TLDR:
The article argues that the sheer volume and complexity of data in today's networks make traditional human monitoring impractical, marking a tipping point for cybercrime. Deep learning and anomaly detection are presented as critical tools for identifying hidden threats—capabilities that extend to transformative applications in fields like healthcare. As cyber attacks grow more sophisticated, AI-driven solutions are set to revolutionize not only security but also other industries, heralding a new era of innovation.
○
A computer winning GO or $101 million stolen from the Bangladesh bank were big news, then as the saying goes “you ain’t seen nothing yet”. In the way sharing economies such as Uber and AirBnB have touched most people's lives in some form or the other in 2015 and 2016, Cyber Security and Deep Learning will touch most people’s lives in 2017 and 2018.
Now the question is, “Why are we approaching a tipping point for Cyber Crime?”
In a nutshell the sheer amount and complexity of data is moving far beyond the realm of humans to monitor and be able to spot interesting behavioural patterns. For example using steganography, it is possible to hide a piece of code or data in your photos and you might never know. Now, imagine you had a tipoff that data was being stolen via a handful of users’ photos, where would you start? Almost every staff member is carrying around thousands, in and out of your building every single day, not to mention SnapChatting and DM'ing them.
Network data is now becoming so colossal that it is way beyond the scope of ordinary humans to oversee. Network forensics can be extremely powerful, but when there are millions or even billions of touch points to look at, then trying to spot the modified file out of sequence cannot be achieved by an engineer looking down a list of timestamps. Of course data can be hidden anywhere, but photos are a nice tangible way of explaining that each of us is generating an ever increasing amount of data in which illicit payloads can be hidden.
Beyond the known issues, companies like Nehemiah Security (nehemiahsecurity.com) are focussed on detecting virgin exploits using techniques such as anomaly detection. This is an area in which Deep Learning excels far more than its human counterparts.
Neural Networks like humans are extremely good at learning patterns and classification. When a dissonant note in a symphony “jars” or an item on a shelf is facing the wrong way and looks out of place we are very good at spotting that. In reality if something looks out of place, not only do we spot it quite easily, we often find it very hard to ignore. We find ourselves putting that out of place item in the right place. Thus, anomaly detection is a part of our primeval defence mechanism. Anomalies sound alarm bells that we need to be on high alert, even if we do not always know why.
Where Neural Networks can help is by doing the boring job of looking for abnormal behaviour patterns 24 x 7 x 365 days a year without ever getting tired, losing concentration and on a scale that is way beyond us “mere humans”.
Will AI be enough outsmart the cyber attacks? Probably not. It simply represents an escalation of the cyber arms race, and the elite will already be looking at ways to both outsmart and utilise the new tools at play. A very old trick hackers use is rather than try all possible passwords on one account, to try the most common five passwords on every account, thus avoiding being locked out. AI can be used to reduce the amount of guesswork based on limited knowledge about each user and that is just for starters.
Cyber security is not the only sector where deep learning is going to have a huge impact. It is highly probable that people with cancer have already been cured with a reproducible set of actions and circumstances. Given the sheer numbers involved, and the disparate way in which data is collected, our human intellects have not been able to put all the pieces of information together and join the dots, but the machines learnig and deep learning can. Google certainly believes so and has done all deal with the UK NHS, why Google should get preferential access to that data is not entirely clear, but at least it is a start.
We know people recover from cancer all the time, but the problem we have is that we cannot accurately predict the circumstances in which we can reproduce the cure. Without being able to consistently reproduce the cure we cannot imply the mechanism. Again the same applies to type 2 diabetes, we have a clue that gastric bands and starvation diets can put people into remission but we are struggling to diagnose what less extreme measures will keep them in remission.
These are some of the problems that Deep Learning will be tackling in 2017. These markets are financially enormous, and they are based on innovation.
Deep learning will ensure exponential progress on those fields that have been hard for humans to crack ever since the industrial revolution by making large scale analysis of data far more profitable and effective than the constant treadmill of clinical trials.
Similar to the dotcom boom of the late nineties and early naughties. Some of the best ideas will emerge from small startups running “out of their garage”. Whenever that happens, it makes the space incredibly exciting to watch. This is a time when small companies can become unicorns…
2017 is going to be an exciting year for deep learning. Learn more at www.thehub.ai