麻豆社区

The development in the field of AI is faster than previously expected

It is not entirely surprising that technological development is faster than expected, and perhaps especially not within the field of AI. In a recently published survey from October 2023 (2023 ESPAI), answered by 2778 researchers who have published articles in highly respected journals within the field of AI, respondents were asked to give their estimates on how AI within a range of areas will develop and at some point, achieve competencies equivalent to a human level, or better. An example in this context could be an AI Researcher. These answers, compared with the answers to the same questions from August 2022 (2022 ESPAI), suggest that development is significantly faster than previously expected, this applies within all the categories asked, some however significantly more than others. For example, it is expected by the survey's respondents, that the limit for a 100% AI generated publication on New York Times’ bestseller list has moved by almost ten years. So this should be a possibility already in 2030 and not in 2040 as assumed in 2022. While an AI Researcher is only expected to be a reality around 2050, however moved closer by about 10 years. The respondents themselves in the survey also express being somewhat concerned about this development, especially in relation to how generative AI can be used to manipulate and spread misinformation (Grace et al., 2024).

With the emergence of ChatGPT at the end of 2022, it is not so surprising that the time limit for a 100% AI generated publication on New York Times’ bestseller list has moved so significantly from 2022 to 2023. Especially not when you look at how the development within this area of AI, the generative AI language models (LLM), is progressing. The first broadly accessible LLM, GPT-3, which was launched by OpenAIand in itself was very groundbreaking, was followed in the spring of 2023 by GPT-4, which in addition to being online and not only building on historical data. GPT-4 can of course do even more, is even better and even faster, and is still being developed and optimized (OpenAI, 2023).

Mollick (2023) conducted in March 2023 an experiment of how much he could do with various GPT-4 based generative AI tools in 30 minutes, based on a business-oriented case. He got an email containing an overall marketing strategy, a website with pictures, a SoMe strategy, with accompanying posts for the different platforms, and a marketing video, and the quality was, according to Mollick's own statement, relatively reasonable. In itself an impressive achievement, but even more impressive is what he almost a year later, with the latest tools, among others Microsoft copilot, generates in under a minute. It is for example a PowerPoint presentation, with notes, based on an AI generated business case about Tesla, a full teaching plan in Word for six teaching sessions on entrepreneurship, including lesson overview, learning objectives and tasks, and a market survey on Virtual Reality formulated by copilot in the Edge browser (Mollick, 2024).

A small look at the area within AI generated images and videos also underlines why there, as mentioned earlier, not at least among professionals is a growing concern. With OpenAI's latest addition, , it is now possible to make 100% AI generated videos from entered text. However, the technology is still in its infancy and still makes a number of recognizable errors with for example hands and general physical laws, but it has been seen before in the beginning with AI generated images of for example . So, it is probably just a matter of time, before the technology is optimized, as is the case for the other tools. Sora has not been released to ordinary users, as OpenAI is trying to incorporate some before this happens.

The e-learning course  provides a basic introduction to what generative AI is (and isn't) and an understanding of what it can (and can't) do (for all employees considering using generative AI).

Pedagogical Opportunities and Challenges

Overall, a fascinating technological development, which in relation to teaching provides a wide range of challenges, including significant challenges in relation to academic integrity, ethics, access to technologies, bias in the AI generative models and not at least examination (The Future of AI in Higher Education, n.d.). In March 2023 we published a newsletter on these topics focusing on Teaching and learning with AI, and the challenges dealt with at that time are certainly no less relevant, in that the tools have become significantly better and more accessible - also for the students who in a way now have a "scientific assistant" right at hand. This can be a pitfall for some students, while for others it can be an invaluable helper.

The workshop offered by SDU UP provides teachers inspiration on how to assess and redesign exam questions with generative AI in mind (the workshop is for groups of teachers from the same faculty).

Since March 2023, however, a lot has happened in relation to the opportunities and any software company with an amount of self-respect, as well as a forest of startups, now focusing on how AI can be used within an existing or new niche of teaching activities. In addition to the previously described activities for content generation and the directly integrated assisting tools in standard products like copilot to and . There are for example also a wide range of products for help with academic writing such as , an old acquaintance now with AI, and that can give references from a given text. In a slightly different area, we have tools like for analysis of qualitative interviews, now also with AI, and for literature search, for which SDU actually has a license - but no data processing agreement with. An exciting product that also deserves to be mentioned is , which can be extremely useful for source criticism of news articles, a field we are some who look forward to being further developed. Digital literacy and thus also the preparation of the students for a working life with the use of AI should be thought into the teaching (The Future of AI in Higher Education, n.d.).

As a teacher, it will, in addition to the previously mentioned tools, also be obvious to look at some of all the new, or old – now with AI, tools for example for feedback activities and automatic corrections, like , , which are now  implemented on itslearning, however without AI functions, or with a mathematical focus, just to name a few.

The workshop offered by SDU UP aims to make it easy and appealing for teachers to start using generative AI tools in their teaching and supervision (the workshop is for groups of teachers from the same faculty).

Remember .

A very important detail is that we at the present moment neither have data processing agreements with or licenses to most of all this wonderful software, which can make it somewhat difficult to use, especially together with the students. SDU Digital is working hard to approve some of them - primarily Microsoft products. When this is said, it certainly does not exclude the possibilities to explore and use different solutions – as long as it is done within SDU's guidelines We 麻豆社区 UP are happy to assist with this and will in this connection be hosts and facilitators on a series of focusing on the use of generative AI in teaching.

The learning circle course offered by SDU UP enables individual teachers - together with other teachers in a small group - to design and experiment with their teaching/supervision with and in generative AI.

Referencer:

Grace, K., Stewart, H., Sandkühler, J. F., Thomas, S., Weinstein-Raun, B., & Brauner, J. (2024, January 5). Thousands of AI authors on the future of AI. arXiv.org. 

Mollick, E. (2023, March 25). Superhuman: What can AI do in 30 minutes? One Useful Thing. 

Mollick, E. (2024, January 31). What Can be Done in 59 Seconds: An Opportunity (and a Crisis). One Useful Thing. 

OpenAI. (2023). GPT-4 Technical Report. arXiv (Cornell University). 

The future of AI in higher education. (n.d.). EDUCAUSE.