Introduction To Artificial Intelligence For Business Professionals
Full Stack Software Engineer skilled in Traditional and Generative AI
AI offers a vast array of tools for any forward-thinking business leader. But, where do you begin?
This article is an introductory guide to AI, making its complex aspects easier to understand and showing how it can be used in business. Whether you're already interested in technology or new to AI, this exploration will give you the information and insights you need to use AI tools successfully in your business strategy.
1. Key AI Concepts
You've probably encountered the terms Artificial Intelligence, Data Science, and Machine Learning. They're often used interchangeably but have unique meanings. Here's a list of essential AI concepts to clarify their differences.
Artificial Intelligence (AI): Refers to the field that is focused on creating systems, robots, and machines that behave more like humans.
Machine Learning (ML): One of the most crucial aspects of AI, Machine Learning employs mathematics, statistics, and probability theories, among other tools, to identify patterns in data. By identifying these patterns, predictions and inferences can be made, also, a deeper understanding of the data.
Deep Learning (DL): A subset of ML, it utilizes a structure called a Neural Network, also known as an Artificial Neural Network (ANN). This structure enables the processing of vast amounts of data, including complex and unstructured data, for example, text in PDF documents, images, and videos.
Natural Language Processing (NLP): Its focus is on enabling computers to effectively respond to human language. Examples include analyzing sentiment in social media comments, summarizing text, and detecting spam, among other tasks.
Computer Vision (CV): This field allows software to extract information from images and videos, such as recognizing faces, identifying objects, or categorizing images based on their content.
Generative AI (GenAI): Utilizes ML to generate new content based on the data the model was trained on. This could include images, text, code, or music, among other things (you'll gain a better understanding of what "model" and "training" mean later in the post).
Data Analysis: It's about collecting, cleaning, processing, and organizing data to find hidden insights, summarize the data, validate hypotheses, or support decision-making. It focuses on understanding the current state of the data and providing insights based on past information. The Data Analyst is like a "detective" who tries to find useful information in the data to answer questions or solve problems.
Data Science: Uses statistics, math, machine learning, and programming to find hidden patterns in the data. These patterns help predict future trends, make decisions, and solve complicated problems. It looks ahead, trying to guess what the data will be like in the future.
Data Engineering: It is focused on setting up the infrastructure and tools necessary for data collection, storage, and analysis. Data Engineers have many important responsibilities. They create and manage data pipelines, keep data accurate and consistent, and make data storage better, among other responsibilities.
Machine Learning Operations (MLOps): The goal of MLOps is to maintain machine learning models in production reliably and efficiently, it is focused on automation and monitoring at all stages of ML system development. This includes integration, testing, releasing, deployment, and infrastructure management.
There are many concepts, but I'm confident that you now have a better understanding of most of them. If you're a visual learner, this Venn diagram should help clarify some of them even more:
1.1. Differences Between A Machine Learning Project And A Data Science Project
As you can read, both Machine Learning and Data Science work with data, but they have distinct objectives and methodologies.
A Machine Learning project is focused on creating models that perform tasks such as classification, regression, or recommendation. For example, a system that classifies emails as spam or not spam, or an e-commerce recommendation system.
A Data Science project employs analysis, visualization, and machine learning techniques. This means that involves building models (not always), but this time the focus is on extracting insights from the data. Usually, the output of the project is a report, presentation, or data visualization dashboard rather than a standalone application.
Both of these concepts are different, but there's some overlap that might be a bit confusing. You might have noticed the word "model" popping up again, which highlights its importance. But don't worry about it, you'll get a better understanding of what a model is as you read further into the post.
I hope the following Venn diagram helps you visualize the relationship between Machine Learning and Data Science.
2. What AI Can And Can Not Do
AI has sometimes been "sold" as a super amazing tool that can do everything and replace everyone. You might have heard about the "I, Robot" movie as a vision of our future. But don't worry, that's just a misunderstanding some people have. As a business professional, it's essential to know what AI can really do, so you can use it confidently in your organization.
2.1. What AI Can Do
I bet you're already familiar with what AI can do these days. Just to give you a clearer picture, let me quickly touch on some of its applications across different industries.
Retail: AI powers recommendation engines, as well as helps predict trends, optimize pricing, and enhance customer experiences. For example, an e-commerce clothing website may suggest complementary outfits and adjust its stock based on real-time demand.
Finance: AI applications in finance encompass algorithmic trading, detecting fraudulent transactions, and assessing loan risks, among other uses.
Transportation: In transportation, you'll find not only the well-known autonomous vehicles but also vehicle and route optimization.
Manufacturing: AI can identify potential equipment failures before they happen, which prevents costly downtime and ensures smooth operations. For example, machine sensors and algorithms can work together to optimize production lines and make them more efficient.
Healthcare: AI assists in diagnosing diseases from images, predicting patient outcomes, and personalizing treatment. For example, AI systems can analyze X-rays or MRI scans to detect cancers or tumors at early stages.
2.2. What AI Can Not Do
AI is definitely an amazing tool, but all the hype around it has caused some misunderstandings. Even with all the progress we've made, there are still some things that AI just can't do yet. Take a look at this image, for instance:
Let's imagine a self-driving car driving around the city when it comes across a scene like the one in the image above. What's the cyclist doing? Is he saying hello, signaling to stop, or pointing out a right turn? And what about the child? Is she just playing or also signaling a right or left turn? It's tough for the autonomous vehicle to figure this out. You have probably understood the idea, one of the limitations of AI systems is that AI systems cannot understand human intentions.
Some other challenges AI systems face are Emotional Intelligence, Ethical Judgment and Morality, Cultural Sensitivity, and Long-term Planning and Strategy, just to name a few.
3. The Importance Of Data
You might have noticed that the word "data" keeps popping up in this blog post, and that's no accident. Data plays an important role in the world of Artificial Intelligence.
High quality and quantity of data are the secret ingredients that transform AI systems into amazing applications. For example, those recommendations you get from social media or streaming services are the power of data at work in AI.
Remember that machine learning is all about discovering patterns, trends, and connections in the data it's trained on. So, the more high-quality data we have, the better our system will be at giving recommendations (continuing with the previous example). But if our data is unreliable, incomplete, or inconsistent, what kind of recommendations will our system provide?
In the last couple of years, the term "Data-centric AI" has taken on a clearer shape. It encourages companies to concentrate on developing better systematic practices for improving data in ways that are reliable, efficient, and systematic, rather than focusing on the code.
"Companies need to move from a model-centric approach to a data-centric approach.”
-Andrew Ng-
It's important to remember that most data comes from us, humans, or from any apps, sensors, or systems we create. This means that there is a high level of probability that the data is biased, incomplete, unreliable, and inconsistent, just like we humans can be sometimes. So, how we can tackle this interesting challenge?
3.1. With Great Data Comes Great Responsibility
Clean, accurate, and reliable data is super important for AI systems. It helps them give valuable insights and make smart decisions. That's where "Data Governance" comes in. Its goal is to keep the data high-quality, secure, and easy to access.
It's essential to remember that when we're working with data, we need to handle it responsibly, if the data is not handled well it can have real-life consequences for many people. This is not a hypothetical scenario, it's happened before, for example, there was this Amazon HR tool that ended up showing discrimination against women.
Responsible AI is a big deal, and major tech companies like Google, Meta, IBM, and Microsoft have come up with their own philosophy for creating AI systems. In 2023, the AI Alliance was introduced with the goal of improving AI capabilities, safety, security, and trust, all while making sure the benefits are responsibly maximized for people and society everywhere.
4. How Machine Learning Works?
It's really important for you as a business professional to know how Machine Learning works. This knowledge can help you identify opportunities to stay competitive, make smarter decisions, and connect better with both technical and non-technical teams. So, let me break down how machine learning works in a simple and friendly way.
You've probably come across the words data, model, and training often, and that's because they are very important in Machine Learning.
In a Machine Learning system, we create a model (sometimes more than one). Think of the model as a little box or function that takes in an input (usually called X) and gives you an output (usually called ŷ). The input can be all sorts of things like text, images, audio, code, or tables of data, and the output is an inference, for example, it could be a number or a category.
So, what's a model exactly? Well, Machine Learning uses math, statistics, and probability theory (among other tools) to identify patterns in data. As you might have guessed, a Machine Learning model is actually a mathematical function. For example, there's this one called Linear Regression, which can help predict future sales or even estimate how much a house is worth.
$$y = β0 + β1x1 +β2x2 + ⋯ + βnxn + ϵ$$
A Machine Learning model is trained. But what does that really mean? Well, it's all about adjusting the mathematical function parameters with a subset of the data (known as a training set). This data needs to be analyzed, cleaned, and prepared beforehand. Once the model is trained, it'll be able to give you the results you want for most inputs, even those that weren't part of the training data.
In summary, a machine learning model is a math equation. During the training phase, we fine-tune its parameters using a carefully prepared set of data. Once that's done, the model can give you the answers you're looking for, even when you feed it new and different information.
I hope the explanation helps you get a better understanding of some core concepts making it all a bit less mysterious. So, next time you hear something like "Rumors say GPT-4 has 1.76 trillion parameters", you'll just know that it's a super huge math equation tweaked with tons of data.
4.1. Unimodal And Multimodal Systems
With all the amazing progress made in 2023, the terms "Unimodal" and "Multimodal" are buzzing around a lot. So, let me give you a quick explanation of what they mean.
A Unimodal AI system works with just one type of data. It could be text, images, audio, or even structured data like spreadsheets. It's focusing on a single data input at a time. For example, imagine a chatbot that only deals with text input to create text responses, that's an unimodal system, totally dedicated to text data.
A Multimodal AI system can handle and understand data from different types all at once. Just think of a virtual assistant that can search or analyze an image based on what you write to it. Sometimes real-world data can be very complex and needs a mix of visual, text, and audio to understand what's going on.
If you're a business professional, it's useful to know the differences between unimodal and multimodal AI systems, mostly if you are thinking about adding AI to your operations, products, or services. If this topic piques your interest, you may want to check Google's cool blog post about interacting with Gemini, Meta's handy usage tips, and OpenAI's breakdown of their famous GPT-4 model.
5. When AI And ML Should Be Used?
Even though you're familiar with AI and ML, it doesn't mean they're the perfect fit for every situation. AI and ML can be very expensive, so it's important to choose the right tool for each use case. Think about the huge amount of data needed, where will you store it? and, the computing power required to train the model, not just once, but several times until it's performing like a champ, do you already have it? Don't forget about tools for monitoring, continuous integration, and training too. Who's going to build all this? Does your organization have the know-how, or will you need to put together a separate team or get in touch with contractors? How many people will it take? As you can see, there's a lot to think about when deciding to use AI and ML. So, let's start by listing some important considerations.
What's the goal?: Before diving into data and algorithms, it's super important to know what you want to achieve with your ML/AI system. By understanding the problem you're tackling and the results you're aiming for, you can make sure your ML solution is both relevant and valuable.
Is your use case a good fit?: Machine learning is fantastic at certain tasks like classification, prediction, and pattern recognition. If your project is all about automating decisions or labeling things on a large scale, and traditional methods take too much time, then ML might be the perfect solution.
What kind of data do you have?: ML algorithms learn from data, so it's super important to have access to relevant, high-quality data. If you don't have any data related to your business problem, or if it's just too hard to collect, then ML might not be the best choice for now.
How many tasks?: ML really shines when you need to automate loads of decisions or tasks. If the issue you're dealing with is a one-time thing or doesn't need decisions made on a large scale, other methods might be a better fit.
What about the social impact?: Think about how it might affect everyone involved, including any indirect effects on society and particular groups of people. Make sure the application is ethically responsible and helpful.
Consider your existing resources: Take a moment to consider if you have the necessary data, computing resources, and know-how to start an ML project. If you find that you don't have the right data, enough computing power, or a skilled team, it might be best to hold off on the project for now.
What's in it for your business?: Think about whether the ML application will bring real benefits to your business or users. It's super important to make sure the solution tackles a real need and that putting your resources into ML technology will lead to a positive outcome.
It is not perfect: Keep in mind that ML systems aren't perfect, and it's important to be prepared to handle and reduce errors. Understanding the trade-offs and setting realistic expectations about the system's accuracy and reliability are key to success.
Cassie Kozyrkov shares some fantastic insights on this topic in her "Making Friends With Machine Learning" course. If you're interested, I highly recommend checking it out.
I hope the table below helps you get an even better idea of how to identify the perfect opportunities to use Machine Learning in your projects.
As you can see in the illustration, if you come across a problem that's low in frequency and complexity, there's a good chance you can solve it yourself. When dealing with high-frequency but low-complexity problems, a rule-based engine can be your best friend, think of daily or weekly reports, can they be automated? For low-frequency, high-complexity problems, Statistical Modeling can come to the rescue, like when you're trying to create a sales agent incentive program. Last but not least, if you're tackling high-frequency, high-complexity issues, AI and ML are here to help, for example, a customer service chatbot.
6. The Machine Learning Workflow
So far we've gone through some key AI terms, discovered why data is so important, learned how machine learning works, and looked at how to identify different machine learning use cases. Now you might be wondering, do AI and ML systems follow a specific workflow when it comes to projects? Great question! Let's dive into some of the workflows used by leading tech companies to find it out.
The following is the recommended lifecycle from Microsoft for organizing data science projects. If you're interested in learning more, check out The Team Data Science Process lifecycle for a deeper understanding.
Another approach comes from Quantumblack AI by McKinsey. Their steps are quite similar to Microsoft's.
And last but not least, let's take a look at IBM's approach to data science, called the Data Science Methodology.
It's interesting to see how different companies like Quantumblack AI by McKinsey, Microsoft, and IBM have their unique approaches. Despite their differences, they share some key steps. Let's check out these common elements together.
Business Understanding/Analytic Approach: In this first phase, we figure out the objectives and requirements from a business point of view. It's all about getting to know the problem we need to solve and understanding the potential impact of a solution.
Data Acquisition & Understanding/Data Collection: In this phase, we gather all the data we need. We'll find the right data sources and make sure the data we collect is related to the business issue we're trying to solve.
Data Preparation/Wrangling, Exploration & Cleaning: After we've gathered all the data, it's time to get it ready. In this step, we'll clean up the data, take care of any missing values, encode categorical variables, normalize the data, and maybe even transform some variables. It's also super important to explore the data using visualizations and statistics to get a good idea of the distributions and relationships within the data.
Feature Engineering: In this stage, we're going to get creative and make new features or variables from the existing data. This will help our predictive models understand the problem even better. Also, we'll do some feature selection to remove any irrelevant or redundant variables.
Modeling: In this phase, we'll try different algorithms and techniques to build predictive models.
Model Evaluation: Once we've trained a model, it's super important to check how well it's doing by using the right measurements.
Deployment: Once we've picked the perfect model, we'll set it up in a production environment where it can start making predictions.
Customer Acceptance/Feedback: After we've deployed the model, it's time to see how it performs in real-life situations. We'll gather feedback from users and stakeholders to find out if the model is really hitting the mark when it comes to business goals.
This is an iterative process, models often require some adjustments over time, which is completely normal. As we gather feedback from users and the business, we will continue refining and improving the model to ensure it truly meets everyone's needs.
Conclusion
In this post, you've discovered some key AI concepts, looked at the uses and limits of AI, and read about the important part data plays in the AI world. You also got a feel for how machine learning works and how to spot use cases for it, and lastly, the steps you'll find in an AI project workflow. As we keep improving and growing these technologies, they have an amazing potential to change businesses and society. Now that you've learned all this cool stuff, you're ready to ride the wave of innovation.
I feel so lucky to have played a part in this learning journey, and I'm excited to contribute even more in the future! Let's take this AI Leap together!
