AI & Big Data I write about the broad intersection of data and society.
As society has increasingly awoken to the dangers of algorithmic bias in the machine learning and AI systems that underlie an ever-greater portion of our lives, it is notable that for all of the attention and funding being focused on AI bias, there has been in comparison a deafening silence on the topic of accessibility bias.
As the web rushes ever faster towards a multimedia-first existence, why is it that there is comparatively so little conversation about making this content accessible to those with differing physical abilities?
AI bias is critically important for society to address, but instead of emphasizing the latest buzzword, why don’t we focus on addressing technological bias in all its forms?
One of the most remarkable aspects of the conversation around technological bias today is how narrowly it has focused on the topic of algorithmic bias. Funding agencies have rushed out call after call to fund AI bias research, major journals have afforded entire issues to the topic, companies across the board are launching high-level initiatives around the topic and the CEOs of the web’s most powerful companies are issuing statements on it.
Where is the same focus on accessibility? Where is the hundred plus million dollars for accessibility that has been spent by just a few major funders alone on AI bias over the last few years? Where are all of the major journal special issues on making the web accessible? Where are all of the top-level accessibility initiatives and statement after statement from top CEOs about their grand visions for making the web more accessible? Mark Zuckerberg’s 2019 vision statement mentions AI repeatedly, but offers not even the most cursory mention of accessibility.
Most major companies today have accessibility groups, there are certainly funders and journal issues focusing on accessibility and there are indeed CEOs that have referenced accessibility. However, the combined level of attention to accessibility issues is merely a rounding error compared to the resources being poured into AI ethics and bias at the moment.
Part of that is simply the fact that AI is the buzzword of the moment.
Another reason is that AI ethics is something the general public has some sense might affect them and can be addressed by companies in a way that doesn’t inconvenience their user base.
The real reason is that making the web accessible within the limits of today’s technology is seen as being in direct conflict with the rich frictionless multimedia world the rest of the web wants.
Who wins in this battle for the future of the web?
The vast majority of the web that forms its most economically important demographic? Or those for whom supporting their accessibility needs within current technology might increase the friction and burden on those who pay the web’s bills? That is sadly the way web companies seem to see the web of today.
Take social media. We love the ability to post beautiful photographs directly from our phones with meme-friendly captions. Who wants to type out a few sentences of detailed prose carefully describing the contents of the image we spent the past hour crafting for Instagram? It’s so much more fun to just put a snarky hashtag so we can achieve viral fame. We adore being able to live stream ourselves at any moment without worrying about closed captioning, speaking clearly or facing towards the camera in good lighting with our lips visible.
Instagram’s meteoric growth and the web’s rapid transformation into a visual-first medium reminds us that the web of today is a multimedia-first world.
It was not until 2016 that Twitter added the ability to add ALT text to images. Twitter touted its new feature in breathless terms of how it would finally allow “everyone [to] be included in the conversation.” Yet, in reality it is truly remarkable that it took more than a decade for Twitter to finally add the most basic of accessibility tools to its platform.
Incredibly, Instagram didn’t add ALT text support until just a month and a half ago, similarly touting it as some form of revolutionary new way of making its platform inclusive.
Neither company responded to requests for comment on why it took them so long to add support for alt text.
Facebook has attempted to add automatically generated descriptive keywords to images posted to its platforms. However, the results to date are rudimentary at best and often sadly comical.
One major US Congressional Instagram star’s images are typically labeled by Facebook’s algorithm as “Image may contain: 6 people, people standing.” How precisely does “Image may contain: 6 people, people standing” help someone understand an iconic inspirational image that supposedly captures the entire spirit of that politician’s story and has been shared half a million times with a caption that offers not a clue as to what it actually depicts?
In an era in which OCR has become commonplace, it is remarkable that neither Facebook or Twitter appear to perform automatic OCR on uploaded textual images to create more relevant ALT text.
Stepping back for a moment, both Twitter and Instagram provide their users with the ability to provide rich alternative ALT text descriptions of the images they share. Yet, spend a few hours browsing through posts from academics, news organizations, thinktanks, major influencers and elected officials and you’ll find that it is the rare image indeed that has ALT text associated with it. It is even rarer to find a live broadcast video with a full transcript provided alongside or burned into the video.
Social media platforms have given us the ability to make our content accessible, but we don’t take advantage of those tools.
Some of it may come down to awareness. Both Twitter and Instagram require users to take additional steps to include the ALT text descriptions. However, in lieu of using their ALT features, users could simply trade their snarky meme-optimized viral captions for descriptive accessible captions, so technology alone does not explain our inaccessible web.
The real answer is because we can’t be bothered.
We don’t need dedicated ALT text options or specialized accessibility tools to make our multimedia content accessible. We could just take the time to type up descriptive captions for each of our images. We could record all of our videos and write up complete transcripts or burn captioning into them before posting them to social media sites. We could stop broadcasting live video and instead record it for post annotation before sharing.
Yet we don’t.
Technology alone doesn’t make the web inaccessible. It is how we choose to use it. Or not use it.
Facebook, Instagram and Twitter may encourage us to use images and video to convey our thoughts to the world. That alone doesn’t make the web inaccessible. If we added genuinely descriptive captions to our images instead of snarky hashtags and captioned our videos the web wouldn’t be quite as inaccessible.
Again, we don’t.
We can’t blame the social media platforms for making the web inaccessible, it is our own refusal to provide accessibility to our content that is the root problem.
This raises the question of why Facebook and Instagram and Twitter don’t force users to make their content accessible.
It would be straightforward for the companies to require users to provide ALT text for each image as a condition for sharing it. Any upload that does not contain a genuine descriptive caption would be rejected. Machine learning algorithms would examine the submitted caption to ensure it is legitimate.
Refusing to upload images that do not contain ALT text would be a simple and trivial way for social platforms to force their users to take accessibility seriously.
So why don’t they?
Neither company responded to a request for comment on why they don’t enforce a requirement for ALT text. However, the vastly increased user friction and cost from requiring users to stop and write up a careful description of each image they share would likely cause many users to share less content. Instead of defaulting to blindly sharing every thought that pops into our heads without context, we would actually have to consider whether we want to spend the few moments it takes to actually describe that image.
Of course, machine learning can help transparently make content more accessible. YouTube automatically generates closed captioning for uploaded videos and even for livestream videos. For many videos the results are excellent, raising the question of why Facebook and Twitter don’t follow suit. Indeed, automated captioning of clear English speech is not that far off from the professional captioning provided by American television stations. Though, just a few dozen of the world’s thousands of spoken languages are supported today by any automated captioning solution.
On the other hand, Facebook’s automatic image keyword generation reminds us of the incredible limitations of current machine vision algorithms. For all their incredible advances and thought provoking technical demonstrations, production AI systems capable of generating fluent visually and semantically descriptive text robustly for imagery from across the world is still quite a way off.
Perhaps someday AI will improve to the point where it can generate accessible forms of multimedia content with accuracy and rendition that exceeds humans. Until we reach that point, we must ask ourselves why we ourselves aren’t taking the most basic trivial steps to make our own content more accessible.
Putting this all together, today our attention is increasingly fixated on algorithmic and AI bias, rather than the broader issue of technological bias as a whole. We rush towards an ever-more inaccessible web and refuse to acknowledge that most of the responsibility for making the web inaccessible lies with us, not the social platforms themselves. Advances in automated image captioning and video transcription will help offset some of this, but even the most sophisticated algorithms can only do much in a world rushing to be as inaccessible as we can make it.
Fixing our ever-more biased web requires more than pointing fingers at technology companies. It requires stepping back and asking how we got here. Why do we have a biased technology landscape in the first place? After all, when it comes to the needs, desires and intermittent whims of the monied and monetizable, technology companies are able to craft ever-more-perfect Utopian worlds.
How do we get companies to see beyond money to view all of us as equally valuable? Is it simply a forgone conclusion that on the web our economic worth defines our “value” to the digital world?
In turn, we reinforce these biases. We refuse to step outside our cultural comfort zones to be a bit less snarky and a bit more inclusive, acting as adults rather than adolescents as we refuse to take the time to embrace diversity and inclusivity ourselves in all its forms.
In the end, bringing the world together requires more than a few lines of code. It requires us to look up from our screens to see the world around us, in all its beauty and diversity.