Competitive Advantage

Competitive advantage is unsurprisingly often stated by companies as their number one reason to embrace Artificial Intelligence technology. Companies and organisations who have started to use AI are already seeing large returns on their investment. Here's how AI offers an advantage over your competitors:

• Make better, informed more accurate decisions and predictions at all levels of business operations
• Increase productivity through more efficient process design and decision points
• Spot positive and negative trends that would be otherwise impossible to observe and understand
• Automatic fraud and anomaly detection
• High level predictive analytics
• Rapid automation of critical data-heavy processes
• Enhanced scalability across all levels of business
• Much greater coverage of decision points and variables through the entire business structure
• Massive fault decreases in areas otherwise prone to human error
• Better understanding of customers, staff and stakeholders through historical analysis and predictions
• Real time response to diverse unpredictable events
• Reduced costs spent on traditional systems and software development
• Minimising loss and maximising gain at all levels
• Constant real time learning and adaptation throughout all business operations
• More streamlined, personalised and relevant services to customers resulting in repeat sales
• Automatic Identification of profitable markets and leads through data analysis
• Solve unique technical, business or operational problems that would otherwise require intense time and cost consuming analysis and resource
• Many, many more...

---

Specific tasks

As well as gaining a competitive advantage in the overall business market, Artificial Intelligence can also be used to greatly speed up analysis, design, innovation, response and automation in a wide variety of tech, science, engineering and related sectors, all of which can be made much more efficient using AI, therefore increasing accuracy and quality at the same time as reducing time, cost and resource requirement:

• Predictive analysis
• Forecasting
• Operation mapping
• Scientific data analysis
• Drug design
• Active risk management
• Image, video and voice recognition
• Cyber security
• Behavioural analysis
• Medical diagnostics
• Enhanced surveillance
• Oil exploration
• Fire protection
• Emergency response
• Engineering design
• Weather forecasting and climate change
• Asset protection and management
• Many, many more...

Tech Basics

At the heart of most AI systems is the very real concept of Machine Learning. Machine Learning is the ability of a system to learn for itself, rather than being told what to do. Machine Learning works by utilising learning-algorithms as opposed to stepwise computer programs. There are many ways of utilising Machine Learning algorithms, ranging from implementation of mathematical formulae derived from Data Science right through to fully immersed Deep Learning processes that make use of Artificial Neural Networks.

This is what makes Machine Learning (especially Deep Learning) extremely powerful: it can work out all the possible combinations, variations and links within and between data extremely accurately - to a degree hundreds of thousands of orders of magnitude higher than any computer programmer could code, and generally faster than any human could do the same task (especially repeated tasks) in the same time frame.

In general, Machine Learning aims to learn how any dataset and parameters within it are intrinsically linked to each other and it does this with varying degrees of autonomy - this is where the lines between Data Science, Machine Learning, Deep Learning and Artificial Intelligence can become a bit blurred and misunderstood, so its worth defining the differences at this point:

Data Science

This is the use of algorithms, generally by scientists and mathematicians to extract meaningful knowledge from data - this is not Artificial Intelligence or Machine Learning as the process is entirely human-led and generally, specific data fields and patterns will be looked for and tuned by a human user.

Machine Learning

Machine Learning is the most important aspect of Artificial Intelligence; it is the ability of a machine (computer) to implement algorithms to carry out a learning process and it can be achieved by implementing specific data-science-derived algorithms or from the running of models than mimic how the brain works (and anything in between). As soon as a Machine starts to learn, whatever method we use, we're in the area of Artificial Intelligence - however the degree or level of Artificial Intelligence with respect to Machine Learning is down to the amount of autonomy.

If a Machine Learning system is simply running an algorithm derived from Data Science, looking for specific patterns, although it is learning, and is therefore Artificial Intelligence, it is low level Artificial Intelligence because a relatively large amount of human input is required and specific fields and patterns can be looked for. On the other hand, if a Machine is learning through modelling of the human brain, as is the case in Artificial Neural Networks, this is high level Artificial Intelligence as the learning is able to function with absolute minimal input from humans - yet is still able to work out links between data.

Deep Learning

Deep Learning is a high level form of Machine Learning/Artificial Intelligence where a machine is able to learn due to the presence of multiple Artificial Neurons - models of human brain neurons that connect with each other using a mathematical algorithm to link with and influence each other using a weighting system (mirroring how how the brain works). Crucially, this network of neurons acts the same regardless of what data you put in - and is therefore very distinct from data-science derived Machine Learning which is usually specific to the data being used. Each artificial neuron receives an input or inputs, changes its state and then gives a relevant output to the rest of the network. Artificial Neurons are arranged in layers: input layer, hidden layer(s) and output layer. The hidden layer is where all the magic happens - and can be a single layer, or is commonly made up of multiple layers; if there are multiple layers it is called a Deep Layer which gives rise to the term Deep Learning.

The data

AI harnesses the power and value of all your data

The data passed into an AI system is what it learns from, and generally speaking the more data, the more it will learn and the more accurate and powerful the AI will be. Data can be anything you choose e.g. visual data, customer data, staff data, financial data, weather data, process data, manufacturing data, legal history data, engineering data....anything - AI will make the links that would be impossible for a human to do in the same time frame.

Supervised, Unsupervised and Reinforcement (reward-based) learning

When an AI system is learning, the term is Training and to get the most out of your AI system, it is crucial to get the training phase correct. AI Machine Learning happens in 1 of 3 ways:

Supervised Learning

In supervised learning - The most common form of AI learning - AI is supplied with data where we know what the inputs and output fields are and the AI uses Machine Learning to link known inputs with known outputs. e.g. in a very simple case we may have inputs of "Age", "Sex", "Location", "Credit Score" and simple outputs of "Good customer" and "Bad customer" based on our historical data. AI would learn the links between all these data so that when we pass it information of a new customer it will be able to tell us if this customer is likely to be "Good" or "Bad". In reality, you would likely have dozens, hundreds or even thousands of inputs from which the AI will learn from (e.g. for image recognition systems where each pixel or group of pixels is represented as a single input you can have hundreds of thousands of inputs) - which is why it's so powerful - making links that a human or traditional computer program would simply not be able to do in the same time frame.

Unsupervised Learning

AI is supplied with data as above however we don't know what the inputs and outputs are defined as - it is up to Machine Learning to work this out. E.g. in the case above AI would look at each field or combinations of fields as inputs and outputs and work the out the links between them all.

Reinforcement Learning (reward-based learning)

Here AI again learns through Machine Learning, however it learns by rewarding itself if it gets a particular action correct. It does this by comparing the environment before and after an action and determining if the action was beneficial, and if so, feeds this back into the learning process.

How do I actually start using AI, now?

Hopefully we have convinced you of the huge benefits of using AI in your company or organisation. However, we understand that starting to use AI for real can be quite daunting for some, a bit of a culture change for many and definitely a thought-provoking investment for all.

Below is our advice on getting started with AI. You can of course do this all yourself, however Fennaio is here to help you at all stages of AI Integration - we have the industry knowledge, and technical and management knowhow to get AI into your business efficiently, including AI design and build, education, provision of all infrastructure, system integration, delivery and support.

(1) Necessity, survey and plan

The first thing to do when considering a move into using AI is to look at your current business operations and identify areas that could be candidates for Machine Learning and automation, of which there will likely be many: typically areas of data handling/analysis, process management, trend spotting, decision making, predictions, forecasting, customer analysis, visuals, sound and biometrics.

This will include a critical appraisal of "where am I now, and where do I want to be?" with the emphasis on improving efficiency - at this stage it's a really good idea to heavily scrutinise your current ways of working and also have a detailed look at what data you have or collect during all your operations.

You know your business better than anyone else, however Fennaio can help you look at your entire business operations in order to identify areas where AI could be efficiently and effectively implemented.

(2) Infrastructure

In order to run AI you will need the correct infrastructure in place which at the very least can be a single computer, an internal server, or alternatively a cloud-based server. In many cases you will need a network to run the AI system over. Obviously this depends on the size of your operations, the amount and complexity of data and the amount of Machine Learning/AI activities being catered for.

A typical set up may involve a cloud based server dedicated to running AI/ML on, with a network connection to your internal network that can access the AI server when required. AI can run on a standard CPU, however complex Machine Learning tasks can utilise the power of a GPU to speed things up by up to 50 times.

Based on the survey above, Fennaio is able to quickly ascertain the level and complexity of AI infrastructure you need for operational effectiveness, durability, scalability and maintainability. Moreover we are able to supply the necessary hardware, relevant network solution and cloud solution if required.

(3) Software

At the heart of AI is Machine Learning software that for example when using Deep Learning, models an Artificial Neural Network using an AI algorithm. The Artificial Neural Network software needs to carry out AI training from a variable dataset and provide an output when supplied with real time data. Typically Machine Learning software is built using Python and R coding languages.

Fennaio provides AI software capable of handling any ML task, no matter how complex the data are. We have a series of in-house built off-the-shelf turnkey software solutions suitable for the majority of AI tasks, that automatically parse data, carry out supervised, unsupervised and reinforcement learning and integrate directly with your existing systems. We are also able to create bespoke AI software depending on your specific AI requirements.

(4) Integration

A key part of AI is the ability to integrate seamlessly with your existing operations, if required. E.g. you may have a powerful AI system that has been trained on a huge amount of customer data, and it is primed for you to query it in order to get some meaningful output or prediction from in order for you to use it beneficially in your systems and operations.

This is where Artificial Intelligence Integration comes in which in essence means linking systems together in real time. There are several ways of doing this including directly coupling software at the code level or linking systems using network protocols - with the sole aim that data can be easily passed between the two (or more) systems.

Fennaio is expert at AI and Machine Learning integration with other systems. We have collectively decades of years experience working with a vast array of systems in diverse industries largely including web-based or network-based software built in a multitude of coding languages. Our particular favourite method of integrating AI is via HTTPS (secure web based protocol) meaning that all systems connected to a particular network can access the AI system using a uniform communication pathway. This is suitable for almost all AI systems where response times in the milliseconds range is adequate. For systems that require microsecond range coupling we can also directly integrate AI software directly into existing code, if required.

(5) Data Preparation

Once you have secured your AI infrastructure and ML software, the next step is to train your Artificial Neural Network, and this means understanding your data and preparing it for use in the AI system. Generally speaking this will involve:

Converting data to numbers-only, and normalising it

Consider an example dataset where you are Pharmaceutical company wanting AI to predict the suitability of a patient for a clinical trial so you don't waste anymore time sending the wrong type of patient. In the past you gave out 10,000 multiple choice questionnaires, each with 100 questions that were answered (A to E) and you followed the patients up a few months later to see their outcome e.g (Suitable and trial conducted, Unsuitable, Suitable but did not attend trial). In this instance you would convert A to E to 1 to 5 and numericize the results as 1 to 3. You could then normalise each representing each value with respect to the highest number in the field list.

Fennaio AI software automatically numericizes and normalises data, and this process can be easily configured using a GUI

Deciding on inputs and outputs

In the simple Pharmaceutical example above it is clear that the inputs are the questions and the outputs are the trial results, which is a classic supervised learning pattern. You may however have no defined outputs in the case of unsupervised learning. Whatever you choose, deciding on inputs and outputs is crucial in order for the AI to learn based on what you will provide the trained Neural Network with in the future for real time results.

Fennaio AI software makes it easy to select your inputs and outputs using a configurable GUI

Formatting the data for use in the AI software

Once you have numericised and normalised your data, and defined inputs and outputs, you will need to pass the data into the Machine Learning software in a way it understands - typically Comma Separated Values (CSV).

Fennaio AI software automatically formats input and output data so it can be directly used in Machine Learning

(6)Training

When your data is sorted out and usable in the AI system, the most important step now commences, training. Here the Machine Learning software will iterate over your data multiple times to learn from it (specifically learn how to get from the inputs to the outputs). Depending on the size of your data, this process can take from a few seconds to several hours. In exceptionally large datasets or for reinforcement learning or unsupervised learning this can take days, depending on the processing power of your CPU or GPU.

The result of training is a fully trained and primed Machine Learning model that is ready to accept real time data and give you an answer. For example in the Pharmaceutical case above you could provide the trained model with a newly-filled-out questionnaire and it will almost instantly give you a predicted output of patient trial success based on the previous 10,000 questionnaires it has learnt from.

Fennaio AI software contains self optimisers that gradually scale up the learning process and once it is self-satisfied it is using the most efficient ML process it will start the full training cycle. This ensures your AI trains as accurately and efficiently as it can.

(6) Monitoring and Visualisation

When your AI system is up and running, and using real time data against a trained model, it is important to monitor it initially and periodically to see if it is giving the results expected and is generally adding benefit to your operations.

Fennaio software provides a range of visualisation, monitoring and test tools in the GUI for you to test the AI system at any time. Visualisations are largely in the form of charts, graphs and text which are numerous and configurable

(7) Further Learning

It is always a good idea to constantly train your AI and Machine Learning systems with current data in order for the model (whether it be an Artificial Neural Network used in Deep Learning or lower level algorithm) to be as accurate as possible. Moreover, if it is clear your AI is "getting it right" - it is useful to start considering reinforcement learning in order to maximise the accuracy of the system.

Fennaio AI software has in-built constant learning, follow-up learning and reinforcement features that enable you to constantly get the most out of your AI system

Concerned over the use of AI? What are the risks?

Despite its relatively increased widespread use, many people are concerned about the use of AI. A few people still imagine AI to be all about robots - and in some communities and individuals there is a genuine fear that "robots will take over the world".

Although that is highly unlikely, and I hope in this section we have described the real-life uses of AI, it is definitely worth discussing the plausible concerns of AI usage - and how we can mitigate against them, to ensure AI is being used responsibly, ethically and for good reason. Key areas of concern for people undertaking an AI journey for the first time:

Internal Security

Risk/concern: There is a risk that AI can be manipulated to be used for nefarious purposes - especially as the AI coding industry is relatively small and niche and therefore developers with specific skillsets could introduce malicious code into AI systems without the owner knowing what's happening.
Mitigate: Use trusted sources, such as Fennaio when anyone is providing AI systems, software and consultancy to you. If possible, always have a member of your own technical team review work carried out by third parties.

Customer Trust/concern

Risk: There is a perceived risk that customers could be put off if they think you are using AI
Mitigate: This is about transparency and understanding. It's your call, but we recommend being open and honest about your use of AI in your company. When customers understand the reality of AI and how it is used, they are accepting - much like the use of Cookies and the common GDPR statement on most websites has become widely accepted now.

Legal and Liability

Risk/concern: Who is liable if AI goes wrong or provides a catastrophic prediction; the AI provider, the company who uses AI or the AI itself?
Mitigate: Like any piece of software, specific terms, warranties and service level agreements should be defined in the contract between AI provider and AI consumer. Always use a trusted provider, such as Fennaio, and always follow careful procedure when starting out in AI. Make sure your dataset is fit for purpose, the AI is scaled up at a reasonable rate, the AI is behaving as you'd like it to, and the AI is integrated correctly with the rest of your systems and operations

Control

Risk/concern: AI completely overtakes a specific process or operation
Mitigate: This can be viewed as both positive or negative depending on what you want from your AI. An AI system is more than capable of taking over a complete business operation if it is set up, taught and integrated in such as way to do this, however it is highly unlikely this would ever happen on its own accord or by pure chance. To mitigate against AI leakage into other systems or unwanted behaviour - proper overall system design needs to be carefully adhered to.

Why are you not using Artificial Intelligence and Machine Learning yet?

When we carried out our initial research in April 2020, we spoke to many companies and organisations of various sizes and staff from all levels within these businesses. Most were not using AI, however most could see where it would be valuable and 75% had an appetite for its use. For the companies who actively wanted to use it (generally medium to large sized companies) the common reasons for not embracing the technology yet were:

* Talent

Simply not enough skillset, knowledge or experience in the current workforce to understand, build and manage and move into AI

> How Fennaio can help with Talent

We are PhD qualified and highly skilled, experienced software developers in the field of AI and software in general, having worked on hundreds of systems at high level from single user to enterprise grade, from initial concept to delivery and beyond

* Investment priorities

For some there was a perception that finances should be directed to known quantities before using AI

> How Fennaio can help with Investment Priorities

We hope through reading this article and increasing your knowledge of AI that we have shown you the massive benefits of using AI. We are happy to demo our software at any time to quickly show you how easy it is to bring in AI and start realising the benefits. Now is the time to invest in AI - companies who are doing so already have and will continue to gain a competitive advantage

* Viable business case

A few companies we spoke with were concerned about putting a convincing business case to shareholders for the use of AI

> How Fennaio can help with Viable business cases

We found this was generally due to lack of understanding of what AI actually was and what it was capable of as many of the business processes we discussed were definitely suitable candidates for improvement by AI. Fennaio can help with full understanding of AI using easy to digest demos and accessible terms. We can show you how the adoption of AI can immediately reduce time and costs of many processes and operations.