Credit: Information Age
AI requires data: lots of data. The way we store that data in future may change, as cloud and edge computing develops. What does the future for AI data storage?
It’s perhaps fitting that a nebulous, hazy term like “the cloud” doesn’t have a definitive date of birth. There are milestones though, of course, in its development.
Many claim the mid-1990s is when the term was first coined. “You can think of our electronic meeting place as the cloud,” said David Hoffman of AT&T, in 1994. Few thought too much of it at the time. In the bustle of Britpop and the launch of Netscape, the cloud seemed like nothing more than a metaphor, rather than a benchmark for how we think of data and online computing. Since those early days though, storage has become an ever-developing topic in the tech world.
1994 was also the year that Jeff Bezos launched Amazon. Back then, it was unthinkable that 12 years later, Amazon would create Amazon Web Services and become a major player in cloud computing. Stranger still, the Seattle-based giant celebrated 25 years in business last week, looking forward to a future where Alexa will potentially lead the world in consumer AI. Surely, not even Bezos could have guessed where his ambition would take Amazon.
However, Amazon’s rise from humble media distributor to the tech goliath of today is not unique. The rise of technology has aligned with an expansion of data and the exploration into AI for so many businesses. More and more companies are requiring more and more data storage: especially if they want to explore machine learning.
Will AI affect cloud architectures?
Big businesses have big data storage requirements. Many rely on a hybrid cloud structure that allows for combining public and private clouds. Services can be easily distributed across data centres and hybrid clouds are particularly cost-effective.
Companies are evolving though. 61% of businesses have already implemented AI into their set-up and it’s hard not to imagine that this number will rise steeply year-on-year. With AI coming to the fore in so many businesses, there comes the need for more data: terabytes upon terabytes are needed just to power a single autonomous car, so imagine if our roads were jammed full of them, for example.
Artificial intelligence will become more human-like: it’s a worrier’s paradise when you put it like that, but it could be a good thing for technology.
Storage has to adapt to this new phenomenon. Surely it’s impossible to fill the cloud with the masses of data that machine learning requires? It’s important too to lead with a data-first approach when it comes to machine learning: Amazon, for example, collected a lot of data before implementing AI in their business.
The economic model of the cloud could well be seriously challenged further in the coming years. Big data environments require lots of servers to support a plethora of devices all processing large volumes of data. Clouds offer economical means to support big data, and using cloud computing for big data implementation lowers the internal processing power commitment.
Moving AI from the Cloud to the Edge
Deep learning applications have been deployed on the cloud successfully already.
There are however issues with data privacy, bandwidth and latency that come from the cloud. These are not new issues but in many ways, the cloud is still a better option for ambitious AI than edge computing. Training complex deep neural networks (DNN) is a difficult process and more suited to the cloud than the edge; edge AI devices tend to operate with smaller memory and power capabilities.
There are a host of reasons as to why AI is better suited to the cloud but equally, there are positives for edge AI too. The edge is far more secure than cloud computing, something that hugely appeals to artificial intelligence. The edge is where a lot of companies are investing and there’s a case that it’s actually easier to develop edge-based apps than cloud based apps.
In the imminent future, however, artificial intelligence will become more human-like: it’s a worrier’s paradise when you put it like that, but it could be a good thing for technology. Think of AI performing in a pyramid: at the peak comes the intelligence that it comes up with from the masses of data that it processes at the bottom.
In the future, AI will rely less on bottom-up big data to and more on top-down reasoning. AI will learn to generalise from fewer and fewer examples. Perhaps the need for more and more cloud space will start to reduce as we develop artificial intelligence further. Rather like when Amazon started, it’s difficult to predict the future.