In this piece, we speak to ITProPortal about the overarching fear that AI and Machine Learning are going to take over people’s jobs and the counter argument.
July 28, 2017
Supporting humans and networks: AI and machine learning
There is an overarching fear that Artificial Intelligence (AI) and Machine Learning are going to take over people’s jobs, but there is a counter argument that their main purpose is to support humans as enabling technologies. In their proponents’ viewpoint, they aren’t disabling anyone. However, organisations that don’t train up their staff now to learn new skills may find themselves left behind. This includes IT, which is of increasingly strategic importance to most organisations today. Both technologies are becoming a fundamental part of our lives, and with the advent of semi-autonomous and autonomous vehicles they will become more so – both in consumer and enterprise applications.
SD-WANs are very good at the branch office level, but as technology moves forward data volumes are going increase and the time to intelligence will need to shrink. Whilst SD-WANs are great for low bandwidth applications, with high bandwidth applications a different approach is needed to move ever larger amounts of data.
Human error
Humans make mistakes – that’s part of our nature, and by using AI and machine learning the risks associated with human intervention can be removed, which could include unexpected network downtime due to the poor manual configuring of a wide-area network (WAN). Thankfully, the concepts of AI and machine learning in IT networking are not science fiction. Rather than making us weaker, they can make us stronger and enable us to increase our performance. They are no Armageddon; they are an enabler that can permit organisations to do more with fewer resources.
The science fiction of autonomous networking, which is spoken about by David Hughes, Founder and CEO of Silver Peak Systems, in his sponsored article for Network World, is already here today in solutions such as PORTrockIT and WANrockIT. They can correctly mitigate the effects of latency without your organisation having to unnecessarily spend money on ever increasingly large bandwidths, WAN Optimisation, SD-WAN and WAN optimisation solutions. With AI and machine learning much can be achieved with what you’ve already got, and an ever larger pipe won’t defeat the laws of physics no matter how much you spend. The problems created by latency will still remain.
Hughes says that many enterprises are using SD-WAN solutions to connect employees consistently and securely to cloud and datacentre applications, but by themselves they do not provide any form of optimisation to enhance the flow of data. You have to add WAN optimisation, which many of the SD-WAN providers do. However, with security concerns requiring encrypted data and rich media being an increasing part of the data mix, they provide little or no performance improvement. He’s nevertheless right to explain that automation is playing a role in SD-WANs to eliminate many of the repetitive and mundane manual steps, which are required to configure and connect remote offices.
He believes it has limitations though: “Automation has its limitations…[it] is not sufficient to translate high-level business goals or intent into specific actions across the network, and automation is not good at dealing with the many unanticipated situations across production WAN deployments.” In his view these are areas where machine learning and artificial intelligence can play a role. With machine learning, WANs can be directed to adapt to changing environments without human intervention.
Data volumes
AI and machine learning techniques permit us to better manage and to cope with the ever-growing data volumes too. Clint Boulton, Senior Writer at CIO magazine, talks about freight forwarding company JAS Global in his 12th May 2017 article, ‘How logistics firm leverages SD-WAN for competitive advantage’, and refers to it taking a gamble on an unknown technology.
The firm is using an SD-WAN to run cloud applications, but hopes to use it as the backbone of a predictive analytics strategy to grow its business. The claim is that JAS Global managed to cut millions of dollars from its bandwidth costs. That’s good.
Boulton also explains: “SD-WANs allow companies to set up and manage networking functionality, including VPNs, WAN optimisation, VoIP and firewalls, using software to program traffic routing typically conducted by routers and switches. Just as virtualisation software disrupted the server market, SD-WANs are shaking up the networking equipment market.”
He will, as many before him have found out once you start down the big data path, find that the volumes of data start to increase exponentially. The need to gather data from further afield at an increasing rate SD-WANs limitations start to bite. There will also be a a need to invest in larger bandwidth capabilities and data acceleration techniques. What’s certain is that data acceleration makes big data and predictive analytics increasingly viable. Machine learning can be used to help us humans to understand what story the data is telling us. Latency on the other hand can lead to inaccurate data analysis.
Go beyond hype
To me this just sounds like hype – particularly as WAN optimisation won’t necessarily increases WAN performance like it should do. On the other hand, data acceleration solutions can create performance increases. Your datacentres and disaster recovery sites don’t need to be situated within the same circles of disruption. Boosted by machine learning they can be placed thousands of miles apart, and as the transmitted data is encrypted it is very secure. The analysis of the network’s performance happens in real-time too, eliminating the risks of being reactive as opposed to being proactive.
Managing network performance, protecting your data, mitigating latency and reducing packet loss needn’t be the gamble that Boulton writes about. Mark Baker, CIO of JAS Global, felt he had to embrace SD-WANs because his company was already supporting global applications and email with MPLS networks and VPNs. The costs of running an enterprise resource planning (ERP) system over them worried him though. The ERP software required a sub-150 millisecond of latency. “Setting up and provisioning an MPLS system also takes several months”, says Boulton. Baker was therefore drawn to SD-WANs from Aryaka.
This is fine, but organisations should also look beyond SD-WAN to a data acceleration solution as it can do more for less. Many of Baker’s goals would probably have been achieved more quickly and more simply with one of them to address the latency challenge of having a global company “go from Atlanta to L.A. to London and Paris”. He adds: “But when you start talking about going across the pond or [to the northern and [southern] hemispheres there is a huge latency challenge to overcome when you’re lacking a traditional MPLS network”. With AI and machine learning, such a challenge is minimised – and that’s simply because machines can support humans effectively and sometimes outperform them. With machine learning behind data acceleration, you’ll always be a step ahead too.
