What is the now and what is the future?

I have concerns about generative AI. I am not a technological philistine, believe me, and my concerns are not actually about the technology itself. It’s not doomsday scenarios of computers taking over the world, or mass unemployment which concerns me (or at least not directly with the latter) – my concerns are centred around how humans interact with the systems and how they understand them. I feel there is genuine ignorance of what these systems actually are. Some of this is exacerbated by the ‘investor speak’ mentioned in the last post, some are from warped ideas people have from fiction and media consumption, and others – speaking as someone with some minor prowess in programming and even just Microsoft Excel has witnessed first hand – it is the ‘wizard’ effect, where anything ‘tech’ or ‘algorithmic’ or even simply ‘generated data’ is like some sort of infallible and incomprehensible magic. 

As much as I would like to write off the explosion of AI as Web 3.0 (cryptocurrency and NFT’s and the like), simply because of the permeating smugness from social media tech-influencers and venture capital types – I unfortunately can’t. Firstly, AI has been embedded into our lives for years now in the form of domain specific AI. This is AI designed to do a specific task where it mimics ‘human cognitive function’ and quite often involves machine learning, where the responses will ‘improve’ and change over time as the system gains more data. This has been a major part of social media algorithms, advertising algorithms YouTube video suggestions, text-to-speech software, predictive text, Siri/Alexa, statistical modelling software, camera filters, facial recognition and smart devices for a long period of time.

Secondly, the ‘big players’ in AI – those who are likely to see the largest gains are already key established players in the tech industry. OpenAI the group behind ChatGPT and DALL-E are bootstrapped with $11billion investment from Microsoft – and prior to transitioning from a ‘non-profit’ to the ‘limited company’ it is today, it had sourced donation support from Elon Musk, Amazon Web Services and Infosys – which those in the UK may recognise as the company founded by the father-in-law of Rishi Sunak. Now, all of these companies and individuals are not going to benefit financially from the growth of OpenAI, the latter investments viewed more as donations – but this was in-part to illustrate that the major ‘players’ in tech have been in on the ‘ground floor’ in generative AI, unlike Web 3.0. This is just one company – when you look at the endeavours of Apple, Meta and Nvidia you see a similar picture – established names in the tech industry with key AI based products or investment portfolio – and that is because the ultimate end goal of generative AI is ‘embedding’. Microsoft are already talking about it – the ghost of the paper clip returning to auto-generate the rest of your angry email to your bank about how you can seemingly no longer contact a human being to cancel a direct debit.

In some ways, this is what has increased the need for the ‘investor speak’ and harnessing of the publics misconceptions to push a narrative about AI, as those entering the market place late need to promise something different/better/beyond – or they’ve already missed out on ‘making it big’.

Realistically the long-term for AI is through embedding – which is why the established companies are all in – it will enhance offers they currently provide. Existing apps and programmes, search engines etc, but also into the ‘Internet of Things’, the collective name for all the stuff you own that for either an entirely logical, or for some unbeknownst reason connects to the internet; Fitbit’s, doorbells, baby monitors, TV’s, cat flaps and coffee machines * etc. How many people have a SmartTV because they were desperate for the technology, or they really believed in it? How many people have a SmartTV because the TV they happened to buy had smart features? You probably also have a use for it now, but until you had it, you may not have.

You will see people throwing out technically true but incredibly misleading statistics about ‘search terms’ or ‘downloads’/‘impressions’ when it comes to AI. Essentially how people searching for AI/ChatGPT is more than those who searched for iPhone at its peak, or comparing how it took only 6-months for ChatGPT to reach 100m users, but essentially a decade for home PCs to reach 100m sales. These comparisons ultimately prove nothing. ChatGPT is  a website to visit. There is no cost, there is no new technology for the user to face, it’s an add-on to what they already know. Yes there does seem to be some sectors that make use of generative AI, and I am not questioning the potential of the systems (see opportunities below) but realistically ChatGPT is the ‘main character of the moment’ to the general public, it is the zeitgeist of now; it’s Joe Exotic, Flappy Bird, Pokémon Go!, swing dance in the 90’s, 3D cinema (both times), it’s “Everything is better with bacon” or “Getting in my 10,000 steps” of 2023. That’s not to say it will not have longevity, uses and potential way beyond what we can do now – but the gimmick will fade, the public will move onto the next thing, but it WILL become normality and another part of our lives, and the Earth will not scorch (well not because of this at least).

So on that high point let’s get to it, opportunities and threats.

SWOT Pt.2:

OPPORTUNITIES:

• Future embedding into ‘traditional software’ and ‘apps’ could support learners, particularly those with additional education needs – providing different explanations/unique scaffolded support (eg. Rewording text, summarising an excerpt etc.)
• Additional scope for personalised learning – numerous platforms (particularly for mathematics) make use of programmed rules-based adaptive learning strategies based on performance – AI could be intertwined in this to further personalise support, for example choosing of specific questions with micro-level shared properties from a bank to practice, not just a ‘random selection’ at a higher topic level.
• Could be used in aspects of language learning – providing ‘live’ conversation opportunities in a more accessible way to learners from a variety of backgrounds who have less natural opportunities to engage in practice within their social circle.
• Potential to support increased statistical literacy and analysis in education – which often falls behind other professional areas – for example it could be used to combat the tendency of treating referential statistics like inferential statistics, which is rife in areas of the sector.
• Use of an embedding a personal dataset using the OpenAI API or making use of an embedding database – either personally or by using a third party, to attempt to focus on specific needs of an institution, qualification or assessment rubric. Seems to be an approach by the amazing folks at Khan Academy and No More Marking – but again, not without issues – find more information here –
• Opportunities to develop for broader assessment styles and update curriculum that can develop further skills and competencies in learners, give a broader scope to assessment methods that could have greater inclusivity long-term – eg. presentations, viva/oral exams, open book exams etc.

THREATS:

• My biggest concern is the potential for policy makers and school leaders to attempt to use ChatGPT and AI as a “silver bullet’ to address workload/stress in the profession. This could potentially lead to no attempt of addressing actual issues, and putting the responsibility and blame back onto workforce for workload as there is definite potential for the “This would fix it, only if you did it properly”.
• A lot of investor speak shrouds the like of ChatGPT and AI in general – and this is unhelpful. It’s always about “It will do this soon” or “You’re just not using it right” or about “Jobs of the future” without actually any of it seemingly coming from the ground in those sectors themselves. Potential is potential, but it’s difficult to see what is actually real and what is just hysteria whipped up for for the capital investor market.
• This hysteria now leads to people believing the systems do more than they actually do, or are better at things than they actually are – the people showcasing what it does often aren’t the ones who “know” how good the output is or how applicable it is to a classroom scenario. The nuance of pedagogical content knowledge is missing from output.
• Despite this, it still being used to put unsuitable teachers into the profession/into certain curriculum areas to address staff shortages relying on this – ultimately providing a lessoned quality of education for learners.
• There is a lot of talk of opening up the ability for ‘developing higher skills’ in students, without explicitly saying what and how, by using what is currently available in the marketplace. There is almost a risk to the extension of the ‘just Google it’ culture that facts and basic skills are unimportant – when they are the necessary building blocks students have to acquire in order to access and develop these higher order skills, but also to actually verify the information that they receive.
• Exam boards and policy makers need to quickly address weaknesses in certain assessment methods in light of ChatGPT influence and develop curriculum to support ethical use in academia and schooling – but quick decision making can embolden 
• AI models have generally been set-up to identify correlations or patterns – why they are there or what they mean are not actually important. Why YouTube suggests you a video is not important to YouTube (as long as you watch it and stay on the site), why a certain item in your purchase history means you are more likely to miss a mortgage repayment is unimportant to your lender when they calculate the risk of the loan. The same thing can be said of these generative models as well, ChatGPT etc. essentially throw words together based on probabilities – why they are associated is unimportant – they just know statistically these words ‘go together’, they follow a pattern on what existed on the dataset and are ‘associated with the words in your prompt’ – there is no comprehension. This leads to potential bias based on the datasets used to train the models – which some recently identified issues regarding large scale synthetic data use to train models could spell problems for models of the future – https://arxiv.org/pdf/2305.17493v2.pdf (This paper does not seem to be peer reviewed to the best of my knowledge).
• The “Do your marking and feedback” promise which is often banded about – maybe by making use of the API to train on specific data – but as the likes of No More Marking have shown, it is not as accurate as a human, nor is the feedback as productive (there are a lot of posts relating to this on their substack – https://substack.nomoremarking.com/p/ai-powered-essay-marking). However, marking and giving feedback in this way also misses a HUGE aspect of the teacher experience – marking is of itself a diagnostic tool, and can be of vital importance for a teacher to understand their classroom and their students. Even a summative assessment can be formative ‘to the teacher’.
• In a broader sense it is a risk of further ‘Balkanisation of the internet’, ring-fencing of knowledge for commercialisation purposes. Why visit information first hand, when this one site can just ‘tell you’?

In a research report published by the DFE in April 2023, they highlight that 86% of respondents say that they experience stress due to their work, 65% say they do not have adequate time in their personal life, 56% say the job negatively affects their mental health and 45% say the job affects physical health. Furthermore, only 17% of those surveyed felt that they had an acceptable workload. There is a retention and and recruitment issue. Whether due to ignorance or something more sinister, as I said in the first bullet point, my main concern is the sector pushes these tools on staff as the ‘fix’ and does nothing to improve address the actual issues of workload, pressures and stress.

In May 2023 Ofsted also released a report based around an ‘Independent review of teachers’ professional development in schools’, in the report they highlight that 87% of respondents felt that workload was a barrier for undertaking Continuous Professional Development. Thrusting AI systems, as user friendly as they are to interact with at a basic level, is only going to exacerbate this. It’s not about the first response, you have to learn how to use the system, to interrogate, to get what you want – it involves new skills. This needs training or time to practice. Even IF this was the magnificent fix (it isn’t) the current culture is not set up to support and promote engagement in this for teachers besides in a superficial way.

In the very first paragraph of this post I said I am not a technological philistine, my concerns are centred around how humans interact with the systems and how they understand them. This is why a range of voices in this DFE consultation is vitally important, otherwise you get the risk of the most enthusiastic group dictating the implementation with little oversight or realism to application. So again, find the consultation here, it is open until late August – https://consult.education.gov.uk/digital-strategy/generative-artificial-intelligence-in-education/consultation/intro/

There are some great things that generative AI can do, but there’s nothing that I feel will immediately change in my teaching due to this besides a summer reflection on my assignment based teaching programmes and what I want students to evidence in their summative assessments and whether the formats are still the best way of doing that given the emergence of generative AI.

I will return to AI – I had a lot more I was originally going to say here with links to research and how a lot of the proposed AI benefits actually contrast with elements of what we understand as high quality teaching and learning. I was even guilty of this myself when listing potential strengths in part 1. Ultimately, this post was getting a bit long and I’m sure you’re bored of me by now.

Until then, I’m off to make a coffee…not over the internet.

Links and Further Reading:

DFE Research Report, Working lives of teachers and leaders – wave 1 (2023) – LINK

Ofsted, Independent review of teachers’ professional development in schools: phase 1 findings – LINK

No More Marking – https://www.nomoremarking.com

No More Marking (Substack – writings of Daisy Christodoulou and Chris Wheadon) – https://substack.nomoremarking.com