Warwick/phd-interview
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Made more changes based on mikes feedback

Browse source
This commit is contained in:
Warwick New 2026-04-23 13:03:22 +01:00
parent e8de0f559d
commit cf9883f7f0
2 changed files with 64 additions and 33 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

BIN
progprac.png Executable file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 79 KiB

97
slides.tex
View file

@ -3,6 +3,7 @@
\usepackage{tabularx}
\usepackage{booktabs}
\usepackage{csquotes}
\usepackage{graphicx}
\usepackage{xcolor}
\newcommand\todo[1]{\textcolor{red}{#1}}
@ -25,16 +26,17 @@
linkcolor=blue,
filecolor=magenta,
urlcolor=blue,
pdftitle={Overleaf Example},
pdftitle={PhD Interview},
pdfpagemode=FullScreen,
}
% Referencing
\bibliographystyle{IEEEtran}
\title{PhD Interview Presentation}
\title{Can the use of a virtual avatar and scaffolding with AI agents improve
first-stage programming students' debugging ability?}
%\subtitle{aa}
%\date{Week 1}
\date{23rd of April 2026}
\author{Warwick New}
\begin{document}
@ -44,22 +46,40 @@
\begin{frame}
\frametitle{What is the problem space?}
Students self perceptions about their programming ability largely affects
several factors in deciding to continue learning to program \cite{lewis2011}.
Novices tend to struggle with programming fundamentals, finding the basics
quite difficult to learn, and so introductory courses tend to have poor rates
of retention.
\newline
Students self perceptions about their programming ability affects student
retention and increases the rate of students that drop out \cite{lewis2011} in
their first stage.
\newline
A particular point of frustration for novice programmers is attempting to
debug without thinking of the problem critically \cite[p.~23]{vickers2008}.
\newline
Recent literature has identified this as a problem space that generative AI
can attempt to help resolve. Renzella et al's work \cite{renzella2025}
demonstrates that there are aspects of learning that traditionally could only
be fulfilled by an instructor outside highly controlled learning
environments (like Gidget \cite{Lee2014}) that can be partially automated with
generative AI. And therefore can be made more available to students outside
time tabled sessions.
%in this case reducing the cognitive load \cite{sweller1988} of reading and understanding compiler messages by providing more human-readable descriptions of compiler errors.
Previous interventions in this space have shown promise such as Gidget
\cite{Lee2014} that has used personification of compilers to great effect and
Metacodenition \cite{Pechorina2023} which helps the student create a mental
model about actions to take when a bug appears.
\newline
Recent literature has begun exploring the use of LLMs but out of the box,
found them in most cases to be worse than the alternatives.
\cite{Pechorina2023}. Though recent literature has found that there is a
benefit to providing code explanation outside business hours
\cite{Renzella2025}.
%Recent literature has identified this as a problem space that AI
%can attempt to help resolve. Renzella et al's work \cite{renzella2025}
%demonstrates that there are aspects of learning that traditionally could only
%be fulfilled by an instructor outside highly controlled learning
%environments (like Gidget \cite{Lee2014}) that can be partially automated with
%generative AI. And therefore can be made more available to students outside
%timetabled sessions.
%%in this case reducing the cognitive load \cite{sweller1988} of reading and understanding compiler messages by providing more human-readable descriptions of compiler errors.
\end{frame}
%\begin{frame}
@ -90,22 +110,29 @@
\begin{frame}
\frametitle{What is the Gap}
Use of generative AI to learn how to debug isn't new in research. But there
are some lessons learned from educational programming interventions applied in
tools that could be made more accessible outside of sessions and scaled to a
wider range of programming environments that we could apply, which has the
potential to improve self-efficacy within the student base. Including:
{\footnotesize
Two practices in helping students learn to debug have shown great promise when
it comes to improving self-efficacy and lowering programming anxiety. I
believe that we could use recent advances in AI to augment aspects of these
interventions.
\begin{itemize}
\item Personification of programming tools as fallible and encouraging, as
presented by Lee et al \cite{Lee2011}, is shown to have a positive impact
on learning motivation and success.
\item Scaffolding what steps to take when you encounter a bug and
encouraging the student to think through them one step at a time, improves
self-efficacy and productivity as demonstrated in Pechorina et al's work
on Metacodenition \cite{Pechorina2023}.
\item Personification of programming tools as fallible and encouraging, is
shown to have a positive impact on learning motivation and success as
shown in Gidget \cite{Lee2011}.
\item Scaffolding what steps to take when encountering a bug, improves
self-efficacy by lowering cognitive load as demonstrated Metacodenition
\cite{Pechorina2023}.
\end{itemize}
Both of these works are restriceted to very specific environments limiting
their application on real world coding projects.
%Both of these works are restriceted to very specific environments limiting their application on real world coding projects.
\begin{figure}
\centering
\includegraphics[width=0.45\textwidth]{progprac}
\caption{Figure 2 provided by Scott et al. \cite{Scott2014}}
\label{fig:question}
\end{figure}
This figure shows how programming self-concept influences programming anxiety
and increasing time on task, retaining more computing students.
}
\end{frame}
%\begin{frame}
@ -182,10 +209,14 @@
cognitive load \cite{Sweller1988}.
\newline
This will be done by walking a student through any issues a
real programming environment can throw at them with the use of an AI agent
based avatar, which will break down the process of solving debugging issues in
their own real world programming environment.
This will be done by walking a student through any issues a real programming
environment can throw at them with the use of a conversational AI agent based
avatar, which will break down the process of solving debugging issues in their
own real world programming environment.
The key feature of the agent being that it has been created with pedagogic
strategy in mind, namely scaffolding and personality to see if AI can help
fill these gaps when structured effectively.
\end{frame}
\begin{frame}
@ -207,8 +238,8 @@
that has then since gained access to the artefact in a longitudinal study
\cite[p.~224]{Coe2025}.
\begin{itemize}
\item Cohort A -- Pre-intervention Cohort
\item Cohort B -- intervention Cohort
\item Year 1 -- Pre-intervention Cohort
\item Year 2 -- intervention Cohort
\end{itemize}
\end{frame}