by Daniel Kahu
reading time: 4 minutes
The views expressed in this post are those of the author and do not necessarily reflect the views and policies of NWACS. No endorsement by NWACS is implied regarding any device, manufacturer, resource, or strategy mentioned.
AI and Word Prediction
When my mum started to lose her voice due to Amyotrophic Lateral Sclerosis (ALS) - also known as Motor Neurone Disease (MND), I spent a lot of time looking into what options were available.
I knew she’d need a text-based app - but I was surprised to find that most of them relied on very simple word prediction.
While word predictions have their uses, it can still take a long time to write out what a user wants to say.
That’s why I started building Alek. Alek is a new kind of AAC tool that listens to the conversation and uses artificial intelligence to help suggest what to say next. Here's what I’ve learned along the way.
1. The Problem: Predictions That Don’t Predict
If you’ve used a text-based AAC app, you’ve seen it happen. You type a few letters, and the app suggests words or phrases. But what shows up often feels off - like it’s guessing wildly.
That doesn’t mean these tools are useless. They can still help a lot:
Completing a word to save time
Reducing the number of taps
Helping users get their message out faster
But when it comes to suggesting full responses, especially in a real conversation, most systems miss the mark.
2. Why It Happens: Static Tools in Dynamic Conversations
Most prediction tools work the same way. They look at the last couple of words you’ve typed and guess what usually comes next.
For example, if I type “thank”, my phone might suggest:
screenshot of phone showing word prediction options to fill in after ‘Thank’ (you, God, all)
This works fine for simple messages. These are common phrases - but conversation is more than just frequently used phrases.
3. The Real Need: Smart Suggestions That Understand Context
Think about the last time you had a conversation. How did you decide what to say? You probably used more than just the last word that you said.
We all use context. We think about:
What’s already been said in the conversation
Who we’re talking to
Ourselves and what’s going on in our life
two people, one looks like they are asking a question, the other looks like they are thinking (What was the question? Who am I? What am I doing?)
These things shape what we say. Think about it, how would you answer if:
Your doctor asked “How are you doing?”
Your best friend asked “What are you doing this weekend?”
Someone you just met asked you “What do you do for work?”
Same types of questions - but your answers would change depending on the person and the setting.
For AAC predictions to truly help, they need to understand that kind of context.
4. What We’ve Tried with Alek
That’s why we built Alek - a new AAC tool that takes prediction further.
Here’s how Alek works:
Listens to the conversation: Alek keeps track of what has been said so far, so it can suggest responses that follow naturally.
Knows about you: You can add details about your personality, common phrases, or topics you care about - and it takes those into account.
It responds to what you’ve typed: Alek will take your lead - you type a few words so it knows more about how you want to respond.
And then Alek uses artificial intelligence to suggest how you can respond.
visual of Alek gathering information and then providing three options
We’re not claiming it’s perfect. Sometimes the suggestions are still awkward or miss the point. But we’re learning fast - and we’re working closely with AAC users, families, and therapists to improve it every week.
drawing of a person in a wheelchair holding a tablet next to a screenshot of the tablet screen of the Alek app in action; The user was asked: “What do you want to do this afternoon?”
5. The Challenges with AI in AAC
Using AI in AAC opens new possibilities - but also brings important challenges.
Privacy is a key concern.
When Alek listens to a conversation, we need to protect everyone involved. Our approach is to keep everything on the user’s device wherever possible, and only send the minimum data required for suggestions. Nothing is used for advertising or stored long-term.
Still, privacy isn’t just technical - it’s social too. We encourage users to let their communication partners know when Alek is in listening mode, and we’re exploring better ways to make that transparent (like on-screen indicators and clearer settings).
If you want more of the nitty gritty details, you can find our privacy policy here: https://alekassist.com/privacy
Other challenges we’re working on include:
Ensuring users always stay in control of what gets said
Making suggestions feel personal — like you would say it
Designing an interface that fits this new kind of predictive help
These aren’t easy problems, but they’re the ones worth solving.
6. The Future of Prediction in AAC
We believe that prediction in AAC should feel smart, personal, and natural. That means moving past generic suggestions and toward systems that:
Understand the conversation as it’s happening
Learn and adapt to each user
Fit the setting, the relationship, and the moment
That’s the future we’re building towards - and we’d love your help shaping it.
Alek is currently completely free and available right now (although there will be a cost associated with it in the future):
In your web browser at alekassist.com
Or download it on the Apple iOS store
If you’re an AAC user, therapist, or carer, we’d love to hear from you. Your feedback is helping Alek get better every week.
Follow our journey:
