Machine learning is continuously unleashing its power in a wide range of applications. As enterprises innovate and develop their machine learning capabilities a lot of questions are raised in the process. Are we ready for artificial intelligence? Can anyone keep up with the likes of Google, Amazon or Apple?

To cut through the noise, Wandera attended the highly anticipated AI Tech World, where we spoke on a panel about the challenges of machine learning at an enterprise level. Here are our key takeaways from the day.

Defining AI

The panel kicked off with a discussion over the true meaning of AI, and this centred on the work of the late, great computer scientist, John McCarthy:

It is the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable

By this, McCarthey meant that we shouldn’t necessarily confine our AI to being ‘like humans’, instead AI can solve the same problems in entirely different ways to the way you or I might.
At Wandera, machine learning is used as a tool to solve specific problems, like malware detection, to then feed an AI in the form of our mobile insights and analytics engine – MI:RIAM. AI isn’t confined by the same biological rules as humans are.
Instead MI:RIAM uses Machine Learning as a tool to draw conclusions for humans to understand and develop.


The traditional computer science and programming principle of ‘garbage in, garbage out’ is overwhelmingly true of any machine learning model. Without some human direction machines can be fed the wrong information and will attempt to solve the wrong problems. Therefore, it’s imperative to tune and update the model regularly, or it will not cope with new situations or data.
The understanding between data and its application is where the true power of data science lies. Without this and the use of tools, mathematics and coding, a data scientist will sink fast.
Wandera’s data science team are able to take the plethora of data generated by the proxy-based solution used by hundreds of global enterprises, and extract it to gain truly actionable insights. From analysis of devices up to anomaly detection, every project is well investigated and assessed until it’s ready for production.
Getting there takes a lot of work and data scientists often struggle with the speed at which they can get new ideas into production. Without real and well timed data a new machine learning model will not be put to test, and this is why many organizations fail in understanding the real value of implementing machine learning.

AI.Monitoring and reinforcement

The idea of (safely) pushing a new model into production as soon as possible comes with its own challenges and having a clear process to utilize feedback within the data science team is the next key step.
At Wandera we define various metrics and processes to ensure MI:RIAM is kept up to date, using a mixture of other machine learning models to run dynamic checks or via our expert threat research team. This allows us to collect statistics and decide when our models need new data or retraining outside of their regular schedules.

Risk, confidence and communication

The final area of interest came from the audience:

How do you get product owners and managers on board and invest in implementing machine learning? And how do you then communicate this to a sales rep etc?

The simple answer is that it’s not always easy. At Wandera we’re lucky to have a skilled team with a wealth of technical knowledge and expertise. Their passion and drive helps in communicating the value of machine learning to other parts of the company. Finding common ground or a common problem you’d like to solve early on helps facilitate this.
As always language and knowing your audience is important. If you can’t communicate the value of machine learning succinctly across the enterprise, then you’ll forever be fighting a losing battle.
Start by implementing features that you can measure easily in ways that everyone understands i.e. getting a probability of outcome, or describable features. Once this is established and confidence in the science is normalised, other AI or machine learning features are much easier to communicate and maintain for future initiatives.

Future of AI

The final point on the panel was the harm that the film Terminator has done for AI. We love Arnie, don’t get us wrong, but SkyNet is so far from the current and future truth that the mainstream media have blown out of all proportion any mention of AI from the tech world.
The power still remains with us to ensure that the ethics of AI are maintained, and at Wandera MI:RIAM is used to defend and secure the mobile fleets of our customers in new and exciting ways. We even have hopes of machine improving how we work in further iterations. But MI:RIAM won’t be taking over Wandera for a long time, or has she just done a panel and written a blog post? I guess you’ll never know.
[text-blocks id=”threat-advisories”]