Artificial Intelligence (AI) is one of the big topics right now, with daily articles and reports published about the great promises it’ll deliver, and how it will revolutionise our existing relationship with computers and technology……. leapfrogging our civilisation into the dawn of a new computer era. All exciting stuff, but before we get carried away with the possibilities being sold to us, let’s cut through the noise and hype and see what’s really going on, and ask ourselves two important questions.
In this blog post we share a technical and practical look at AI and specifically in Machine Learning, understanding what it is from a theoretical and philosophical perspective before doing a dive into the everyday applications and how this can transform your business operations.
Not many people realise the term of Artificial Intelligence as a concept is much than they thought. Ancient Greeks used the term “automaton” to describe self-moving and/or self-willed
entities, both in the real and mythological world. Over the centuries, this concept was almost entirely consigned to the philosophical discipline of science.
With the advances made in technology, engineering and in miniaturisation, Gottfried Wilhelm Leibnitz developed what is now considered the first mechanical computer capable of executing all four arithmetic operations – as early as 1671. During this time, Leibnitz also conceived concepts of machines capable of solving algebraic functions, encryption, and even differential equations.
Modern AI, however has its undeniable roots in the proceedings of the now famous 1956 Dartmouth conference, where John McCarthy, Alan Turing, Marvin Minsky, Allen Newell, and Herbert A. Simon among many more gathered to put in place an understanding of what AI would mean and set the foundations in place for what it means today (our emphasis):
The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.
Since that eventful conference, the field of AI maintained two fundamentally different schools of thought:
While it is arguably endlessly fascinating to delve into either topic in detail, Machine Learning through its techniques and technology is much better suited for most contemporary IT systems, and that is our focus on from hereon in.
Machine Learning is fundamentally based on two elements:
Neural networks are a simulation of actual networks of neurons found and extensively researched in the field of biology. They are a simulation, not only because it translates a biological system into a comparatively artificial system, but it also implements certain simplifications to make it more compatible with how computers work and process data.
Feedback loops also simulate a biological behaviour, expressed as adaptive behaviour. Succinctly speaking, feedback loops in ML tell the system the answer to the question “How good was I”, with the answer often being tripartite expressed as numbers, e.g. positive numbers
representing a positive or reinforcing feedback, and negative numbers representing a negative, dampening feedback and effect.
This feedback is then used by the ML system to adjust its internal parameters so as to improve its answers adapting to the feedback given – an adaptive system.
Often, ML systems are ‘trained’ before they are exposed to live data – the training phase comprises often extensive sessions of feedback-driven trial and error attempts before the supervisors deem the results as sufficiently accurate. This is called supervised learning, where the ML system is kept under guidance and detailed introspection of its human controllers.
Opposite to that are unsupervised training sessions, that are frequently employed to avoid human intrinsic confirmation bias that would otherwise skew the outcome. But unsupervised training of a ML system has its own issues and challenges. We leave that discussion for another time.
As we have seen, ML requires vast amounts of data to be trained on, and similar amounts of data to make sufficiently accurate decisions, or predictions depending on the use case.
Most obvious applications of ML in public services are predictive text suggestions in chat and collaborative text processing applications such as Facebook Messenger, Google Hangouts, WhatsApp, Microsoft Office 365, Google docs, and similar applications. Trained with vast amounts of real-life chats and documents, the system is trained to predict what you will type based on the last few words. This goes beyond autocorrect (which has its own Internet meme corner due to its failures; search for ‘damn you autocorrect’ in Google Search of Bing) due to its predictive behaviour. Critics may argue that ML is hence a glorified and overhyped statistical analysis, and there is a conversation to be had about this. But there are many more applications of ML that have real business value.
Pattern recognition; detecting hand writing faster and more accurate than humans
Consider the two use cases of a postal service with hand-written addresses, and HMRC dealing with large volumes of hand-written corporation tax forms CT600 (and all its supplemental forms) that cannot be submitted electronically. In both situations, with properly trained handwriting recognition systems, both accuracy and speed of processing can be greatly increased.
Thinking about your business; can you identify areas where large volumes of hand writing needs to be processed?
Pattern recognition; analysis of complex and often complicated symptoms
Pattern recognition is not restricted to visual applications. Consider a conversation between a GP (or even a specialist) and a patient. There is a language barrier in that patients often struggle to give accurate information (after all, how many patients are medical experts?) or information at all simply because they do not consider some symptoms important or related to the potential illness or ailment that plagues them.
Secondly, there are frequently hugely overlapping sets of symptoms between different illnesses, and it is often very difficult to come to an accurate diagnosis and subsequent path of treatment. Properly trained ML systems can help fill in the gaps in the patients’ narrative, suggesting to the GP to ask the patient confirmatory questions, and to offer a selection of possible diagnoses on a range far bigger than human GPs often are able to consider.
Consider a complex IT landscape where a fault in the system is difficult or outright impossible to explain or even fix. With similar complexity in overlapping symptoms of faults in complex IT systems, an ML based IT faults diagnosis system can help solving these problems much quicker.
Consider the challenge to protect IT estate from intrusion. How do you detect your IT infrastructure is under attack when complex and nuanced/sublime techniques are used (instead of rather blunt instruments like DDoS and brute force password guessing)?
ML intrusion detection systems such as Amazon Guard Duty come to mind.
ML has made huge advancements in recent times that by combining pattern/symptom recognition engines with natural language processing systems, large parts of first line customer services can be automated, given a significant amount of cases are of comparatively simple nature. This is often implemented as chat bots on an organisation’s support website. Being capable of handling mundane and often repetitive support requests with (very) acceptable natural language capabilities, customers often don’t realise they are chatting with an automated system. Ethical and workforce issues aside, ML backed chat bots can actually increase the customer experience with support staff not being bogged down by mundane and repetitive support issues but instead dealing with the intellectually more interesting and challenging complex issues.
We have all but given you a glimpse into the world of Machine Learning – we intentionally avoid referring to AI: While AI is as much or as little useful as ML, it is much less applicable to a broad spectrum of business use cases than Machine Learning. Not surprisingly, Machine Learning starts to make its way into virtually any aspect of business IT.
For example, use of ML in enterprises helps business operations by integrating algorithms into applications that enforce organisational processes while enhancing them – Enterprise cognitive computing (ECC). ECC apps like PS (project system), SAP, PLM, ML (material management) etc. are used to automate numerous mundane tasks and in doing so deliver great improvements in the field of business analysis and information accuracy. For example, ECC customer support application can be programmed to pick up calls within the five seconds on a regular basis. Another way to save time is document management automation. If your business requires constant document updates and exchange – document management software can automatically send documents to designated addresses while keeping track of all the changes and revisions. Many document management solutions incorporate and automate appointment scheduling.
Fixed assets can be easily managed with a solution that uses barcodes to track and update spreadsheets and documents. Human resource management solutions, help keeping track during the recruitment, performance reviews and leaves.
At Digital Craftsmen consultancy services, we aim to develop solutions that are trained, rather than programmed. As an MSP we are familiar with data management and in ML techniques which help you define a business case and support your operations in AI.
If you are recognising yourself and your business in the examples we illustrated, or if you have a use case that was not covered, please get in touch, and we are more than happy to explore the opportunities of ML with you tailoring it for your own specific needs.
You can always a contact our team on: firstname.lastname@example.org or call us directly on +44 (0)20 7345 7706.
We are ISO 27001 and Cyber Essentials Plus accredited, proud to be awarded Investors in People certification for our training and investment in developing the skills and expertise of our Craftsmen, because robots coming or not, there will always be a place for highly skilled experts in the technology sector.