artificial intelligence in the workplace
PHOTO: Shutterstock

This is part 3 of a 4 part article series, sponsored by Alfresco.

In this third article, we wanted to lift the shroud of mystery a bit and explain some simple and practical concepts to help you get started on your own AI journey.

The term AI (artificial intelligence) is a bit of a blanket phrase that is used to cover so many different things today. In our mind at least it is analogous to saying “baking”, we all know what baking is, but are we talking about, bread, cookies or cakes? Moreover, if we are explicitly talking about baking bread, is that sourdough, rye, flatbread or dinner rolls? The term AI is a lot like that; we could be talking about deep learning, machine learning, neural networks, or anything in between.

Knowing the basic differences between each is essential, not just to the data scientist but to the practitioner that may be buying or using AI in their business. The good news is that you don't need to know anything about algorithms or advanced math, you need to know why each is different and where and how you can effectively apply each one. For once in our lives, a little knowledge can go a long way.

Machine Learning

The most likely form of AI that you will encounter in the enterprise is Machine Learning (ML), as the name suggests it's a system that learns over time. It does so by learning to identify patterns in data. ML tools use different models to undertake different tasks. There are many such models, but common ones include, Decision Trees, Random Forests, Hidden Markov, Naïve Bayes, and Support Vector Machines. Each works quite differently from the other, though they can also be used in conjunction with one another.

A random decision tree - artificial intelligence
PHOTO: Shutterstock

Decisions Trees go through a process of answering a hierarchy of yes and no questions to get to a result. For example:

  • Is an income higher than x? (yes or no)
  • Are the outgoings lower than y? (yes or no)
  • Did the income drop in the past year (yes or no)?
  • And so on.

This is a simple but powerful and highly intuitive method that finds regular use in rules-based business activities like credit risk assessments, or for that matter dating applications.

Does this person like heavy metal music (yes or no), do they like foreign travel (yes or no). You get the idea, it's pretty simple, and something we all do ourselves every day when we make decisions. Do I need exercise (yes), do I want to go to the gym (no).

In contrast, Hidden Markov Models (HMM) are a machine learning clustering method that is used, for example, in weather predictions. HMM use what are called hidden state analysis to predict observable states. An example of a hidden state is high/low air pressure. By contrast, an example of an observable state is sunny/rainy weather. Another example of a hidden state could be the emotion anger — an observable state may be a lowered brow in a face or a raised voice. Audio waves are also hidden states we can cluster and then associate with an observable state like specific sounds or words. In short, there are many different models of ML that work to solve different problems, these are the tools you will use most often in enterprise content and process jobs in the back office, and in customer experience situations.

These ML are, depending on the quality of your data, highly accurate and efficient. They can transform many currently tedious, slow, and manual activities. However, they are not the whole of the story; we do have technologies that are by many measures much smarter and approach more closely the concept of mimicking human intelligence. Here is where we get into the world of neural networks and deep learning.

Neural Networks and Deep Learning

There is a big debate as to whether these advanced technologies really do mimic human intelligence, either way, they can do many of the things humans do and at least on the surface do so in a way that seems to mirror human thinking. For just like the human brain, artificial neural networks, link together multiple functions that primarily communicate back and forth with one another depending on the task in hand. Each function takes an input, applies some logic to it, and outputs a result to the next function in the network. The significant advantage here over ML techniques is that we train ML systems, neural networks train themselves. They can learn and model highly complex relationships and concepts. For example, you might use neural networks to identify a specific face or object in a picture or video. A task that is much more complex and difficult than identifying a word, letter or a number in a document.

Neural Networks
PHOTO: Shutterstock

We think it is important that you the buyer and user understand that there are different tools, approaches and models for AI & Machine Learning, as that empowers you to effectively start and expand your use of these services in your organization. Most of the time though your first encounter with an AI will likely come through a packaged application. For example, Alfresco Intelligence Services, that pre-integrates third party AI services, such as Amazon Web Service’s AI services for NLP (Comprehend), Image Analysis (Rekognition), and extracting Text and Data from images (Textract). Such a service normalizes and stores the AI output and allows you to call up and uses wherever it is needed in your business processes.

Decision Making Technologies That Can Learn

Once again, AI is a blanket term for a lot of different technologies, that can learn and make decisions. Some systems are relatively easy to unravel and understand. We can figure out with some systems why a loan or insurance claim got declined as we initially set the parameters for the decision making. In more sophisticated neural networks, networks that continue to evolve by themselves, we may not be able to unravel why they came to a particular decision.

Sometimes, sophisticated and powerful may not always be better; we have to choose our approaches, tools, and models carefully to ensure they are the right fit for our particular requirements. We have to decide how “transparent” the system needs to be, can we unravel it, or don't we care? Some systems require much more human supervision than others; some require more data, some less. Some require much more computing power than others, and some are much more expensive to run and manage. It's a constant trade-off to come to a decision on which system works best for you and your needs.

Chances are you need help and guidance from experts, you will likely use pre-integrated applications, optimized to meet specific needs. Even so a little homework in advance can ensure that you ask the right questions and can be fully engaged in the process of leveraging AI in your organization.