Welcome back, fellow SLPs! In our last two posts, we tackled the crucial topic of AI and Clinical Data, as well as AI Truths & Myths. We learned that responsible AI use starts with safeguarding our clients’ privacy and understanding how AI works in practice.

We’ve debunked some common myths and explored the complex questions surrounding AI and data security. Now, let’s pull back the curtain and take a look at the mechanics room. In this post, we’ll demystify how AI tools actually work, providing a clear and simple explanation for SLPs.

Artificial intelligence can sometimes feel like magic, especially when it comes to generating coherent text for notes, therapy ideas, or reports. You type in a prompt, and seemingly out of nowhere, clinically relevant content appears. But behind this impressive capability lies a fascinating, albeit complex, system built on specific AI principles.

As SLPs, understanding the basics of how AI tools function is essential. It can not only alleviate some of the mystery, but also empower us to use these tools more effectively and with realistic expectations. So, let’s peek under the hood and demystify the mechanics of AI.

The Foundation: Large Language Models (LLMs)

To truly understand how AI works, you must first understand its base. At the heart of most AI tools you encounter that generate text are Large Language Models (LLMs). Think of these LLMs as incredibly advanced prediction machines for language.

They’re not thinking, feeling beings; instead, they are sophisticated algorithms. They are often built on a “transformer” architecture, that have learned to predict what word or sequence of words is most likely to come next. This ability to predict is the foundation of how they generate paragraphs, articles, notes, and even initial drafts of reports.

The Training Process: Learning from Text Patterns

The secret to this predictive power lies in the training process. LLMs are trained on truly massive amounts of text data – billions upon billions of words scraped from books, articles, websites, code, and just about any other digital text you can imagine. For models used in healthcare or clinical research, this could also include vast quantities of de-identified research papers, medical journals, and clinical guidelines.

During this training, the AI doesn’t process this raw text directly. Instead, it goes through a crucial step called tokenization.

What is Tokenization?

Tokenization is the process of breaking down the text into smaller, manageable units called tokens. These tokens can be whole words (“the,” “client”), parts of words (“-ing,” “re-“), or even individual characters or punctuation marks. The way text is tokenized can vary between different AI models, but the fundamental idea is to convert human language into a numerical format that the AI can process.

Think of tokenization like the raw data stream in the movie The Matrix. Instead of seeing words and sentences, the AI sees a constant flow of these individual numerical tokens. This numerical representation is the fundamental language that the AI understands and manipulates.

Each unique token in the AI’s vast vocabulary is then assigned a numerical ID. Therefore, AI operates at this granular level of tokens and learns statistical relationships between them. It doesn’t typically “hold” or store entire books, articles, or clinical records as copies. Instead, it retains the learned patterns and probabilities of token sequences.

How is Tokenization Used?

Once the text is tokenized and converted into these numerical representations, the AI analyzes this vast ocean of language, identifying patterns. It learns the statistical probability of certain tokens appearing together. This includes the common structures of sentences and paragraphs, and typical ways humans express different ideas in various contexts (e.g., academic writing, informal speech, or clinical documentation).

For example, the AI learns that after the tokens “Client” followed by “demonstrated,” the token “progress” has a very high probability of appearing next in a clinical note. It learns the statistical relationships between diagnostic terms and common intervention strategies.

It’s all about recognizing these statistical relationships between these numerical tokens. It’s crucial to remember that the AI doesn’t “understand” these relationships in the way a human clinician does. It doesn’t grasp the clinical meaning, the client’s emotional state, or the unique context of a therapy session. It simply recognizes that, based on the numerical sequences it has processed, certain token sequences are more probable than others.

Mechanics of AI Text Generation: Prediction in Action

So, how does this learned knowledge translate into generated text that can assist an SLP? Let’s discuss the AI principles at work. First, you give an AI a prompt – a request, a question, or even just a few keywords (e.g., “Draft a SOAP note for a 5-year-old with articulation goals for /r/”). Then, the AI first tokenizes your input.

After that, it processes its learned patterns and the numerical IDs of these tokens. It then predicts the most likely next token (represented by its numerical ID). This prediction is heavily influenced by the parameters of the LLM. Think of these parameters, often billions or even trillions, as the internal “knobs and dials”, like the Oz Wizard. The AI adjusts these during its training to learn the strength of the statistical relationships between tokens. The more parameters an AI has, generally the more nuanced and sophisticated its predictions can be.

The Output of the AI Machine

The AI takes that newly predicted token. It compares the original prompt and the token it just produced. Then it predicts the next most likely token, again relying on its vast network of parameters. This process continues, token by token, until the AI determines it has reached a stopping point. Finally, these numerical tokens are converted back into human-readable text.

Think of it like a super-powered version of the autocomplete feature you might use when typing. However, it operates at the level of fundamental text units, and is guided by an incredibly complex web of learned relationships stored within its parameters. The “creativity” or “clinical relevance” we sometimes see in AI-generated text isn’t true clinical reasoning in the human sense. Rather, it is the result of the sheer vastness of the data it has been trained on. It has learned the intricate statistical relationships between these tokens. It is all governed by its numerous parameters.

What AI Isn’t: Understanding the Clinical Limitations

Finally, it’s vital to reiterate what AI isn’t, especially from an SLP perspective. Even understanding the mechanics of AI, its human-like responses can often muddy the waters.

• It isn’t conscious. It doesn’t have emotions, personal experiences, or clinical judgment to draw upon.
• It cannot understand or provide the nuances of human communication. Sarcasm, irony, or the subtle emotional context of a client’s utterance are often difficult for AI to understand. They are even harder for AI to create (don’t ask it to generate jokes).
• It is limited in understanding highly nuanced linguistic elements like prosody, intonation, or specific phonetic realizations (e.g., truly hearing and analyzing a distorted /s/ or a lateral lisp). These areas are critical for SLP assessment and intervention, but challenging for current AI to grasp fully and accurately.
• It cannot build genuine therapeutic rapport or provide the empathetic connection essential for effective therapy.
• It cannot make a differential diagnosis or independently formulate a comprehensive, individualized treatment plan based on dynamic clinical interactions.
• AI can’t handle very large chunks of information. This is a significant limitation to remember, especially when working with long documents. More advanced and often paid AI versions, which are hopefully HIPAA-compliant, can process larger numbers of tokens.

Its output, however impressive, is ultimately based on the patterns it has learned from the text it has been trained on. You must remember that it is processed at the level of tokens and statistical probabilities, not genuine clinical understanding. You MUST always double-check output for accuracy, completeness, and bias!

Looking Ahead: Understanding AI as Your SLP’s Toolbox

Understanding the core mechanics of AI—that it’s a sophisticated text prediction machine built on statistical patterns—is the key to unlocking its potential without falling for the hype. This knowledge empowers us to see AI not as a magical entity, but as a powerful tool in our clinical process. It’s an assistant that can handle routine tasks, potentially allowing us to save precious time and redirect our energy to where it matters most: our clients.

Now that we’ve peeked under the hood, what are your initial reactions? Has demystifying AI’s inner workings changed your perception of its potential uses (or limitations) in SLP? Share your thoughts in the comments below, and then weigh in with your perspective in our quick poll! Results will be shared at the end of the series.

AI & SLPs Poll 3

I’d love to hear your thoughts after reading this post! Please answer these 4 questions; your valuable input will help shape my series’ final insights. To keep the poll fair and ensure unique responses, a Google account sign-in is required. Be assured, your email address is neither collected nor visible to me.

The AI & SLPs Series: Your Comprehensive Guide

Welcome to the AI & SLPs Series! Over the next eight weeks, we’ll delve deep into how Artificial Intelligence is shaping the world of speech-language pathology. Here’s what you can expect:

• Part 1: AI & Clinical Data Privacy
  • This foundational post explores AI training data, client privacy, and HIPAA compliance for SLPs, including the non-negotiable role of BAAs.
• Part 2: Separating Truth vs Myth 
  • We debunk common AI myths in SLP practice. Get a realistic understanding of AI’s true role and capabilities.
• Part 3: How AI Tools Work (current post)
  • Get a clear, jargon-free explanation of how large language models function. Understand their capabilities and limitations.
• Part 4: AI for Clinical Spark & Efficiency
  • Discover ethical ways to use AI. Brainstorm, overcome planning hurdles, and refine non-clinical communications.
• Part 5: Mastering Prompts 
  • Learn prompt engineering. Communicate effectively with AI models to get tailored, useful results for SLP needs.
• Part 6: Compliant Platforms & Tools
  • This post guides you through AI tools. Learn key factors for ethically and compliantly selecting platforms for your SLP practice.
• Part 7: Ethical & Responsible Use
  • This crucial post delves into broader ethical responsibilities for SLPs using AI. It covers principles beyond data privacy.
• Part 8: The Future of AI
  • This concluding post explores emerging AI trends and future possibilities in Speech-Language Pathology. Prepare to adapt, innovate, and lead responsible AI integration.

Stick around as we keep figuring out this whole thing together, giving SLPs the knowledge they need and helping us all find a balanced way to think about AI in the future of speech-language pathology. There is a lot of gray area and strong opinions, and I hope I can provide some facts to help you make informed choices that correspond with your own values.

Keep on clickin’!

Social Media Icons: designed by rawpixel.com – Freepik.com

The post AI & SLPs Series, Part 3: How These Tools Actually Work appeared first on Mrs. Speech Online.

Motivators

Just like our younger clients, older students love a good reward at the end of a challenging session. The beauty? They’re experts at navigating YouTube, so they can easily cue up their favorite vids. But wait, there’s more! I love introducing them to some lesser-known gems that pack a powerful punch:

Flexible thinking and how the mind learns new things:

Backwards bicycle: this was mind-blowing to watch!

How mistakes are okay and help you learn:

OK Go Rube Goldberg Machine (I talk about how you can see the evidence of how many previous tries – 85 to be exact – they had to get it right, but how awesome it was when they got it right.  BTW, the song itself has a great message of how bad things are only temporary.) It’s also useful to see the TEDTalk on how it was engineered. (Caution, the TEDTalk has one minor cuss word and the band is having a drink of champagne at the end.)  If you don’t know what a Rube Goldberg machine is, it is a YouTube rabbit hole.  I warned you.

The Power of a Stutter

A TedTalk on how accepting your insecurities and even embracing them allows us to become better communicators.  How it frees us from worrying about hiding them to think about how we can speak to and help others. (BTW, she has a very noticeable stutter so this is an excellent video for your fluency students.)

Teaching Support

How to make sounds

I’m probably not alone when I say that I turn to these when I have a kid I just can’t get to make the sound, especially now that I am wholly teletherapy.  My camera just isn’t as professional as what I can find on YouTube.  The two best sources for sound instruction I’ve found are Sharon Turner, who has a animated video model of the tongue inside the mouth for each sound, and Peachie Speechie, who has great teaching videos for each sound.  But there are many, many options to choose from.

Read-Alouds

I’m using short stories in most of my older groups.  When we first read a story, I want it to be a dramatic, cohesive reading, so I actually don’t want the students to read it.  I want them to listen and be drawn into the story.   But you know, a day of that and my voice gets tired.  Cue YouTube read-alouds!  You can find an audiobook of most popular short stories on YouTube pretty easily.  And bonus – most you can turn on Closed Captioning so students can read-along if you don’t have a text that matches well.  And the students actually love being read to.  If I have a student who needs to read for a goal, then I can stop and have them read a paragraph or so, then return to the read aloud. I have playlists of some of my favorites at http://www.youtube.com/@Mrs.Speech-wk4mr.

Background Knowledge

To go along with the read-alouds, a well-made video to give background knowledge is worth so much more than me droning on about what life in a lighthouse is like.  You know what they say, “A picture is worth a thousand words.”  Well, a video must be worth a million, right?!  And they are so much more willing to watch a video than listen to me (and let’s be honest, it gives my voice a break).

Movies

And if at all possible, I like to find at least a clip of a movie or animation to go with the story.  It provides so many opportunities for language-rich discussion, with opinions (Which did you like better, and why?) comparing and contrasting, and inferencing (Why do you think the director made this change?).

Social Skills

There are SOOOO many social skills videos out there.  Some of them are tailored toward our autistic population, some are more for introverted young adults, and some are for the ESL population.  Don’t just stick to the videos that tell how-to do something.  Sometimes the awkward videos offer great teaching moments too.  This one is one of my favorites.  While I do touch on what he does wrong, I mainly use it to talk about other people’s responses, and how they are generally kind about it and make an effort to not make him feel bad even though everyone feels awkward.  People are typically kind, so you might as well go out there and try.

Pro Tips 

Now, I mentioned in Read-Alouds to turn on the Closed Captioning if you don’t have a text that matches the reading.  It might be useful to do so with the other videos as well, to make the speech/text connection and fill in any missing gaps in the auditory information.  

I also suggest you turn the speed to 75%.  This is usually not noticeable to the kids, but it allows them more time to process what is being said so they get a better understanding of the information.  

I have more tips in my previous post: Unlocking Language Learning with YouTube: A Speech Therapist’s Guide.

Update 12/24

I recently found the free internet browser Brave and have been using it the past couple of weeks.  It was simple to move my Google bookmarks into Brave, and I am using it solely for use with the kids, so it streamlines my therapy sessions too.  I haven’t had a single youtube ad, and I haven’t noticed other ads popping up on websites.  I will say that, while it does block the ad on a website, there is often still the space where the ad would be.  Often, a reader extension like ReadBee will remove those spaces and focus on the text.

So, there you have it, folks! These are just a few ways to leverage the power of YouTube in therapy sessions with older students. But hey, the possibilities are endless! Got some killer ideas of your own? Drop ’em in the comments below!

Until next time, my e-speechies – keep on clickin’!

Social Media Icons: designed by rawpixel.com – Freepik.com

The post YouTube in Therapy: Engaging Older Students with Smart Strategies appeared first on Mrs. Speech Online.