Byte-sized

edtech research
AI for School Teachers

Introduction
The following text has been adapted from AI for 1. Use AI safely and effectively to support their students’
School Teachers (Luckin, George, Cukurova, learning
2022), available to buy from routledge.com 2. Help students understand how AI works and how to
use the power it brings ethically
The authors joined forces because they want to convey
With each passing day, the advances humans make in
the ways in which AI can help you, and they want you to
AI change the world. Perhaps most importantly, these
understand how to work alongside AI, so that it augments
advances are transforming the world of work. This
what you are able to achieve as a teacher, school leader,
transformation has significant implications for everyone,
parent, school governor, or trustee. Most of all, they want
but as long-time educators, the authors of the book AI for
you to realise how amazing your own human intelligence
School Teachers, on which this Byte-Sized EdTech Research
and brain power is, so that you can keep AI firmly in its
is based, are particularly concerned about the implications
place as a tool to make you, and those around you, smarter.
for teachers. Teachers are responsible in large part for
The aim is therefore to help teachers understand enough
preparing students to act wisely and well – now, and in
about AI to buy tools that will be useful and appropriate
the future. Increasingly, this demands that teachers have
to their context, to use AI effectively, and to help other
a working understanding of AI, so that they can do two
people to do the same. Educational examples are used
things:
throughout to illustrate the authors’ points.

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 Meet the authors
• AI for School Teachers was co-authored by 3 • Karine George is an award-winning educationalist
experts in education and Artificial Intelligence, and an active research practitioner. She was a
•providing teachers an all-access pass to one of the headteacher for more than 20 years in a school rated
most important public conversations of our era as ‘Outstanding’ by Ofsted
•
• Rose Luckin is Professor of Learner Centred Design at • Mutlu Cukurova is a Professor in Digital Technologies
University College London and Founding Director of in Education at UCL Knowledge Lab, University College
EDUCATE London

AI and you: what AI am I using already?

A good place to start in our discussion on the searches to online shopping and phone apps that
understanding of AI is to consider what AI you’re help you navigate, AI is being used all the time, and it is
using already in your daily life and workplace already being used in schools
• There are adaptive platforms that help students on an
• Many people use speech-to-text when they dictate a individual basis through an AI that adjusts the tasks the
note on their mobile phone. The software that turns student is asked to complete, and the amount of help
the sound made by their voice into the words and they are given, for instance
spaces that appear in the note is a form of AI • When anybody decides to spend money on AI they
• Similarly, you might be using a voice-activated should understand what the AI is doing and what the
personal assistant in your home or on your phone. Siri, AI can do
Alexa, and Google Home are all examples of this kind • This does not mean that they need to understand how
of device. AI analyses the sounds that come in through to build an AI system or to write computer code, it just
the microphone, interprets the meaning of these means that they need a general understanding of what
sounds, constructs a response, and expresses the is happening inside the “black box” of AI
response as spoken words
• What are the educational outcomes we hope to
• There are many, many ways in which AI is being used achieve for our students and school with the help of
by billions of people across the world. From Google AI?

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 Explanations, definitions and some history
Good Old-Fashioned AI
• In 1964, a system called ELIZA, a computer programme that development team
played the role of a psychotherapist, was devised
• This need to have an exhaustive amount of information
• A potential patient was asked to type in their symptoms and preprogrammed is precisely the problem with what is known
ELIZA searched for keywords, and returned stock phrases to as Good Old-Fashioned AI (GOFAI)
progress the conversation
• In chess, there are thousands of ways a game could look just
• This approach was called production rule-based pattern two moves ahead. If you need to predict what the board
matching, and although ELIZA’s operation sounds primitive, looks like twenty moves ahead, the number of possible
several systems evolved from this approach became configurations stretches into the quadrillions
sophisticated enough to accomplish advanced activities,
such as diagnosis and treatment recommendations from a • There is a severe limit to the intelligence this style of AI can
set of symptoms achieve. Once the knowledge is written into the computer
program code, the system cannot be updated without going
• The pinnacle of this pattern-matching approach was IBM’s back and changing the code. No matter how many disease
Deep Blue, a computer system that beat Grandmaster cases they diagnosed, or gas pipe fractures they identified,
Gary Kasparov in several games of chess in 1997. It did this or games of chess they played, GOFAI systems couldn’t
by analysing its gigantic library of possible moves that it improve
could make, all of which had been preprogrammed by its

Machine learning and aIgorithms

• The approach that many modern AI systems use, however, is • When it comes to human learning, you probably learned
called machine learning and most of the AI that we use daily concepts or the names of objects from people around you
uses machine learning to produce its ‘intelligent’ behaviour when you were young

• The instructions within a machine learning AI system, • Without humans spending time correctly labelling things,
sometimes referred to as the algorithm, need to be trained many machine learning algorithms can’t learn at all, so the
point to note is that to make a machine behave intelligently,
• In the same way that humans are ‘trained’ to recognise a lot of human intelligence is required
similarities and differences as we grow up and get more
experienced, AI algorithms can be trained • Unlike Good Old-Fashioned AI, once a machine learning
system has been trained to a level that its developers are
• Preparing an algorithm to start to learn requires that the data confident is accurate enough to be used in the real world,
fed into the machine learning system is correctly labelled the system continues to learn and crucially, it improves

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 Explanations, definitions and some history
Transparency and the Black Box
• Now you might think with regard to GOFAI, why call an AI the algorithm in the machine learning AI system has done
that cannot learn artificially intelligent? The thing is, GOFAI or why it has reached a particular decision
comes with some significant advantages
• Machine learning is fast, and it learns, and has many
• GOFAI creates a trail of decision-making points as its rules advantages over GOFAI. However, the disadvantage of
are fired. Therefore, any decision that a GOFAI system not being able to provide an explanation for its decisions
makes can be explained. This makes GOFAI systems highly represents a significant problem, particularly for education
transparent to outside viewing and training activities, where it’s really important to be able
to explain, and often justify, why a particular decision has
• Machine learning AI, on the other hand, has no rules. It been made
can therefore be extremely hard to know precisely why a
machine learning system has made any particular decision • In an educational setting, it would be an absolute disaster
if you couldn’t justify or explain why certain actions were
• Machine learning AI systems are what we call ‘black box’ taken!
systems. There is no transparency available to show us what

Data and AI
• One of the other challenges that most machine learning AI • For example, data about the temperature of a classroom,
faces is the need for enormous amounts of data from which or about the time it takes people to get from one part of
the AI can learn the building to another, or data about light levels in each
classroom, or about which pupil regularly sits next to which
• As an example, let’s take an essay grading machine learning other pupil
AI system. To grade essays accurately a machine learning
AI would need to have processed millions of essays across • We appreciate that teachers are time-poor and thinking
the full range of possible grades. Do we have millions of about data sources may not seem like a good use of any time
examples of graded essays that cover the full range of they do have, but the truth is that it is a good use of time
possible marks? Are they in a digital format that can be
labelled and made accessible to the machine learning AI? We • In particular, if you are a school leader with many competing
may be able to collate sufficient examples, but it isn’t easy! demands on your time, it is easy to overlook the value of
The hefty data requirements of machine learning AIs are a data. We know that the very fabric of the building and all its
key restriction on their application resources are important assets in supporting the delivery of
learning
• So understanding the importance of data is key to engaging
with AI. Think about what data you have in your educational • Schools have inventories, logbooks, and financial data that
setting document the purchase of everything from PE equipment
and technology to the replacement of carpets and even the
• This could be data about individuals, such as how they number of chairs that are ordered. Schools are a treasure
perform academically. Or perhaps it’s data about how those trove of underutilised data
individuals are feeling, whether they are anxious, or whether
they are feeling confident • Dealing with issues as they arise can be costly and other
budgets can suffer as money is pulled from one source to
• There are many sorts of data available in any organisation, support another. Therefore, schools need to really examine
and quite often, some of the most obvious kinds of data are all the data they have, as often the solution to many of the
not really thought about when we ask the question: what logistical nightmares schools have can be found in the data
data is available? they hold

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 What AI can do in educational settings
The ETHICAL AI Readiness Framework has been In the first part of this document there was a small history
developed to prepare an organisation to leverage the lesson and an exploration of AI - an education on the subject.
transformational power of AI In the second part, the question is raised: what can AI do for me
and my school?
When it comes to engaging with AI in educational settings, the To answer that, you must be introduced to the the ETHICAL AI
authors of the book this Byte-Sized EdTech Research is based Readiness framework, and must ask yourself what challenges
on have developed the ETHICAL AI Readiness framework, a you face in your educational setting, and if they are something AI
7-step process for getting an organisation ready to leverage the can help you address.
transformational power of AI.

The 7-Step ETHICAL AI Readiness Framework

The AI Readiness Framework

Exploring the steps Each step must be carried out with rigorous ethics in mind

STEP 1:

STEP 1
The first step is to involve your organisation’s whole team in
learning about AI. What it is, what it can and can’t do, and how
it might help in your educational setting. Until now, teachers
have been in a position similar to the majority of the general
public. AI was a “too busy to bother” area that they might not
have understood or particularly trusted, and given competing
priorities, did not feel great pressure to understand. But with
educational technology being thrust front and centre during
the pandemic, there has been a shift. AI is now increasingly
E: Educate, Enthuse, Excite seen as a valuable tool for teaching and learning, and a critical
component of the knowledge people will need to flourish

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
Exploring the steps

STEP 2
STEP 2:

The second step is to focus your thinking around a particular

challenge your organisation faces, and consider whether it is
even something AI can address. 10 key prompts exist in the
framework to help you deep dive into your challenge, and will
TH: Tailor & Hone work to filter out what can and can’t be done and how AI might
help

STEP 3:

STEP 3 The third step comes into play once you’ve decided which
challenge you’re going to tackle. You and your colleagues must
think about what data you have to hand, and what state it’s in.
It’s likely your current data is spread all over your organisation,
and you might not immediately see the wealth you’re sitting on,
I: Identify so spending time identifying it, and making it fit for purpose, will
reward you tenfold in the long run

STEP 4
STEP 4:

The fourth step is to begin to collect new data as it pertains to

your challenge. This can involve data capture methods such
as surveys and interviews, but it must be done ethically, with
consideration for the people and sensitivities involved, as you
C: Collect must ensure bias does not creep into the dataset that will skew
your results for the entire process

STEP 5:

STEP 5 The fifth step is to apply your AI techniques to the data and start
to see what results emerge. Although it seems like this would be
the climactic part of the process, only 20% of your time will be
spent doing this, with the other 80% spent identifying, cleaning,
A: Apply and collecting all that data that powers the AI!

STEP 6
STEP 6:

The sixth step is about seeing the results of your AI applied to

the data from the challenge, and learning from them. This is
the most valuable step, as findings from the process will help
not only shape how an approach or solution is created for the
L: Learn challenge, but will help you understand your challenge in a
whole new way

STEP 7:

STEP 7 The seventh and final step (which doesn’t have a letter in the
ETHICAL acronym) is about looking at the findings you’ve just
received, and going back to step 2 again to check from the
ground up that your challenge and your AI are as specific and
informative as they can be. Remember that AI alone will rarely
yield a solution – far more likely it will be that AI combined
Iterate with human intelligence will have to work together to achieve a
solution

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 The 7 steps in context
that unites the commonalities at school, such as issues
Step 1 - Educate, Enthuse, Excite around attendance, attainment gaps, budgets, behavioural
interventions, essay grading, homework, recruitment and
more

• An AI mindset allows you to be open to the possibility that

• A brief history and exploration of AI and some of its terms
actually, if your data was studied, it might reveal something
can be found in this document in the previous sections, but
unexpected that challenged your preconceptions about what
it is important to reiterate that the first step of the 7-step
otherwise seemed like an obvious dilemma at school that
AI Readiness process is about educating, enthusing, and
you just hadn’t got around to investigating
exciting your team and colleagues so that you can view
yourselves through a ‘data and AI lens’ • Key Takeaway:
• You will develop an ‘AI mindset’ that allows you to view • Talk to your colleagues about data. AI is often a nebulous
the challenges in your educational setting as those which term with broad connotations that either scares or is too
may be solvable using AI. That allows you to see your rich, vague to be of meaning to people, but collecting and
complex data as not just siloed, unconnected, impenetrable connecting all that disparate data in your school could help
repositories of spotty information, but connective tissue you and your colleagues immeasurably in the long run

• A key aspect of the AI Readiness approach is to help you

focus your thinking on a thorny challenge you are currently Step 2 - Tailor & Hone
facing as a teacher, headteacher, or possibly a school
governor or trustee

• Maybe the challenges that you face are systemic and • The AI Readiness Framework uses ten key prompts to
process-based. Or perhaps they are rooted in an attempt to help teachers and heads focus on one particular challenge
shift your school from a traditional pedagogic approach to that might be addressed and better understood by the
an enquiry- or project-based pedagogy application of AI. In the book this Byte-Sized is based on,
you can find these prompts and the scoring metrics to help
• Maybe you’re:
prioritise them, which will help you decide on what to do
• Trying to recruit, train or retain the best staff next

• Analysing attainment gaps • List your assumptions during the exercise. Assumptions
can be hard to make explicit, as they’re often unwritten
• Concerned about bullying best guesses that have never been formalised, but they’ll
nevertheless shape the way in which a potential solution
• Trying to address the gender gap in maths starts to emerge
• Understanding learning analytics • Key Takeaway:
• Worried about pupil attendance • Your head might be swimming with all the different issues
• Unsure as to whether your marking and feedback engages you can see in your school, but performing an exercise to
students or actually helps them learn identify just what’s possible, and how much of an appetite
you and your colleagues have for the tackling of the
challenge, will help ground your expectations, and direct
your approach

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 The 7 steps in context
• Data can be unimodal or multimodal, quantitative or
Step 3 - Identify qualitative, structured or unstructured

• It is also important to avoid the ‘streetlight effect’ –

searching for data under an illuminated spot because
that’s the only area in which you have visibility, rather than
• In education and training, data is everywhere. You might
searching in the surrounding shade, which happens to be a
have data about the physical learning environment, the
much larger area
virtual learning environment, the curriculum, the pedagogy,
the use of resources, and much more besides • Key Takeaway:
• In addition, the connections that exist between these things • There is an understandable reticence about data collection
are also a form of data, as are the connections that exist within education: a worry that people who should not be
between these things and the people who are learning able to access the data that’s been collected will end up
seeing it, or that it will be misused. It is therefore extremely
• Ask yourself:
important that ethics is at the forefront of our thinking
• Can I learn about the sort of data I should collect for my when data collection and collation are being considered
challenge from existing research on the topic?

• What relevant data can I currently access?

• It is not always the best idea to select data sources that are
easiest to access. Many times, it is worth going to a little
more trouble to access data sources that are of the greatest Step 4 - Collect
relevance to our challenge

• Once we have identified works of relevance and we

have identified relevant data sources, the next step is to
synthesise all this data and information together. And don’t • Data can be collected by some common means, such as:
forget, it is not just the data that you yourself might have
• Surveys
collected or that from others in your organisation, there are
sources out there that are publicly available and that can • Interviews
contribute to your understanding
• But there is also multimodal data, such as video-enabled
• A few initial questions that need to be addressed for new platforms, which can allow analysis and reflection of
data collection could be: language used in classroom organisation or the addressing
of misconceptions, or the style of scaffolding provided to
• Who is going to have responsibility for collecting the
individual students, or student-to-teacher conversation
data? Perhaps it will be a course or module leader, or a
class teacher, a head of the department, or a teaching and • Key Takeaway:
learning policy lead
• Data collection needs to be designed carefully and
• Timeframes: when will the data be collected? Today, next must complement that data which is already available.
week, next month, next year? Thoroughly examining your challenge in the earlier steps
should allow you to frame exactly what data you do and
• Over what period of time will the data collection happen?
don’t want, so that you’re not wasting your precious time
• And it is important to consider that this data collection could or opening yourself up to risk, collecting something that
be occuring whilst teachers might be trying to keep up with will later turn out to be irrelevant
their schedules

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 The 7 steps in context
• Eye-tracking data from live online teaching
Step 5 - Apply • Survey responses

• Preparing this data is key and deciding what machine

learning AI technique to apply to it will depend on the
• With a number of the examples featured so far in this piece, context
the kind of machine learning AI that has, as a default, been
discussed, has been supervised machine learning • You may not want to use all the data you have with just
one AI technique anyway. You may end up applying some
• This is the type of machine learning used when you want to machine learning, and with the remaining data, using some
train the AI to find something specific in the data, such as a more traditional methods not based in AI
child’s face, or particular grade of exam script
• Human intelligence will need to be used to help clean –
• However, we do not always know exactly what we want label – the data as well, in removing errors, and feature
the AI to find in data, so we need another type of machine engineering
learning that can find patterns in the data: unsupervised
machine learning • Feature engineering is where humans help describe
patterns so that the AI isn’t scrambling about identifying
• This is the tool we use in a situation where we do not know commonalities with the data that make absolutely no sense
what we are looking for and so we cannot get the algorithm
to learn what the target data we want to find looks like • Key Takeaway:

• With unsupervised machine learning, the algorithm looks for • Unpacking what AI can do with the data that you’ve got
patterns, searching for similarities that might surprise us will let you make greater sense of both the data you’ve
collected, and the challenge itself. It may even reveal
• Data that might be fed into an unsupervised machine something in the data you had no idea was there. But it
learning algorithm could be: takes a lot of time to prepare the data, and if it isn’t clean,
you can get a lot of nonsense information out the other
• Log data from interactions with an online learning platform
end. With an increased understanding of your challenge,
such as mouse clicks
you will be in a much better position to select the AI tools
• Audio from student conversations in breakout rooms in and products you need to make your life easier in your
Zoom educational setting

• Performance data from tests and exams

• Once machine learning has been applied to your data, and

you’ve identified the patterns and relationships between Step 6 - Learn
data sources, you can use human intelligence to relate these
patterns to the particulars of your challenge

• For a simplified example, in an educational setting, that might

you hadn’t even thought to look for in, for instance,
mean discovering several different patterns, such as student
budgeting and spending, or teacher or student confidence,
profiles, through your application of machine learning, and
learning behaviours, recruitment, marking, or administration
then relating those profiles to answers from a student
survey. You might discover: • Acting on those findings, and being confident at an
organisational level that how you are tackling your challenge
• How confident each of the different student profiles are
is supported by the data, is a great and rewarding feeling,
when in particular teaching sessions
and moreover, it can help you organise your institution
• Do some profiles interact or learn more in peer-to-peer in a tighter, more methodical manner, and understand
sessions? the challenge better, so that you are not falling into traps,
constructing misconceptions, or making mistakes with
• How do the different profiles respond to feedback; is there similar challenges in your school
a correlation between engagement with feedback and
increased knowledge retention? • Key Takeaway:

• Has any additional technology boosted student confidence • Including human intelligence at all stages of the process of
based on those profiles? leveraging AI, from articulating your challenge, to cleaning
and organising your data, will mean that at the end of the
• Imagine how useful it might be to have all that siloed, AI Readiness exercise, you’re able to materially address and
unconnected data that you had at the beginning of the improve the conditions in which your challenge appears,
whole process now working in concert to show you patterns you’re able to understand it in greater depth, and actually
do something to solve it for people

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’
 The 7 steps in context
• Key Takeaway:
Step 7 - Iterate • Unless you do take your time to probe, prod, deconstruct,
and scrutinise the problem, it will be all too easy for you to
be persuaded that an AI product that automatically does
the marking for teachers, for example, is the solution. AI
• The step of the ETHICAL AI Readiness framework without a
is extremely accurate and fast. It uses absolutely the latest
letter in the acronym, step 7 is included to keep you and your
and most sophisticated deep learning to produce its marks
colleagues learning
• However, in this example, if the real problem is about the
• You should repeat steps 2-6 as many times as you need
quality of feedback that pupils require, then an AI marking
to, to increase your confidence that the findings from the
product is not going to help you. Similarly, if the real
application of AI are correct and effective
problem concerns the types of activities and assessments
• We can always learn more about the problems and that pupils are required to complete, then the AI marker
challenges that we face as teachers by exploring the relevant is not going to crack your problem either. And if the real
data, applying AI thinking and AI tools and AI techniques cause of the problem lies in the need for them to be able
to explain and justify their decisions, then forget the AI
• Once you see the problems and challenges that you face in marker, no matter how fast and accurate it is, as it will
your educational setting through a data and AI lens, then not be able to help you when it comes to justifying and
you will be able to make better decisions about how these explaining the marks it has allocated, let alone why they
problems and challenges can be addressed support a decision to revise learning goals

Conclusion
Much of the above text was adapted from AI for To find out more about how you can benefit from examining
School Teachers (Luckin, George, Cukurova, 2022), your institution through a ‘data and AI lens’, and leveraging the
available to buy from routledge.com transformational power of AI to tackle your challenges, order
your copy of AI for School Teachers at routledge.com, and
contact the AI and Data Science Team at EDUCATE Ventures
The above was a very brief summary of the 7-step ETHICAL AI Research at hello@educateventures.com
Readiness Framework and how it could help tackle some of the
challenges in your school. Remember, however, that this Byte-
Thanks for reading!
Sized EdTech Research piece is based on a whole book, and in
that book the authors use a holistic and iterative organisational
framework that has been developed by the team at EDUCATE
- The EDUCATE Ventures Research Team, March 2022
Ventures Research over several years.

Read more Byte-Sized EdTech at: www.educateventures.com

Cite as ‘Byte-Sized Edtech Research, AI for School Teachers, EDUCATE, 2022’