58.000
and increase the size of an item in your long-term memory. You could learn about

0:03:58.000,0:04:01.680
related things in your long-term memory such that they effectively get merged together.

0:04:02.560,0:04:06.320
So by learning more we get larger items in our long-term memory

0:04:06.320,0:04:10.240
and that means that our working memory can actually hold more knowledge overall

0:04:10.880,0:04:17.360
so that we end up with some spare capacity to hold even more things. So our lesson there is

0:04:18.320,0:04:22.880
increasing your knowledge actually increases your processing capacity for dealing with problems.

0:04:25.840,0:04:31.840
So your brain is very eager to link related facts together in order to fulfill this process. So

0:04:31.840,0:04:35.600
for example when I was showing this slide a moment ago you may well have been thinking,

0:04:35.600,0:04:40.400
ah, so working memory is a bit like CPU registers or like a cache, and then there's these kind of

0:04:40.400,0:04:45.120
pointers pointing off to long-term memory which is this sort of other data structure. So you're

0:04:45.120,0:04:49.520
very keen to kind of link your existing knowledge about compute - how computers work - to knowledge

0:04:49.520,0:04:54.320
about how the brain works. But we have to be a bit careful: your brain does not work exactly like a

0:04:54.320,0:05:00.400
computer. The sort of differences - it's not quite as neat as the way that the computer is organized,

0:05:00.400,0:05:05.040
and it can have sort of different capabilities that the computer. And this is an example of

0:05:05.040,0:05:09.920
a more general point, which is that sometimes existing knowledge can trip you up. You can form

0:05:09.920,0:05:14.080
bad links between things you already know and things you're trying to learn. So for example,

0:05:14.080,0:05:19.360
you might learn how inheritance works in Java then you come to learn inheritance in C++, you

0:05:19.360,0:05:23.760
try and sort of use all your existing knowledge, and later you realize that actually there's some

0:05:23.760,0:05:27.920
finer points that don't match up and there's some key differences there, but your brain was very

0:05:27.920,0:05:33.040
eager to sort of link all this together and so it can lead you into sort of incorrect knowledge.

0:05:35.840,0:05:41.200
So the last thing I'm going to talk about is to do with problem solving. So a common misconception as

0:05:41.200,0:05:46.000
to how this works - so there's various domains where you might need to solve a problem,

0:05:46.000,0:05:49.680
so in your programmer - in your day-to-day life you might want to fix the performance

0:05:49.680,0:05:54.240
of a web service, for example, but there might be other things you want to do that involves solving

0:05:54.240,0:05:58.640
problems. You might want to make a move in a chess game, or you might want to solve a brain

0:05:58.640,0:06:03.760
teaser - so weighing a jumbo jet was kind of a common interview question a few years ago, an

0:06:03.760,0:06:09.200
example of sort of these teaser type problems. So the misconception that people have about problem

0:06:09.200,0:06:14.000
solving is that you have sort of one problem solving process in your brain and it's like a

0:06:14.000,0:06:17.920
sort of procedure that you can apply to all these different things that you might need to tackle.

0:06:19.120,0:06:23.440
And the implication of this was, therefore, people thought that maybe if you, for example,

0:06:23.440,0:06:27.600
learned more about how to do chess that improves your problem solving and therefore will improve

0:06:27.600,0:06:33.440
your programming. Also, they might think that by testing you on how good you are at brain teasers

0:06:33.440,0:06:37.680
they can test your problem solving module and thus test how good you are at programming.

0:06:38.320,0:06:45.440
But this is not the correct model of how things work - so this is wrong. How it actually works

0:06:45.440,0:06:51.120
is that we have totally separate domains and we learn problem solving per domain. So if you want

0:06:51.120,0:06:55.520
to fix the performance of a web service then you learn how to solve that problem, but your

0:06:55.520,0:06:59.680
knowledge is specifically constructed around that domain and your problem-solving ability

0:06:59.680,0:07:04.400
is based in that domain. It's then separate from how good you are at making a chess move,

0:07:04.400,0:07:10.080
and separate again from how good you are at brain teaser problems. So learning one won't improve

0:07:10.080,0:07:15.840
the other and testing one won't necessarily reveal very much about one of the others.

0:07:17.760,0:07:22.480
So this leads to our third lesson which is that problem solving is not a generic

0:07:22.480,0:07:26.880
skill that's applied across all different domains. It's actually acquired per domain

0:07:26.880,0:07:33.360
through generally practicing on that domain.
So how can we use all these lessons to sort

0:07:33.360,0:07:37.520
of supercharge our ability and to sort of learn functions and solve problems?

0:07:38.560,0:07:42.240
So the lessons we had were to use varied examples to learn and explain

0:07:42.240,0:07:47.280
abstract concepts - your brain likes to learn the abstract by joining together concrete examples.

0:07:48.640,0:07:52.720
We could increase our processing capacity by increasing and strengthening our knowledge,

0:07:52.720,0:07:57.040
so by learning more in long-term memory and increasing the sort of size of the items in

0:07:57.040,0:08:01.200
our long term memory, we can actually free up working memory capacity to deal

0:08:01.840,0:08:06.800
with more problems. And this is another way that you know your brain is not like a computer:

0:08:06.800,0:08:11.840
this is quite sort of counter-intuitive if you think of it from a, sort of, how a CPU functions.

0:08:13.680,0:08:18.480
And finally you can improve problem solving within a specific domain but it's not shared

0:08:18.480,0:08:22.480
between different domains, and if you want to know how good somebody is at solving problems within

0:08:22.480,0:08:26.720
a domain - a particular programming language or framework or whatever - then you should look at

0:08:26.720,0:08:31.120
that specific skill, not by trying to test it - by sort of coming at it from another angle.

0:08:32.320,0:08:35.360
So those are my lessons and there's my contact details

0:08:35.360,0:08:39.840
in case you want to know any more about this. Thank you for listening.