AI as a Service

How common is the problem you need to solve? When our client came to us asking for help filtering out inappropriate adult content from their site, we were pretty confident they weren’t the first to need this tech.

However respectable your business and the legitimate users of your website or service are, if it involves uploading content then this always introduces an element of risk. Let’s assume that 99% of people out there would use your functionality exactly as you intend. It’s just that 1% who try to upload pictures of naked ladies for lolz you’ve got to worry about. It’s been the same since the dawn of time, or at the very least, the dawn of the internet. Or social media.

Solving common problems is easy, when you know where to start. That’s where we come in. By using Microsoft Azure Cognitive Services, we were able to provide an 'AI as a service’ (or ‘AIaaS’) solution for our client in just three days.

The big-hitters in cloud services like Google, Amazon and Microsoft have already spent ages solving problems like these. Setting up a pre-trained AI integration meant our client got all the benefit of this tech, but without the years of development, or billion-dollar price tag. And no more nudes. Win!

The challenge

There’s loads of brilliant businesses around like the one we helped, which provide really clever services for users online. The problems arise when you consider that not all ‘users’ will ‘use’ a service or product in the way it is intended, especially where content and imagery are concerned – think of the last social media platform you visited, and how hard they must have to work (or not!) to police their user generated content. Tricky, right?

You might think on the surface that our client’s problem was ‘How can we stop people uploading adult content to our site?’. You’d be right. However, there’s other underlying problems too, which also need asking first. (And answering. Obviously). For example:

What actually constitutes ‘adult content’? Who decides, and how?
How to avoid any filter being too heavy handed, impeding legitimate site users?
Should there be a human in the loop? Is live moderation by an actual person required?
For content which IS graded unsuitable, should further action be taken, and if so, what, and how? i.e. Is it automated? Or by a real person?

Not all adult content is created equal, unfortunately. There’s ‘unsuitable’ and then there’s ‘actually illegal’, which also poses the moral question of whether sites should have a duty to report the illegal activity a content filter catches, although that’s perhaps a whole other blog for a different day…

The solution

Philosophical and ethical questions aside, because we quickly understood how our client’s existing systems operated, we were able to identify and integrate the right AIaaS tool to filter any explicit content, which in this case was Microsoft Azure Cognitive Services. It involved machine assisted moderation, as well as human-in-the-loop processes.

The system used two separate metrics to ‘grade’ images, with the combination of both yielding a score as a percentage. From this, thresholds could be set for automatic exclusion – (e.g. anything with a score above 90% would almost certainly be the bad stuff) or queued into a workflow for review by a moderator team.

What was important here was getting a solution in place quickly (for obvious reasons).

We’ve written before about how AI can be used at different levels, depending on quite how particular your problem is. In this case – and for other common problems – rather than spending hours trying to re-do on a smaller budget what the big guns like Microsoft have already achieved, an easy access, pre-trained AI solution was the best bet.

Of course, knowing this is half the battle. Because at Fuzzy Labs we also work on much more bespoke applications of AI - using Machine Learning as a service (MLaaS) for example, using datasets to train custom algorithms rather than off-the-shelf solutions - we knew AIaaS was the best way to get the problem solved quickly, and within the client’s budget. The explicit content filter did the job right, with a minimum of fuss, and without compromising the final site experience.

Our client got big AI functionality with a low barrier to entry, and frankly, that’s what we’re all about.

Our other case studies:

How to innovate

Using AI to improve wound care

Where do new ideas come from? Collaboration is a good place to start when it comes to innovation, as we discovered when we talked to our medical client about their wound care provision

Making better decisions

How AI improved personalised healthcare using genetic data

If you identify a gap in your services, could AI be used to speed up the decision making process on what a potential solution or product could look like? That’s exactly what we helped our healthcare company client to achieve.

Reaching big goals with simple automation

How OCR unlocks possibilities in medical tech

Sometimes the way to achieve the really clever, complicated stuff is to actually do one very simple thing first.

Do more with your data

Speed up reporting with data warehousing

How much more could you be doing with your data? This was the question we asked our client, Regital, a digital agency specialising in programmatic advertising, when they asked us to help speed up their reporting processes.

Automate to improve

How AI helped enable faster benefits and council tax claims

Lots of clients know that AI can probably help speed up their processes, but aren’t sure what is possible, or how to start. Demonstrating how automation can improve efficiency for local government.