19th December 2016

Artificially Intelligent beings will rule us all

A blog written by James Hickman, CTO at Virtual1

Not just a Science Fiction author’s prediction, but a one-time concern of Steven Hawking. A future the famed physicist is looking to avoid, with the first step being the recent opening of an AI research centre, funded by the Lever hulme Trust, at Cambridge University – a collaboration between Oxford, Cambridge, Imperial and Cal-Berkeley, which will look at issues such as long-term implications and uses, as well as regulation. The Government, conversely, is predicting that UK citizens do not have the skills to fully exploit advances in this area – perhaps they should speak to Professor Hawking!

Whilst we might laugh off the idea that androids will be doing our housework or deciding to wipe us out in some apocalyptic future, there is a trend towards a general acceptance of letting AI take more control of our world around us. The most obvious development are a is in autonomous cars with the likes of Google and Tesla seeing it as a fruitful area of research and future revenues. Ironic really, as Elon Musk, Tesla CEO – who, with other concerned investors, formed Open AI with a $1 billion investment last year to promote the open-sourced and safe growth of AI – had previously cited Google as the one company he feared could develop AI too powerfully, too quickly. Whilst in other areas, the US air force is also sending unmanned space craft into orbit whilst ocean-going survey vessels are more often looking for objects on the sea floor with the help of submersibles totally disconnected from a human operator for hours at a time.

This is all very interesting from an engineering perspective but in the world of networking, there is a lot less noise around similar initiatives. But you’d be wrong to think it isn’t taking root in a multitude of small ways.

I was speaking at LINX94 this year under the title “Humans are overrated”. The topic covered our projects to bring automation to the delivery of new circuits as well as for dealing with change requests in a way that reduces or removes the need for a human engineer. These systems are smart, but not true AI in the sense that researchers in the field would accept. In reality, there is limited need for a true AI mechanism in what can for the most part be transactional events with the occasional need to bring in an expert to break a complex change down into simpler ones and then enact them in the right order.

One area I do look forward to getting some AI help on though is in the area of diagnostics. When a handful of alarms go red on the NOC wall board, being able to quickly identify the source of the issue and apply any triage is the ideal outcome for both us as the provider and the end customer on the other end of the wire. This still relies on skilled engineers with a working knowledge of the infrastructure. To assist them, we are working on better correlation tools that gather all the alarms and other data to make a “best guess” as to the device or circuit causing the problem. This, along with some focussed diagnostic testing tools that they can run against the devices they know are in the path, all helps improve the service we provide to our customers. But we still need a human to review the outputs and use their years of experience to work out what is wrong. And the reassurance of speaking to a human when you call us with an issue, probably doesn’t hurt either.

Whilst AI might end up driving our support guys to work, we still need some good networking brains to keep the board green. At least until our robot overlords decide they don’t need us anymore.