Year 9 Computing: Summer Assessment

An algorithm is a sequence of steps that can be followed in order to complete a task or solve a problem.

The computer first works out the value on the right hand side, adding the current contents of total to the current contents of price. That result is then assigned to the variable total, replacing whatever total held before. The equals sign here means assignment, not equality.

A for loop repeats a set number of times that is known before the loop starts, so it is count controlled. A while loop keeps repeating for as long as a condition is true, and nothing decides in advance how many times that will be, so it is condition controlled. A for loop is the right choice for working through every item in a list of forty names, because the number of repeats is known. A while loop is the right choice for asking a user to enter a password until they get it right, because you cannot know in advance how many attempts they will need.

A list stores many values under a single identifier and gives each one an index, so a program can hold thirty scores in one place instead of thirty separately named variables. That means a loop can work through every score with two or three lines of code, where separate variables would need thirty lines that all say almost the same thing. It also means the program still works when the number of scores changes, because nothing in the code depends on there being exactly thirty of them.

First, work out who is going to use the app and what they actually need it to do, and write those needs down as success criteria, because everything afterwards is judged against them. Second, sketch the interface before coding it, as wireframes or a storyboard showing each screen and how the user moves between them. Sketching is fast and changing a sketch costs nothing, whereas changing a built screen costs hours. Planning the algorithm and the test plan at this stage matters for the same reason: mistakes found on paper are cheap and mistakes found in code are not.

Machine learning is an approach in which a system is trained on a large quantity of example data rather than being given a set of rules written by a programmer. During training it finds statistical patterns in that data, and it then applies those patterns to make predictions or decisions about new data it has never seen before. Nobody writes a rule saying what a cat looks like: the system is shown many labelled pictures and works out the pattern itself.

A shortlisting system is trained on data about who the organisation has hired and promoted in the past. If most of those people were men, then the pattern in the data is that successful candidates look like the men already there, and that is precisely the pattern the system learns. The system has no concept of fairness and no way of knowing that the historic decisions were themselves biased: it simply finds the statistical relationship in what it was shown and applies it. The result is that a bias which used to be one interviewer's is now applied consistently to every single application, which makes it both harder to spot and much larger in effect.

A bitmap image is stored as a grid of pixels with a colour value recorded for every single pixel, which is why photographs are bitmaps. A vector image is stored instead as a set of instructions describing the shapes it contains, such as the coordinates of a line's endpoints and the thickness and colour to draw it in. Because a vector is redrawn from those instructions at whatever size is needed, it stays perfectly sharp at any scale, while enlarging a bitmap simply makes each pixel bigger and the image looks blocky.

A vertex is a single point in three dimensional space, fixed by its x, y and z coordinates. An edge is a straight line joining two vertices together. A face is a flat surface bounded by three or more edges, usually a triangle. The complete set of faces is called the mesh, and the mesh is what gives the model its shape, which is why a model with more faces can show finer detail but takes longer to render.

Rendering means calculating the final image from the model, and it has to be done separately for every frame. An animation runs at around 24 to 30 frames every second, so even a two minute sequence is several thousand complete images. Each of those images requires the computer to work out, for every visible surface, how light falls on it, what shadows are cast, how the texture appears at that angle and what is reflected in it. Multiplying that amount of calculation by thousands of frames is why studios render on large farms of machines rather than on one computer.

The case for it is strong on practicalities. An automated marker can process thousands of scripts in the time a human takes to mark a handful, it applies exactly the same standard to the first script and the four hundredth, and it does not get tired or distracted, which removes a genuine source of unfairness in human marking. It is also far cheaper, and results could be returned much sooner. The case against rests on what marking actually is. A written answer can be correct in a way the mark scheme did not anticipate, and a system trained to recognise expected phrasing may not credit an unusual but valid response, which penalises exactly the strongest candidates. There is also a risk of bias: if the training data under represents certain groups of students, their writing may be systematically marked lower, and because the same system marks everything, that bias applies at national scale. The most defensible position is not to choose one or the other. Automated marking is reasonable for short, factual responses where the mark scheme is closed, but extended answers should still be marked or at least moderated by humans, with a sample checked every session and a clear route for a student to have a script looked at by a person.