Generative AI experiments 1

Generative AI

robot, by Terry Freedman

In a recent interview in the Financial Times, the science fiction writer Ted Chiang said that AI models are trained by toggling “weights” or the strength of connections between different variables in the model. He went on to suggest that a better name than artificial intelligence, albeit not as sexy, would be Applied Statistics. (FT Interview)

I think he's right. I'm as guilty as anyone else at using anthropomorphic language to describe what apps like ChatGPT do -- although I usually go to the trouble of placing words like "written by..." in quotation marks.  But ultimately they can't think for themselves. In my opinion, the biggest danger of generative AI is not the fear of its becoming conscious and taking over the world, but of its users or "recipients" failing to use common sense and critical thinking.

As a small example of this, I asked ChatGPT to draw up some course outlines, and compared them with my own versions. In one of them it "misunderstood" (there we go again) the nature of the course, but that was probably a reflection of my insufficiently thought-out prompt. Overall, it did pretty well, but failed to insert any breaks. It did once I'd asked it to, but my point is that although I can see the potential of AI helping to create a course outline, or a computer programming debugging test (see below), it would be a grave error to delegate our own responsibly to it entirely.

Embedded generative AI

A development that I think is potentially interesting is the embedding of generative AI in the tools we already use. I have to say that I haven't been too impressed by Bing's AI Chat or Google's Bard, in terms of the answers it gives, but I will say that it's nice tio have the answers served up rather than a list of websites where I have to extract the information myself. 

I think it's a good idea to compare the results of different applications as well as applying your own judgement. For example, I recently asked both Google Bard and Bing Chat how we can increase the number of teachers who are qualified to teach computing. This is how they answered:

Bing

There are several ways to increase the number of teachers qualified to teach computing. One way is to provide financial support to people training as computing teachers and recently announced partnerships with industry leaders to train over 45,000 computing teachers1. Another way is through the BCS computer teacher scholarships which seeks to grow the number of specialist Computing teachers in England2.

I hope this helps. Let me know if you have any other questions.

Bard

Here are some ways to increase the number of teachers qualified to teach computing:

  • Offer more training and development opportunities for current teachers. Many teachers who are not qualified to teach computing may have the skills and knowledge necessary to do so, but they may lack the teaching experience or pedagogy. Offering more training and development opportunities can help these teachers to develop the skills they need to be effective computing teachers.

  • Create more pathways to teacher certification in computing. Currently, there are a limited number of pathways to teacher certification in computing. This can make it difficult for people who want to teach computing to get the necessary credentials. Creating more pathways to certification would make it easier for more people to become qualified to teach computing.

  • Increase the salaries of computing teachers. The salaries of computing teachers are often lower than those of teachers in other subjects. This can make it difficult to attract and retain qualified computing teachers. Increasing the salaries of computing teachers would make the profession more attractive and help to ensure that there are enough qualified teachers to meet the demand.

  • Partner with industry to develop training programs. Industry partners can play a valuable role in developing training programs for computing teachers. They can provide access to experts, resources, and equipment that can help teachers to develop the skills and knowledge they need to be effective.

  • Make computing a more visible subject in teacher education programs. Computing is often not given the same level of importance as other subjects in teacher education programs. This can make it difficult for teachers to develop the skills and knowledge they need to teach computing effectively. Making computing a more visible subject in teacher education programs would help to ensure that all teachers have the foundation they need to teach computing.

By taking these steps, we can increase the number of teachers qualified to teach computing and help to ensure that all students have access to a high-quality computing education.

Comparison

Comparing the results, Bing was less detailed but included links to the original information. It also included a handy button to copy tghe text, and one to download it.

On the other hand, the Bard version is more detailed, but I have no idea where the information came from (well I do, but you know what I mean).

Both versions have come up with points that could be discussed with the appropriate people.

Embedded AI

Embedded AI is also making its appearance in applications. It can be used in Excel, for example (I haven't tried that myself yet, but I intend to). A newer version of Mailerlite, which I use for writing and sending out this newsletter, has an AI bot in the text editor. I haven't tried that because I don't wish to "upgrade", but Squarespace, which I use for my websites, has introduced something similar. I can’t see me using it.

Recent articles: 

Introducing ChatGPT. This article is for Digital Education subscribers. It looks at aspects such as whether an AI-written article is detectable, and checks for plagiarism.  It's in the Digital Education Supplement section.

The following articles recently appeared on the ICT & Computing in Education website:

The Magic Of Human-Computer Communication (Revisited)

Why Schools Should Have A Hyperlocal Blog (Updated)

Testing Pupils Using Stories (Update) -- in which I used ChatGPT to create a debugging test.

This article first appeared in Digital Education, my free newsletter.