October 21, 2020

How 2 predict grades badly...

29 minutes

Zoey and Bia discuss some of the mistakes that Ofqual made in their algorithm, how using “complicated” maths is not necessarily better, and share some anecdotes of their experiences with teachers and dealing with (un)conscious bias.

Timestamps

00:20 – Introduction

01:54 – Initial thoughts

02:42 – Mistake #1 – Their approach

04:43 – Mistake #2 – Data leakage

05:15 – Mistake #3 – Emphasis on the rank

06:57 – Mistake #4 – Ignoring outliers

08:31 – Mistake # 5 – No peer review

09:16 – Mistake #6 – Too precise

11:14 - Mistake #7 –Disregarded unconscious bias.

12:53 – Mistake #8: Education system in the UK.

13:30 – Ofqual considered edge cases – (almost a positive thing!)

15:00 – How we might have handled this situation

17:39 – Another example of algorithmic bias – Accounting system the Post Office used.

18:53 – Challenge: “Prison Break”. This based on “Liar’s paradox” attributed to Epimenides (amongst many other philosophers). For more challenges, presented in a more visual manner, check out our Instagram.

25:52 – Anecdotes of experiencing bias from teachers.

Useful links:

Ofqual’s report

Bristol University's study on unconscious bias - http://www.bris.ac.uk/media-library/sites/cmpo/migrated/documents/wp221.pdf

Tom SF Haines’ post (Lecturer in Machine Learning at Bath University) - http://thaines.com/post/alevels2020

...more