Taking a look under the AI hood

In my personal blog, 21st Century Chalk, I talked in some detail over two posts about AI in society in general and AI in the education sector more specifically. For a longer look at my thoughts and research, check it out. But in summary, I mentioned how it is a great technology, but not the saviour of all things that some may be professing it to be. I highlighted many of its issues and described its shortcomings. And in this age of data-hungry systems, it is important that people know what they are using and how they are being used. After all, ‘if you are not paying for the product, you are the product’.

It is for this reason that I think it is important for developers to be open about their data use, and in this current era, their AI use. And not just with how their apps use AI, but also how they used AI to develop their apps.

The Three AI Features 

1. The first implementation of AI is in Story Mode. The words from chosen word lists are fed into an AI, along with some optional prompt words and a story is generated that the user can spell along to. 

2. The second implementation is in the daily challenge word definitions. This is a minor use of AI but extremely useful. Once a word is spelt correctly, the word is first fed into Merriam Webster dictionary API to fetch the definition. The definition that comes back is usually written at too high a level for younger spellers to understand so the definition is then fed into an AI to rephrase the wording, making it more accessible. It also generates an example use for that word, for further understanding. This is extremely useful for many red and black level daily challenge words.

3. The final implementation is the smart hints system. This is only available when you get a word wrong. It is most useful after repeated errors as it takes all previous errors and information about the word list and the AI generates some advice for the speller. This has the potential to be very useful depending on the learner and the teacher managing the learning.

My need-based approach

I am considering other AI uses, but with care. There are a lot of features that I thought would be totally controlled by AI in SpellMe, but they are not. For example Mastery mode, where the word levels dynamically change depending on student performance, is done with zero AI. I don’t even think using AI would enhance the feature. 

There are far too many apps and companies forcing AI into their products just for the sake of it, without that feature being useful to anyone or anything. It’s a tool looking for a problem to solve, rather than a tool for an existing problem. Not everything needs AI. So, for now this will be all the AI features that will be implemented in SpellMe. Any others will undergo scrutiny to see if a clever algorithm can do the same job or if it is even necessary. There will be no AI-powered emojis for me! (at least not yet :P)

In my next post I will discuss how I use AI to develop SpellMe, and the difference between vibe coding and collaborative AI coding.