Respuesta a las preocupaciones sobre Scribe

Recently on the Mosen At Large podcast, a listener wrote in with a thoughtful question. He wrote:

“I’d like to share some thoughts / questions about Pneuma Solutions Scribe and its promise to make documents accessible, hoping to get a broader understanding and perhaps develop a fair opinion. Here is another company that promises fully automated accessibility for an affordable price, but I wonder what they really mean by accessibility, how far their AI can go and how much of the remediation process they have been able to automate. I cannot imagine it going beyond OCR, automatic image descriptions and the addition of some structural tags for headings, lists and possibly simple tables. Can they fix color contrast issues? Can they check that the font is appropriate? How do they deal with data presented in diagrams and charts? I don’t know of any AI that can make data visualization accessible. How do they deal with situation where the document contains explanations based on color or graphical elements such as your typical product manual? In summary, is what they offer true accessibility or just a handful of automatic improvements for screen reader users? Are the files they generate accessible to everyone or just to screen reader users through a visually hidden link? During the interview, Mike pointed to the fact that users can request a human remediated file when there are problems. I would say that, as a blind screen reader user, I don’t know what I don’t know. In other words, in a document or a website, there are often pieces of information which are inaccessible up to the point where we don’t even know we are missing something. How are they protecting us from that? Even if their AI is so great that it can truly make documents accessible, they promote a model in which Pneuma Solutions establishes itself as a middleman between the content provider and us. Do we really need / want something like that at this point? They oppose AccessiBe and even developed an extension to block it, but I’m finding many similarities between their solution and that of AccessiBe. I know I’m somewhat negatively predisposed here, and that’s why I wanted to run it by you and the awesome audience of Mosen at Large.”

Thank you for asking us about the current limitations of our document remediation technology. We understand and appreciate your concerns.

You’re right about the current scope of what our technology can do automatically, specifically OCR, image descriptions, and adding structural tags.

In that sense, the current results that we can deliver automatically are equivalent and sometimes better than other processes on the market that require people and time, so they cost more to implement. This makes it a financial decision for an organization as to what documents they can afford to make available in an accessible format. In many cases, that means that because of the restrictions of budgets, information is not released to anyone.

This is not so true for new information, but legacy or archival information is often overlooked in these discussions. Many large libraries of archival material can take millions of dollars and thousands of hours to make accessible. With our technology we bring that cost way down to pennies per page and weeks instead of years for full library conversions.

It doesn’t make sense for us as a community to expect that companies spend hundreds of thousands of dollars to make documents accessible that may or may not be consumed. I would rather we have access to everything with a reasonable level of accessibility now, and if we need to have a better version, then the owner of the content should be required to make it available within a reasonable time period.

Now, back to the technology and why ours is better than current solutions. We deliver results across platforms, in multiple accessible formats including Braille, large print, and multiple languages to anyone in the World with a browser, with our usual focus on simplicity and ease of use.

And we want blind people to be able to benefit from this technology for any digital content they find on the web, without requiring them to invest hours to learn about document accessibility first. That’s what makes us different. I see no difference in what we are doing than with any other assistive technology. We have never promised a silver bullet.

We are developing an emerging technology that is only going to get better with use. But we’re not stopping there. As we discussed in the interview, we’re building and training a machine-learning model that will let us go beyond the current state of the art. That’s why it’s important that people start using our solution. We need to collect as wide a variety of real documents as we can in order to feed our machine-learning model.

Some of these documents will need to be remediated by a human, and as we said during the interview, the results of that remediation will help us and everyone who uses our service for years to come.

We’re working on automated remediation of diagrams and charts, as well as mathematical notation and other specialized notations. We understand how important these things are both in education and on the job. We’re also not the only ones working on these things. Other companies and organizations have been working on access to these things for quite a while, and we’ll incorporate the best available solutions into our automated remediation process. When we do, you’ll get an intelligent and easy-to-use product that gives you the best available automated access to a variety of document formats, all in one place.

As for color contrast and fonts, by default, Scribe preserves the original visual appearance of the document, within the limitations of your chosen output format. But you can choose to override this. For example, by checking a single checkbox, you can request a large print version of the document. Our goal is to serve the needs of people with all types of print impairments, not just blind people using screen readers.

That’s also why we don’t use hidden links. When Scribe is embedded on a website, the link to get an accessible document with Scribe is clearly visible.

You asked if our product can protect you from situations where you don’t even know what information you might be missing. It’s true that we can’t yet do anything about that. For now, the best we can offer is that if you feel that you are missing information you need out of a document, you should probably request human remediation. But, most of the time what we provide within seconds is enough. I assure you that every .PDF a sighted person looks at isn’t perfectly formatted every time. Formatting is important but, so is equal access to the information that is being published right now instead of in some indeterminate future.

About us being a middleman between the content providers and blind users, it’s important to be realistic about doing what we need to do to make information accessible. It would be nice if all content providers implemented accessibility standards, but we all need access to these documents now, not when they get around to it, or are forced by court order to do so. We would never discourage content providers from directly making every one of their documents accessible when created. But we know it’s not always feasible for many of them to do so. That’s why we’re offering our solution.

In the end, it’s more important to have timely access to information we need than it is to go through an intermediary.

Finally, you compared us to AccessiBe. Remember that we are blind people first; we need the technology we’re developing as much as you do. And we value the input of the blind community because we’re a part of it, unlike AccessiBe, which seems to view our community as a problem that customers can pay AccessiBe to sweep away. So, while we appreciate your concerns about the limits of our technology and the risk of over-hype, we hope you’ll understand why we’re different from AccessiBe and other companies like it.

Thanks again for voicing your concerns, and we hope you’ll be willing to give us a chance.