So, You're Being

Told to Implement

Released: November 6th, 2023

TL;DR (Too Long Didn't Read): Artificial Intelligence (AI) is a powerful tool with the potential to transform businesses by making processes efficient, decisions informed, and customer experiences personalized. However, implementing AI is complex and not one-size-fits-all. It includes various technologies like Machine Learning, Deep Learning, Expert Systems, Natural Language Processing, Computer Vision, and Robotics. Understanding these concepts is crucial. AI has limitations such as hallucination, bias, and transparency issues, which must be addressed. Implementing AI involves inspecting current processes, defining goals, aligning solutions, validating results, and deploying carefully while tracking metrics for effectiveness. A well-defined goal is essential for a successful AI strategy.

Let me start this off by saying that I have never seen a topic instantly take so much popularity and cause as much confusion as Artificial Intelligence; or as the kids say, AI. If you haven't heard of it, it's a transformative force that has the potential to revolutionize the way businesses operate, making processes more efficient, decisions more informed, and customer experiences more personalized. It comes to no surprise that company boards and executives are nominating an “AI Champion” and pressuring this person to implement it within the business. Now there’s one major problem, it’s not a one-size fits all solution, or even close. Now I don’t want to scare or deter you from going down this path, because there are many substantial upsides; but I most certainly would like to set expectations.

We will start with a brief explanation of AI, introducing some key terms and differentiate perception vs reality for some assumptions people are making. We will also dive into some things to watch out for that could cause a liability. Then we should establish a framework for implementing AI, because why not… everything else has a framework. Cybersecurity has many like NIST, ISO, etc, cloud has multiple versions of the well-architected framework depending on who you ask, and there’s others like FinOps, Digital Transformation, ITIL, I’m sure we could go all day and I rest my case. So, are you ready?

Understanding the Basics

Let's talk AI. What is it exactly? AI is a blanket term used for anything digital that performs some form of human task. It's code; it's a series of If/Then statements that progress from if this happens, then do this, and then do another action once that's complete. That is a very oversimplification of under the hood and this model has evolved since its inception to have the ability to train itself on what to do next depending on a confidence score of an output. Effectively, it does the work on an input, applies quality assurance, and then self-improves in a continuous workflow. That is called Machine Learning and connecting multiple iterations together to train a centralized hub or brain is called a Neural Network. This is where we achieve Deep Learning; check out the visual below to help explain the difference. In the current world of AI, everything falls under Stage 1 to the left. Even though machines can perform tasks much faster, they still don't have the human capability of feelings, instincts, etc so they can only make decisions based on calculated metrics. In the words of John Searle, "The reason that no computer program can ever be a mind is simply that a computer program is only syntactical, and minds are more than syntactical. Minds are semantical, in the sense that they have more than a formal structure, they have a content.” So this raises the question, will we ever have true AI?

Before diving headfirst into the world of AI, it's crucial to understand some fundamental concepts. It encompasses a wide range of technologies, and understanding their functions and how they play together will allow you to determine what you can use, where you can use it, and most importantly why. Let's break them down within their primary categories; Machine Learning, Deep Learning, Expert Systems, Natural Language Processing, Computer Vision, and Robotics. Everything is based on Machine Learning or Deep Learning but it's important to understand their concepts and values because the use cases are nearly endless.

Machine Learning - Comes in 3 flavors to describe how the models are trained. Models are a series of algorithms designed to enable machines to learn and make decisions based on data. These are supervised, unsupervised, and reinforcement, which are described below. Reinforcement Learning is becoming the most popular due to its "observe and learn" methodology.

Supervised

Training an algorithm using labeled data, where the desired output is already known. It's used for calculating decisions around known metrics like future sales projections.

Unsupervised

Relies on unlabeled data, where the desired output is NOT known. This allows the algorithm to identify patterns and structures within the dataset. Less around projection and more around correlation.

Reinforcement

The models learn by interacting with their environment and receiving feedback in the form of rewards or penalties. Like ChatGPT, you giving positive feedback on a response teaches the system it's a correct response (reward).

Deep Learning - Simulates the human brain with a heirarchial structure of Machine Learning algorithms. This enables the processing of extremely complex patterns and calculations with quality assurance. Deep learning allows an algorithm to perform a calculation on a set of data, deny it through QA, and send it back to the previous stage to run through a different algorithm through a process called back-propagation. Some use case examples are Generative AI, Natural Language Processing (NLP), and Image Recognition. Spoiler Alert: You will see these expanded further in the near future.

Expert Systems - Emulates human expertise in specific domains offering intelligent support and decision assistance. This is done via a human created Knowledge Base (although machines can search, transform, and aggregate the data) with a programmed set of rules to what information is to be displayed (Inference Engine). Various industries like financial, medical, law, and computer programming use these systems.

Natural Language Processing - Focuses on the interaction between computers and the human language. NLP is a Large Language Model (LLM) that enables machines to understand, interpret, and generate text or speech in a way that is both meaningful and useful. NLP starts with transcription, where there is a multistep process to break up the speech to understand meaning and intent; then it sends through whatever processing engine is applied. You will recognize this in use cases like chatbots and ChatGPT, Sentiment Analysis (understand emotion like happy, sad, upset) and Text Summarization, submitting or retrieving a bunch of text and writing a concise summary. Text Summarization will look extremely familiar to a TL;DR.

Computer Vision - Aims to replicate the human ability to perceive, interpret, and understand visual information from the world by processing and analyzing images or videos. Computer Vision algorithms can extract valuable insights through Object and Entity Detection (objects like people, animals, stop signs, etc, Entities are known objects like the Statue of Liberty) and Optical Character Recognition (letters, numbers, and symbols). These capabilities are used in Facial Recognition Software (FRS) and scan-to-text, License Plate Recognition (LPR) and handwriting analysis systems.

Robotics - Bridge the gap between the digital and physical landscapes and integrates AI with mechanical design and engineering. Robots can perceive their environment, process information, and execute actions based on AI-driven decisions. These AI decision capabilities are generally a combination of the above. Is the robot able to recognize the environment around it? Computer Vision. Can it listen and respond to human speech? Natural Language Processing with potentially a mix of an Expert System. I know the photo to the left looks like a human/robot slap-box competition but just think of the possibilities if we can react to AI in the physical world.

All of these sound amazing, right? Why not immediately implement them within your company? Believe it or not, AI does have its drawbacks and limitations even though it is being improved upon every day. For instance, a few terms you may have heard of, Hallucination, Bias, and Transparency. Hallucination is when an AI system, most likely NLP, hasn't been trained on a specific request so it just makes crap up. Bias means an AI system was only trained on an incomplete data set so it's outputs aren't very accurate. One of the primary concerns here is training a Computer Vision platform with only data from a specific ethnicity so it is inheritly racist. Same rules apply to any data analysis applications, crap data in equals crap data out. Last is Transparency, much like the concerns of Hallucinations and Bias, the need to see and understand how an AI model calculated its output can ensure these issues are mitigated. This is why the AI Ethics list came out to ensure developers creating these models ensuring Fairness & Bias (everyone has access to and the data isn't trained to be discriminatory), Transparency & Explainability (its decisions are clear and interpretable), and Privacy & Security (protecting user data and privacy).

Let's Talk About How to Implement!

Like most frameworks, this one is a revolving circle to show continuous improvements. Don't think once you deploy that there aren't any other improvements to be made as processes and technology will change. Go into this with an overall objective and ask yourself "What are the goals of implementing AI?" Are we going to improve someone's satisfaction like employees or customers, or improve efficiency and ROI? With AI, you can have both! Let's break these down and take a look.

Inspect: As always, before any decisions can be made, first you need to know what's going on. All frameworks start with a different wording of "learn the current state". It's very important to get multiple perspectives, not just leadership, to see what is happening in reality and not just ideally or perceptually. Get granular here too. If someone opens an email, copies the data to an application, and runs a series of steps to get the data entered into the end platform; then know each step individually. Processes can't be improved unless they are fully understood, this is why we perform Process Mining. Same goes for technology; is someone watching cameras or scanning images to manually look for certain object or people, what do they do when they identify something? This is the important data we are looking for.

Define: This is where we take the processes from our Inspect stage and look to improve and optimize. Using the information from above, think about what can be accomplished and don't be afraid to play the game "what if..." or "wouldn't it be nice if...". Be sure to Define what it should not do as well. Many AI conversations start with a blanket discussion on the current process pain points or what would improve customer or employee experience if they either didn't have to do a specific task or if the task can be done much quicker. In the Define stage, you aren't aligning solutions yet, but you are describing what the solution would need to do. So we aren't solving, we are characterizing. Does the solution need to identify, automate (Business Process Automation or BPA), or does it need to have the ability to interact with humans real-time? Sounds frustrating I know, because we are excited and want to talk solutions but they won't be 100% effective unless we know what and how we are solving the problem. Another important item in this phase is to Define what the current important metrics are and what the expected improvements should be. This will be your quantifiable success data and will directly impact ROI.

Align: Once you Define what a solution should do, now is the time to Align. Are you able to use a pre-trained model that has solved a use case similar to yours or do you have to customize? Everyone assumes there's a pre-trained model for everything and unfortunately that's not the case. Nearly every conversation I've had is, what AI software does this for us or works in the <enter vertical here> space. That would be nice, now wouldn't it? I mean, if you use your favorite search engine and type in Maufacturing AI for instance, you will get populated with multiple solutions with websites ending in .ai. Now will these fit the bill? Some are an obvious no, but we won't know until we move to the Validate stage. Also a good thing to remember, if the solution is custom, then it's generally more precise and accurate to business needs but also more expensive.

Validate: Like all frameworks and development lifecycles, before implementation you need to check yourself before you wreck yourself. Also, word to the wise...don't use production data to Validate, use fudged numbers or identification data to ensure there's no leakage. If the results pass your approval analysis, it could be qualitative or quantitative depending on how exact you want to be, then it's time to move forward. If not, then let's go back to the Align stage and find something better. If you are choosing a custom AI module, then what data are you going to train it with? Make sure it's complete and won't train to Bias and the results match expectations.

Deploy: Now we are ready for primetime, all validations passed and we are ready to implement into the business. Be sure to train and document for stakeholders prior and ALWAYS schedule the implementation to minimize downtime. Most companies will have a Change Management schedule and tools for tracking but if you don't, then create one for this. Trust me, it's important! Also, after implementation, you will need to track metrics for ROI to determine how effective the AI changes are. This is what the business truly cares about and if you have the data readily available then you will be a rockstar!

As you can see, implementating AI isn't as simple as a single solution but there are amazing benefits as long as you define the goals and objectives. Ask yourself: What specific problems do you want AI to solve? Whether it's optimizing internal processes, enhancing customer or employee safety and support, or gaining insights from vast amounts of data; a well-defined goal will guide your AI strategy and investment decisions. Now....JUMP!!!

You Can Find More Info on AI/ML Here.

Page updated

Google Sites

Report abuse