Automating Accessibility

AI-Powered Audits for Frontend Development Workflows

If you are a Frontend Developer, a UX/UI Designer, or perhaps a DevOps Engineer, most likely you have faced the task of ensuring accessibility of your interfaces. As a society, we reached the point where making your web apps meet diverse user needs is a must. However, developing frontends accessible to everyone has always been and still is a challenging task. Moreover, it’s not always easy even to detect a problem with accessibility, let alone find a fitting solution.
Luckily, we live in the age of digital transformation, and new technologies – even ones that weren’t conceived with accessibility issues in mind – are there to help us. One universal technology that can be applied in accessibility testing is Artificial Intelligence. There are plenty of AI-powered tools that you can integrate into your development workflow and dramatically enhance your ability to conduct accessibility audits.

If you encounter accessibility barriers in front-end projects and are interested in implementing AI-driven accessibility audits, this article is for you. Let’s explore together these tools and methods and see how they can transform the accessibility of your projects.

Applying AI in Accessibility Audits

A Little About Traditional Audits

Seeing as you are still reading this article (after all, it is a niche subject), my bet is you belong to one of the categories I listed above. As such, I’m sure you’re accustomed to conducting accessibility audits to check if your web apps comply with standards like WCAG. This job usually means we have to manually review elements from text alternatives to keyboard navigation to semantic HTML – a monotonous, time-consuming, and painstaking work. Moreover, it’s not always easy even to detect a problem with accessibility, let alone find a suitable solution.

Automated tools (think WAVE or axe-core) offer a faster and significantly more efficient alternative to manual audits. However, although these tools do streamline the process and catch trivial errors, they lack the nuance of a live developer's insight and often fail to detect more subtle, context-specific accessibility challenges. Therefore, although automation accelerates the process of accessibility audits, it doesn't eliminate the need for manual review.

Perfecting the Process with AI Tools

AI-enhanced tools are in every sense transforming the way we deal with accessibility audits. Employ Machine Learning to analyze UIs, these tools detect accessibility barriers unnoticeable with standard checks. And they go beyond basic error detection: AI-driven tools learn from each interaction, providing over time more and more accurate assessments. This is particularly useful in dynamic content scenarios where user interactions vary significantly, and we need to ensure that accessibility adaptations are not just reactive but proactive.
Sophisticated AI technologies are indeed the next step in tool evolution. Microsoft and Google are developing some very promising open-source solutions in this area, watch them if you want to stay current. And if you want more versatility and robust community support – you can use Tensorflow.js to train and deploy custom ML models specific to your needs.
By the way, your tool of choice for accessibility audits may already be beginning to incorporate AI features. The already mentioned axe-core is a good example of such integration.

Now, let’s get physical and see how to integrate such tools into our workflows!

Integrating AI Tools into Development Workflows

Let’s review how you can set AI-driven tools to automate accessibility audits within your CI/CD pipeline in five easy steps.

Step 1: The Choice

As I already said, there are a lot of AI-enhanced accessibility tools to choose from. Here, I will focus on the AI-enhanced axe-core as a familiar ready-made solution with a comprehensive rule set allowing a decent level of customization.

Step 2: Setting Up AI Tools in Your Development Environment

Let's assume that you’re already more than familiar with basic setup and move to more complex aspects. In the example below, I'm integrating axe-core into a React application tested with Jest (a setup I often use):

`import { axe, toHaveNoViolations } from 'jest-axe';
import { render } from '@testing-library/react';

expect.extend(toHaveNoViolations);

describe('Accessibility checks', () => {
test('Must have no violations', async () => {
const { container } = render();
const results = await axe(container);
expect(results).toHaveNoViolations();
});
});`

This test checks for accessibility violations directly within the component rendering, making it an integral part of the development and testing lifecycle.

Step 3: Automating Tests in CI/CD Pipelines

Let’s go further and integrate our tests into our CI/CD pipeline so that they run on every commit. For this, we will use Jenkins:

pipeline {
agent any
stages {
stage('Install Dependencies') {
steps {
sh 'npm install'
}
}
stage('Run Linter') {
steps {
sh 'npm run lint'
}
}
stage('Run Accessibility Tests') {
steps {
sh 'npm run test:accessibility'
}
post {
always {
junit 'results.xml'
}
failure {
mail to: '[e-mail]',
subject: "Accessibility test failure in ${env.JOB_NAME}",
body: "An accessibility test failed, review the test results."
}
}
}
}
}
This script includes stages for installing dependencies, linting, and running accessibility tests with detailed post-stage actions, including notifying the development team about failures. This approach ensures not only that accessibility is tested but also that you’re immediately aware of any issues.

Step 4: Version Control Hooks

Diving deeper, you can set up a simple pre-commit hook in your version control system. In the example below, the hook prevents commits if the tests fail, reinforcing adherence to accessibility standards before the code even reaches your rep:

echo "#!/bin/sh
npm run test:accessibility || exit 1" > .git/hooks/pre-commit
chmod +x .git/hooks/pre-commit

Step 5. Real-Time Feedback

Last but not least, you can use tools like TensorFlow.js to configure real-time feedback mechanisms in your IDE. For instance, let’s set up a custom model to predict accessibility issues as we write our code:

editor.on('codeChange', async (code) => {
const issues = await predictAccessibilityIssues(code);
if (issues.length > 0) {
displayIssues(issues);
}
});

Challenges and Solutions

Like with any other instrument or technique, deploying AI-driven accessibility tools is not all roses. Let’s take a look at some possible technical issues you may encounter in the process and review our options for solving them.

Challenge 1: Dynamic Content Handling

AI-driven tools often struggle with dynamic content that loads asynchronously or changes based on user interaction. Because of this you may miss issues on parts of your code that weren't fully loaded and/or interacted with during the testing.
To solve this, you can enhance AI tools with event simulation capabilities or integrate them with frameworks that support dynamic content management. There are browser automation tools that simulate user interactions and ensure all dynamic content is loaded before the accessibility tests are run. In the example below, I’m using the Puppeteer library:

`const puppeteer = require('puppeteer');

(async () => {
const browser = await puppeteer.launch();
const page = await browser.newPage();
await page.goto('http://yourapp.com');
await page.waitForSelector('#dynamic-content'); // ensure dynamic content loads
// Now run your accessibility tests
const results = await axe.run(page);
console.log(results);
await browser.close();
})();`

Challenge 2: Legacy Systems

Legacy apps and systems may lack support for modern AI-driven tools, which sadly can impede your attempts to move your accessibility testing to the next level. Bridging this gap typically requires implementing a middleware solution. In this regard, you have two options:

1. Implementing an adapter pattern to modify the interface of your legacy system to make it compatible with new AI tools. This pattern acts as a translator, allowing old and new systems to communicate without requiring changes in architecture.
2. Or you can write an API that would act as a middleware interface. Here’s how you can do this:

`// Pseudocode for a middleware API that adapts legacy system data for new AI tools
function legacyDataAdapter(legacyData) {
// Transform legacy data format to be compatible with new AI tools
return transformData(legacyData);
}

// Use the adapted data with your AI tool
const adaptedData = legacyDataAdapter(oldSystemData);
aiTool.process(adaptedData);`

Challenge 3: Ensuring AI Model Accuracy

Even AI models don’t always predict or identify accessibility issues accurately. This can lead to false positives or missed issues.

To overcome this issue, continuously train and fine-tune your models on a rich dataset covering various accessibility scenarios. Feedback loops for flagging inaccuracies can also help you improve model performance:

aiModel.predict(issue).then(prediction => {
if (prediction.isIncorrect()) {
aiModel.retrainWith(feedbackData);
}
});

Conclusion

Dear reader, I hope the insights I shared in this article will help you enhance the accessibility of your web projects. However, the strategies and explanations I’ve covered here are far from an all-encompassing study of this topic. To go beyond the basic solutions described above and to discover the full potential of AI in improving web accessibility, you will have to experiment with AI-powered tools and solutions hand-on. I wish good luck to you and your projects, whatever you do!

Comments

[Ivan Isaac]

This is really insightful! How do you see the future of accessibility tools evolving beyond AI integration? Would love to read more about potential advancements.