Artificial intelligence (AI) and machine learning (ML) seem to be all anyone is talking about these days. Much of what is being said concerns the consequences and drawbacks of using these relatively new technologies, but they can also create helpful efficiencies and improve processes when leveraged correctly.
Although artificial intelligence, machine learning and even deep learning are terms that may seem vague or interchangeable, there are some distinctions.
AI refers to a computer’s ability to learn and reason like humans. For example, through chats, it can interact and seemingly have a conversation with a person. Through visuals, it can scan a document and report back the information that is in it. Through audio, it can listen to something and dissect the words. In general, it uses our senses to interpret the world around us and translate that data into the computer.
ML involves algorithms that can learn without being explicitly trained. They don’t need to be programmed but they are trained and given inputs which allow them to predict what would happen from that standpoint. Many AI applications use machine learning models to interpret the question or input information.
Deep learning is a part of machine learning in which artificial neural networks adapt and learn from large amounts of data.
In other words, ML is a subset of AI and deep learning is a subset of ML.
A misconception about these technologies is that they are brand new; they have actually been around in some capacity for decades. In the 2010s, advances in processors and access to data accelerated progress in these areas, leading us to where we are today.
For years, AI and ML have been used in things like online targeted product recommendations. A shopper places an order for one thing and a “suggestion” pops up based on the buyer’s previous purchases. Or, if you search for something on one page and then start to see ads related to that on any subsequent pages you visit. At this point, nearly everyone has encountered the customer service chat bot. These are all examples of AI and ML in action.
While those are a few ways companies are harnessing AI and ML, there are less obvious applications. Operationally, a company can use it for process mining, looking at logs of what people are doing within a software application or on the company’s network to identify and build out things like their collections process. Personally, people may not even be aware they are using these technologies through their email provider, particularly if it suggests focus blocks because it is using their calendar to make informed decisions.
That said, generative AI is the technology that is really grabbing the headlines as of late. With AI and ML, one would develop a model, train it and put in new outputs to get a prediction out of the model. Generative AI takes it a step further, creating articles, images and other types of data that look or sound like the real thing, using whatever information is fed to it, fake or not. ChatGPT and DALL-E, the image generator, are examples of generative AI.
The results are impressive, producing lifelike pictures of people who don’t exist and music that wasn’t written by a human. They also have “authored” articles about historical events that sound accurate but are factually incorrect and have developed videos of things that look real but never happened.
Like many things, the technology is only as good as the information it is given. If used in nefarious ways, it can quickly go from extraordinary to alarming.
AI and ML make up the foundation of decision automation. Automating decisions gives companies the ability to execute faster and more consistently. Companies are already seeing ways they can make efficiency and productivity gains, improve speed to business, expand capabilities and business models, enhance customer service and upgrade monitoring.
These technologies can write computer source code, proposals, article summaries and can locate key information from written documents, making businesses more efficient and getting them to the market more quickly. For example, a company uses a large language model. To get answers to their questions, they would need to ensure the model is using information from a definitive source. There’s an opportunity for new business models that use curated datasets that could be used to source answers from large language models that are factually evaluated.
They could also assist with customer service interactions, answering questions faster and at all hours. It can help companies better monitor what is happening within their systems, including with their customers, allowing them to have a stronger pulse on what is happening around them.
Some industries will feel a different kind of impact. The reality for the law profession is that some large language models can be fine-tuned with a library of case law books and the related federal, state, or local statutes and ordinances and be able to provide an interpretation of the law in layman’s terms. That doesn’t mean a human isn’t involved. Many AI and ML applications use a human-in-the-loop (HIL) approach to send low-confidence results to a human to provide feedback. The HIL process is one way a model can improve over time. Additionally, a large language model could go through a number of legal briefs faster than a human could. It could also draft contracts or other documents if terms are specified.
Construction is another industry that is already seeing the benefits of AI and ML. Beyond employing robotics into a build, the vendors who do the 3D model renderings are already integrating this type of technology into the tools they have. A builder could enter in the necessary information, such as the size and height of the building and the type of business. The program can create a model without having a 3D or CAD engineer create the drawings. It can even tell them where they should put the electrical and plumbing.
Healthcare may be one of the more significantly affected, though. Data in healthcare is still not being used as effectively as it could be. ML could help pinpoint different new drugs that are available for patients, and eventually, AI could take vitals, considering the individual, their exercise and eating habits, and other factors to determine a recommendation specific to that person rather than to a cohort.
As exciting as these new applications of these technologies are, there are plenty of risks as well.
AI models can’t always be taken at face value. As it relates to performance, there is risk in what data the model was trained on and potential errors in the data. For example, early interactions with Chat GPT were not promising, with it giving bogus information and incorrect answers to questions. There are also security risks around what is being fed to the model and how it’s being used.
As the model starts doing more for individuals, there is a loss of control, which can mean a loss of trust and those using it may be wary of what information the model is outputting.
As alluded to earlier, certain industries may experience a negative economic impact since there is a high likelihood that some jobs will become obsolete with AI and ML.
Finally, ethical risks have been a major drawback of using AI and ML. Unfortunately, there are many examples of biased data being inputted into models, causing them to discriminate against certain individuals. Recognizing the possibility of bias and examining what may be causing it, is the first step in promoting inclusivity and belonging when utilizing AI and ML tools. Whether it be validating information, finding inaccuracies, weeding out bias or just connecting the dots that the model may be missing, the human element is still a vital part of the AI and ML journey. Developing a thorough audit process can help mitigate biases and better the tools in the long run.
AI and ML are already creating efficiencies and productivity gains, improving speed to market, expanding capabilities and business models and enhancing customer service for many organizations – and it is just the beginning. While these tools have many benefits to offer, they will impact the way we currently work and will require extensive due diligence to ensure we’re monitoring risks and using them appropriately and ethically.
Getting started with these tools can be daunting. Baker Tilly Digital has a three-step process for helping clients get started on AI and ML: ideation, building a minimally viable product (MVP), and then operationalizing and deploying the AI or ML model. We help companies identify and narrow in on opportunities that return tangible business value, identify the data inputs and other policies and regulations that may be required to follow, identify reporting metrics to be used as a baseline, and then determine candidate modeling approaches. Then once we have a working MVP we then show clients how the model can be put into production or deployed to be used in a line of business. This often involves evaluating AI services or selecting a foundational model and fine tuning the model with your data.
Beginning to leverage AI and ML in your organization involves many parts within each of the steps and often takes several months to complete. An organization that takes the time to fine-tune a model that is accurate and tied to business metrics will be able to use AI and ML to transform their way of doing business.
Connect with our team to learn how your organization can drive value from leveraging AI and ML.
This article was derived from the Leveraging AI and ML to drive value in your organization webinar, watch the full recording below.