Appu Shaji, CEO of Mobius Labs: It’s Time We Democratised ML Algorithms

How Computer Vision Has Evolved

Although AI has been around in some form for decades, we’re only now starting to see it kick into high gear, argues Appu. It’s an interesting time to be working within AI, with a lot of the technologies starting to mature as more data to train machine learning (ML) algorithms becomes available. But the reason that so much can be done with AI now is because of the hard work people like Appu have been putting in for the past 20 years.

When Appu started out in computer science, he was drawn particularly toward computer vision. Possibly influenced by his family’s creativity – his father was a filmmaker – Appu found a way to connect his passion for science with a visual element. However, as he soon discovered, simply because you have a camera and you have a computer, making a computer understand what the camera sees is easier said than done. Appu likened finding a solution and ensuring a computer can understand what it sees to Alice going down the rabbit hole – it’s impossible to tell what you’ll find or how deep the rabbit hole goes.

But what was impossible then has become possible in today’s world. Image recognition software is now capable of recognising and analysing faces, animals, plants, objects, weather, geography and much more. It’s able to sift through millions of images and categorise them in seconds, far beyond the capabilities of humans.

And while in the past such powerful technology was only available to specialists and large enterprises, Appu points out that this is no longer the case…

Democratisation of Machine Learning Algorithms

A big driver behind Mobius Labs is the idea that computer vision software should be more freely available to non-specialist users. In other words, Appu is a champion of the democratisation of machine learning algorithms.

“You shouldn’t need to have PhD knowledge to use this technology anymore,” Appu said. “We want to empower people to build those applications” that create new business value, and it’s important that complex technology doesn’t stand in the way of that aim.

For Appu, the first step for any enterprise hoping to adopt computer vision tech is to very clearly identify the problems they want to solve and then move from there. To create real business value, it has to start with addressing a problem, and not purely because of innovation for innovation’s sake. In this way, you can truly build something that people will use and succeed with.

Where computer vision can help most is in bringing “superhuman” vision capabilities to the enterprise. Take the example of a news organisation, something Mobius Labs has a lot of experience helping. A news company has a mountain of content in its database, including both photo and video content. Too much content, in fact, making it impossible for any single human or team of humans to go through and categorise everything.

This is why, for Appu, it’s vital that we make AI solutions more open to users. “We want to work with non-technical users,” Appu told us. In that way, Mobius Labs can help businesses of all sizes operate in a content-rich world.

What About the Ethics Behind Computer Vision?

There have been countless worries about the ethical implications of AI, especially when it comes to computer vision software. On the one hand, you have the potential for unconscious biases in the development team to rub off on the machine learning algorithm, resulting in discriminatory behaviour from the AI. On the other, you have worries about the exploitation of AI in a surveillance state.

At all times, those who develop and use AI solutions should bear these ethical implications in mind, Appu argues. One way to do that is by bringing people together from a range of perspectives and backgrounds to work on the development and training of ML algorithms. Issues that may escape one individual or group are therefore more likely to be raised by another.

Where it comes to state exploitation, or even corporate abuse, Appu suggests that accountability must be woven into the processes around AI usage to uphold an ethical stance. “The algorithm doesn’t have a sense of accountability built in,” Appu pointed out to us. Instead, it’s up to data leaders to make up for this shortfall.

For instance, AI has the potential to help in policing. But instances of bad publicity, including facial recognition misidentifying or discriminating against ethnic minorities, can and should be offset before a solution is even rolled out. Appu demonstrated that in a sensitive area like policing, automating decision-making should be extremely caveated – or removed completely from the process. In this way, you can re-introduce accountability into the process and offset some of the ethical problems that could arise from biased software.

But one area that computer vision solutions can help rebalance the ethical battle over AI is in combatting deepfakes. As Mobius Labs works heavily with journalists, they understand just how important it is to sort through content and verify its authenticity. “We live in a very pixel-polluted world, where it’s difficult to understand what is actually in the pixel itself,” Appu said. “Our AI can really bring people value in clearing that up.”

Appu highlighted that current computer vision solutions are able to detect highly unethical deepfakes, but that this is a game of cat and mouse. This current generation of algorithms isn’t smart enough to mimic humans 100%, but behaviourally they may soon be. This is an AI arms race, and to root out deepfakes computer vision must keep evolving. Appu is confident that deepfakes will remain detectable for the foreseeable future – but that isn’t cause to rest!

The Potential for Geospatial Monitoring

Perhaps most excitingly for our editor, Appu also shared details about the potential for computer vision software to help with global issues, like climate change.

It doesn’t take long to find worrying news about extreme weather events in today’s world. But as these events become more frequent, people will need all the help they can get to prepare for and deal with the fallout. Adding value to society is very important to Appu and Mobius Labs – and this is clear from the geospatial work the company has done so far.

In 2020, there were 5,774 satellites orbiting the Earth, many of them fitted with cameras pointing back to us. Combine that amount of coverage with that provided by drone technology, and you have an unparalleled visual database from which to monitor the environment and the weather.

Appu told us about the potential for computer vision technology to detect wildfires in remote locations, which the authorities can then flag. Constant monitoring of the situation can then help track the wildfire. Activities like this could help to save lives and whole ecosystems from devastation. Likewise, the technology could be used to track deforestation, which contributes to 10% of global greenhouse gas emissions. Tracking illegal deforestation could help to limit these emissions by empowering authorities.

Overall, it’s clear that computer vision is a complex but evolving area of technological innovation. But, when handled correctly and the ethical implications of its misuse are properly considered, there’s great potential for computer vision solutions to bring about societal good. Whether that’s by improving journalism, monitoring extreme weather, or by rooting out harmful deepfakes. And for the CIOs of growing, multi-national enterprises, these kinds of technologies will be all the more necessary in the years to come.