AI in HE
UR Generative AI Day
8 June 2023
Prof Miles Berry
These slides: bit.ly/uraidayslides
For some, or for all?
Children need to be adequately prepared for working with, and using, AI. For a proportion, this will mean a thorough education in AI-related subjects, requiring adequate resourcing of the computing curriculum and support for teachers. For all children, the basic knowledge and understanding necessary to navigate an AI driven world will be essential. In particular, we recommend that the ethical design and use of technology becomes an integral part of the curriculum.
AI fo accessibility and inclusion
The future is already here — it's just not very evenly distributed.
William Gibson
ZPD
What the child is able to do in collaboration today he will be able to do independently tomorrow.
Vygotsky, 1978
Images
Cloud Vision API
Dual coding…
Dual coding theory (DCT) explains human behavior and experience in terms of dynamic associative processes that operate on a rich network of modality-specific verbal and nonverbal (or imagery) representations. We first describe the underlying premises of the theory and then show how the basic DCT mechanisms can be used to model diverse educational phenomena. The research demonstrates that concreteness, imagery, and verbal associative processes play major roles in various educational domains: the representation and comprehension of knowledge, learning and memory of school material, effective instruction, individual differences, achievement motivation and test anxiety, and the learning of motor skills. DCT also has important implications for the science and practice of educational psychology — specifically, for educational research and teacher education. We show not only that DCT provides a unified explanation for diverse topics in education, but also that its mechanistic framework accommodates theories cast in terms of strategies and other high-level psychological processes. Although much additional research needs to be done, the concrete models that DCT offers for the behavior and experience of students, teachers, and educational psychologists further our understanding of educational phenomena and strengthen related pedagogical practices.
Speech
3.6 To access the curriculum, early literacy provides fundamental knowledge; reading comprises two elements: word reading and language comprehension; systematic synthetic phonics is the most effective approach for teaching pupils to decode.
Languages
Google Translate
Bellos, 2011
A new common language?
Google’s researchers think their system achieves this breakthrough by finding a common ground whereby sentences with the same meaning are represented in similar ways regardless of language – which they say is an example of an “interlingua”. In a sense, that means it has created a new common language, albeit one that’s specific to the task of translation and not readable or usable for humans.
Language
GPT, Bard et al.
Other tools are available
GPT3.5, GPT4 -> ChatGPT
GPT3.5 -> Bing Chat
GPT4 -> Copilot
LaMDA -> Bard
DALL.E 2 / Bing Image Creator, �Firefly, Midjourney, Stable Diffusion
Prompting well
Start with a verb (command words)
Provide context
Role play
Use references
Use quotation marks for phrases
Be specific
Give examples
Give an example of how you have (or how you could) use Chat GPT or similar in your work as an academic.
bit.ly/uraiday
Jill Watson
Dubbed Jill Watson, the AI teaching assistant was based on IBM's Watson platform. Jill was developed specifically to handle the high number of forum posts by students enrolled in an online course that is a requirement for Georgia Tech's online master of science in computer science program.
Georgia Tech, 2016
Coming soon…
CS50 … plans to use artificial intelligence to grade assignments, teach coding and personalize learning tips, according to its Professor David J. Malan... Even with more than 100 real-life teaching assistants, he said it had become difficult to fully engage with the growing number of students logging in from different time zones and with varying levels of knowledge and experience. "Providing support tailored to students' specific questions has been a challenge at scale, with so many more students online than teachers," said Mr Malan, 46.
His team is now fine-tuning an AI system to mark students' work, and testing a virtual teaching assistant to evaluate and provide feedback on students' programming. The virtual teaching assistant asks rhetorical questions and offers suggestions to help students learn, rather than simply catching errors and fixing coding bugs, he said.
Bloomberg, 3/6/23
Some academic uses
Module specifications etc
Teaching materials
Readings
Feedback on work
Proposals and bids
Abstracts, précis, critique
Lecture or course planning
Keep people in the loop
The data subject shall have the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.
GDPR, 22:1
There are limits
It doesn’t really understand
It doesn’t really think - problem* solving is a problem
It’s not great at maths
It’s over-confident
It does make things up
GPTn are not up to date
Reliability costs
Uses
Applying
Suggesting, summarising, simplifying and critiquing readings
Suggesting ideas and structures
Helping with data analysis
Improving writing
Weekly reflections and directed tasks
Cheating
Personal statements
This form of assessment has gained popularity due to its numerous benefits, which include efficiency, objectivity, and adaptability.
Firstly, MCQ assessments are highly efficient in terms of time and effort. They allow for the examination of a vast range of content within a relatively short time frame, providing a comprehensive assessment of a learner's knowledge and understanding (Simkin & Kuechler, 2005). This is particularly advantageous when testing large groups, as it reduces the workload for both the test taker and the evaluator (Scouller, 1998).
Secondly, MCQ assessments offer a high degree of objectivity. Since multiple-choice questions have a predetermined set of responses, scoring is straightforward and less susceptible to subjective bias (Martinez, 1999). This ensures that all test-takers are evaluated fairly and consistently, promoting a sense of equity among examinees (Burton et al., 1991).
MCQ assessments are also highly adaptable to different learning contexts and subject areas. They can be used to assess a wide range of cognitive skills, from basic knowledge recall to more complex problem-solving abilities (Haladyna et al., 2002). Additionally, MCQs can be easily integrated with technology, facilitating the implementation of computer-based testing and adaptive testing models, which can further enhance their efficiency and effectiveness (Wainer et al., 2000).
Despite some criticisms, such as the potential for guesswork and the focus on lower-level cognitive skills, MCQ assessments continue to provide valuable benefits for educators and test takers alike (Tarrant & Ware, 2010). Their efficiency, objectivity, and adaptability make them an indispensable tool in evaluating a wide range of knowledge and abilities.
The Imitation Game
I believe that in about fifty years’ time it will be possible to programme computers, with a storage capacity of about 109, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent, chance of making the right identification after five minutes of questioning.
Turing, 1950
Is it cheating if ChatGPT…
Explains something to you?
Gives you ideas for an essay?
Suggests how to improve your essay?
Writes the essay for you?
Is it cheating?
3.2 In particular, academic misconduct may include, but is not limited to:
a. Plagiarism: presenting another person’s published or unpublished work in any quantity without adequately identifying it and citing its source;
b. Duplication: resubmitting work in any quantity without acknowledgement or without adequate redevelopment to make it novel and appropriate to the assessment, including the resubmission of work which was previously submitted at another institution;
c. Falsification: inventing or altering facts, data, quotations or references without acknowledgement;
d. Collusion: assisting another student, or being assisted by another person, in gaining an unfair advantage in an academic assessment;
e. Failing to comply with ethical guidelines or requirements, including those set out by the University and any relevant external bodies;
f. Cheating: engaging in conduct that sets out to undermine the security, integrity or fairness of an assessment; this includes obtaining, introducing, using or sharing information or materials without permission.
g. Contract cheating: contracting with another individual or body to receive or provide work in exchange for compensation of any kind, including payment.
What students see
“I understand that academic misconduct is strictly prohibited including plagiarism, essay mills, generative AI, collusion, impersonation, and falsification or any other action which might give me an unfair advantage. I confirm that all the work is my own unless collaboration and/or generative AI has been authorised by my tutors. Where I have drawn on source material, that material is referenced according to the University guidelines. This work has not been submitted as part of another assessment.”
Burden of proof
If the student admits to using generative AI, or the evidence indicates a case to answer, a referral should be made for a formal investigation under the Academic Misconduct Procedure via the Programme Convenor. If they don’t admit to using generative AI and the academic and disciplinary officer have insufficient evidence to pursue a case against the student, the work is marked as usual and there are no penalties applied.
The sum of the primes below 10 is 2 + 3 + 5 + 7 = 17.
Find the sum of all the primes below two million.
Any teacher* that can be replaced by a computer, deserves to be.
Arthur C Clarke�David Thornburg
Any questions?
m.berry@roehampton.ac.uk
@mberry
These slides: bit.ly/uraidayslides