Generative AI and the Science of Reading

Learning to read requires excellent texts that are carefully constructed to combine a few elements, interest, introduction of new vocabulary, and control over the level of difficulty, including the use of specific phonological and orthographic patterns for early readers. This has always been a source of difficulty, book publishers have done a reasonable job, but the costs for schools and classrooms have been prohibitive, especially when considering the need to send students home with texts to practice reading.

TextProject from Elfrieda Hiebert has been producing such balanced texts and making them available for free.

Generative AI presents the opportunity to create such texts readily.

for example I asked chatGPT to create a first grade level text about dogs:

All About Dogs

Dogs are fun friends to have. Some are big and some are small. Some dogs have long hair and others have short hair.

Dogs love to play. They like to run, jump, and fetch balls. Some dogs like to swim in water, while others prefer to stay dry.

They have wet noses and wagging tails. When a dog wags its tail, it is often happy. If a dog shows its teeth, it might be scared or upset.

Dogs can hear very well. They also have a good sense of smell. This is why some dogs are used by police to find things.

It's important to be gentle with dogs. We should give them food, water, and a cozy place to sleep. In return, dogs give us lots of love and joy.

Dogs are not just pets; they are part of the family!

I followed up with a request to make it rhyme.

**Doggie Delight** Dogs are friends, both big and small, With furry coats, short and tall. They love to jump, run and play, Fetching balls and barking away. Some love the splash, some stay neat, With wagging tails and tiny feet. When tails wag, they're feeling glad, But bared teeth mean they're somewhat mad. With ears so sharp, they hear a pin, And noses that sense where they've been. Police dogs with a sense they hone, Find things hidden, all on their own. Treat them right, with food and rest, For as our pals, they are the best. Not just a pet, but family too, Dogs bring joy in all they do!

You will notice that the vocabulary is no longer as simple as in the original text. 

Finally I asked for a rewrite using short vowels:

Pups at Play

Dogs are pals, big and small, With fur that's short, or long and tall.

They jump and run, and play all day, With a ball to toss and fetch their way.

Some dogs swim, some sit and nap, With a wagging tail and a little yap.

With ears to hear and a nose to sniff, They find a scent, then off they zip!

Feed them well and give them a bed, Pet their head, and they're so well-fed.

Dogs are fun, they make us glad, Best little pals we've ever had!

In carefully examining the results, it is clear that ChatGPT can do some of the work for us, BUT we need to make sure that the text is consistently doing what we actually need. Right now, even with prompting, the results need "massaging" depending on what we are teaching. S teachers use but beware!

AI and Academic Publishing

 Like many others, I have been playing with generative AI for the past few months. I am an author of scientific papers and, even more so a frequent reviewer. I have been elated by the potential of generative AI to bridge the gap between English knowledge and conducting high-quality research, especially for international scholars whose first language is not English. This is an opportunity to level the playing field and allow equal access to academic publishing, which is predominantly conducted in English. Many times I have reviewed articles with good ideas but really hard-to-understand language that required many rounds of review and editing before it was publishable.

On the other hand, generative AI is quite as capable of generating data that isn't there (often referred to as hallucinating). For example, after being asked about my publications Chat GPT 3.5 spit out this list: None of the publications are real! This will require our publication engines to allow us to track every in-text reference with quick access so reviewers can check the veracity of such claims that may be "halucinated".

Even more challenging is generative AI's ability to "hallucinate" research studies. In a manner of a few minutes, I was able to have Chat GPT generate two potential studies about reading instruction (synthetic phonics and reading recovery) with ANOVA designs, including result tables. I even got Chat GPT to design and execute a study about the impact of a Wind Surfing intervention on Math achievement of second graders. For example, examine this paragraph generated after I requested a qualitative study instead:

"As this study focuses on qualitative exploration, the quantitative results will not be the primary focus. However, to provide a broader context for the qualitative findings, basic descriptive statistics of math achievement scores may be reported for both the windsurfing instruction group and the control group. These scores will be collected through pre- and post-intervention math assessments administered to all participants. The quantitative results will be used to complement and contextualize the qualitative findings, providing a broader perspective on students' math achievement in relation to their windsurfing experiences."

I am sure that generative AI will create an increase of papers submitted for publication. To prevent science from being overwhelmed and suspicious we may need to write new rules and accelerate existing trends.

1. Demand researchers pre-register their research.

2. Ask that each paper submitted will include a statement about the use of generative AI and will include the transcripts of their use.

3. Create ethical standards for AI use in scientific publishing AND teach about it in graduate schools.

4. Create reviewing mechanisms that allow easy tracking of citations to the source.

Tech EDGE Art TEAMS - Exhibitions of Learning

Here is our latest [podcats from Art TEAMS