Imagining a world without artificial intelligence (AI) seems next to impossible today. From influencing businesses to making our daily lives seamless, AI has done all that and much more, especially over the past decade. It is important to understand the workings of AI, its advantages and challenges, and how it affects humanity to truly leverage its power. In an insightful session as part of Xperience AI, Former Head of Tensorflow - Rajat Monga - touched upon these issues in much detail. 

Before we move on how AI has evolved over the past few years, let’s first delve deep into how this technology works. 

Workings of AI

During his session, Monga broke down the various components of AI. This technology takes into account deep neural networks. For instance, if there is an input in the form of an image, then each layer goes through certain functions that are applied to it. Furthermore, they are stacked on top of each other, which is how it gets the name ‘deep’. 

“The outer layer makes a prediction on this. If it’s correct, that’s great. But if not, it learns from that incorrect behaviour, and with time, updates its parameters to really learn what’s going on,” he explains. 

Further deconstructing it, Monga speaks about how algorithms and data come together through compute. “Over the last few years, we have seen custom hardware being built for this. All these three aspects have come together over the years, where data, compute and algorithms are enabled by softwares such as Tensorflow that allow us to leverage their power together,” he adds.

Evolution over the years

That being said, there has been steady progress being made in the world of AI. Citing an example, Monga says that sometime back in 2011-12, there was research at Google that training networks across thousands of machines were leveraging YouTube videos. 

“It turns out that the algorithm was not really told what’s in there, but just from seeing so many videos of cats, there were neurons in there that would understand that. Around the same time, there was other research going on too that was leveraging the power of what was called convolutional networks in data sets; there was a huge jump in accuracy. Over the next few years, there was more improvement. As this moved forward, what really changed was how software was built,” adds Monga.

Another common example is that of Google Photos, which is utilised by people all over the world to automatically label pictures. But this wasn’t the case earlier. There were several softwares that had their day, including Picasa, which Google bought at some time.

“From manually labeling earlier to how it automatically labels today, this automatically changes how you build software. But what if you could say out loud that you want to see pictures of cats? That’s exactly what deep learning has allowed us to do. The same kind of algorithms are being applied to speech datasets,” explains Monga.

This progress has also led to a surge in virtual assistants like Alexa and Google Assistant, where instead of using typing or some other way to enter text, speech can be leveraged to carry on tasks. 

A blend of good and bad

Just like everything else, AI, too, has its cons. One such example is public cameras, which are being used all over the world today. This enables the automatic identification of where the places are, and there’s also a face recognition data set. Although this can help to track criminals, it could also be misused in other ways. 

There are several other challenges to humanity that AI poses. “More than a million terajoules of energy every single day is being used to boil one pan of water. Imagine every single minute, every person is boiling that water. Most of that energy is coming through the use of fossil fuels. There is an increase in climate change, drought and floods. There’s a lot being done, and AI is helping. But the important point to remember is to use it effectively,” says Monga. 

As they say, it’s always better to be safe than sorry!