Blurring faces in Fulcrum with AI

Q: What does this new feature imply about Fulcrum?

This new feature demonstrates one way Fulcrum is innovating by tapping the power of AI for automated image processing.

(If you’re the type of person who wants to jump straight to the doc, go for it.)

It’s a cool feature, for sure, and a great application of AI in particular. But why start with privacy? Why are photos a privacy issue? Why use AI instead of having human beings review photos? And what does all of this imply about Fulcrum?

The short answers: Privacy regulations such as GDPR and CCPA are developing teeth in the form of fines and significant penalties. Since faces can be used to identify people, these regulations require organizations to blur faces in photos. AI is actually better than humans at identifying faces at scale. And this new feature demonstrates one way Fulcrum is innovating by tapping the power of AI for automated image processing.

For the longer answers, let’s start with a broad perspective and drill in.

How most of us will use AI

The majority of us will never create new AI functionality. It still takes a lot of knowledge to do so. Not only do humans need training in AI algorithms and tooling, the AIs created need to be trained too. (“Training,” for an AI, is looking at lots and lots of examples of the thing you’re applying AI to.) While AI is often hyped as semi-miraculous, focusing on practical, well-understood use cases maximizes its value.

AI-based computer vision

Computer vision is one of those specific use cases that’s worth focusing on. It’s a form of AI-based image processing that enables a computer to tell us something about the contents of an image. We can use it pretty reliably to identify specific items in photos, such as soda cans, cranes, or faces. And because these are pre-built, pre-trained AI models, there’s no code necessary to use them.

For example, this is a photo taken by Fulcrum, in the Fulcrum labs, where we use artificial intelligence to identify a beer can and the text that’s written on it.

(This is something that we’re shaking out. If you’re interested in working with us on it, contact the Product Manager, Dale Wilson, to discuss your potential use cases.)

Where AI does better than people

Humans still tend to be far better at general-purpose applications than AIs are. That’s why visual CAPTCHAs are so common, and why we do better than AIs at games like “Muffin or Chihuahua?”

But a utility plant or office looks nothing like the carefully laid-out set of muffins and dogs we see above. Instead, it contains things like cranes, transformers, and desks. If there’s something that looks like a face to an AI, it’s almost certainly a face. Leveraging AI to scan photos for faces eliminates reliance on humans. Software can then blur out the faces detected, ensuring compliance with privacy regulations.

In fact, humans are arguably much worse at this kind of task than an AI. We may be better at general-purpose image identification, but we’re lousy at repetitive tasks. People get bored and hungry. They want to go home on a Friday afternoon. They daydream and start flipping pictures without really looking at them. Stuff gets missed.

We’re only human, after all.

But once you put an AI to work, it works tirelessly. Here’s a photo taken in Fulcrum that has been de-identified with no human intervention whatsoever.

Targeting privacy

To get some perspective on why we should place such a high priority on using AI for facial blurring specifically, I spent some time with Kristin Carrington, the founder and CEO of Carrington Risk. Her company provides risk management for commercial properties, which includes using Fulcrum to perform property inspections.

Kristin told me that it has become absolutely essential for her business.

Privacy laws, like GDPR in the EU and CCPA in California, tighten regulations on personally identifiable information (PII). A face in a picture qualifies as PII as per the EU definition. Soon, Kristin predicts it will be challenging to conduct global business without de-identifying faces in photos. Manual de-identification processes are cumbersome and may not be permitted by clients.

Moreover, manual review is an additional, unnecessary risk. It’s better to allow the AI to take over, thoroughly reviewing every photo equally. This mitigates potential human oversight after a long day.

Fulcrum’s AI-based feature is the obvious answer due to the high risk of human error and poor productivity. It opens more business opportunities for her company. This ensures highly productive engagement with those opportunities.