trailhead

Discover the Science Behind AI

Learning Objectives

After completing this unit, you’ll be able to:
  • Describe what machine learning is.
  • Articulate the different flavors of machine learning.
  • Understand how to bring AI into your business.

Getting to Know the Science Behind AI

There are a lot of technical concepts that go into truly understanding AI, but the most important thing to remember is that AI is about building intelligent computer programs that carry out tasks like:
  • Visual perception
  • Speech recognition
  • Sentiment analysis
For example, self-driving cars require computers to have visual perception capabilities, and voice-activated applications like Siri require computers to have speech recognition capabilities.

Photograph of a person speaking to a smart watch. The speech bubble says, "Hey Siri, start my coffee machine!"

The brains behind AI include modeling techniques such as machine learning, deep learning, and natural language processing. Let’s start with machine learning.

Machine Learning for Beginners

Machine learning is the core driver of AI. It’s the process of using algorithms to tell you something interesting about your data without writing code specific to the problem you’re trying to solve. Said differently, it’s a way to have computers learn from data with minimal programming. Instead of writing code, you feed a machine data and it builds its own logical function based on this data.

So how does this work in practice? Well, it all starts with “training data.” That’s a set of data you give to your data model to help it learn. The more data you feed your model, the stronger it gets.

When you feed training data to your machine learning model, that data is defined by a set of attributes and characteristics. It’s up to the model to determine how to make sense of all these attributes.

So how does the model decide what attributes are most important to build the best model? The algorithm “weights” the different features of your model to determine the best set of attributes, that when combined into an equation, solve a specific problem.

Graph representing an algorithm with three function inputs, weighted at 0.4, 0.3, and 0.6, that result in the function output.

Finally, ever heard the phrase, “Garbage in, garbage out”? That’s especially important here, when you’re preparing your training data set. The better quality your data, the better your model.

Deep Learning and Natural Language Processing

We just learned that typical applications of machine learning involve selecting the best features to produce the most optimal model. We also learned that an algorithm can only be as good as the data that goes into training it.

For these very reasons, the performance of machine learning algorithms can weaken when key information is buried in unstructured data. That’s where deep learning comes into play.

Deep Learning

Deep learning is great at automatically learning the best features from somewhat noisy data (read: unstructured) so that algorithms can learn more effectively. It also:
  • Uses complex algorithms to perform tasks in domains where it actually learns the domain with little or no human supervision.
  • Learns how to learn. For example, consumer apps like Google use deep learning to power facial recognition in photos.

Natural Language Processing (NLP)

Natural language processing (NLP) is a form of machine learning that recognizes language and its many usage and grammar rules by finding patterns within large data sets. It also:
  • Can perform sentiment analysis, where algorithms look for patterns in social media postings to understand how customers feel about a specific brand or product.
  • Handles speech recognition, providing a text summary derived from “listening” to an audio clip of a human speaking.
  • Conducts question answering, typically handling those questions with a specific answer (for example, What is the square root of 4?), but also exploring how to handle more complex and open-ended questions.

There are other types of machine learning, but we wanted to cover the two most prominent ones here. If you want to geek out on this topic further, read this or watch this.

Illustration of a network of interrelated spheres. Each of the three input layer spheres is connected to each of the four hidden layer spheres, which are in turn connected to each of the two output layer spheres.

How to Integrate AI into Your Business

Now that you know more about the science behind AI, let’s talk about different ways to integrate AI into your business.

The first step is understanding where it makes sense to initially deploy AI. Organizations have tons of data, so establishing quick wins and building trust within your organization with a focused approach is key.

For example, let’s take email metrics as a possible predictor of customer behavior. You can start with machine learning to answer questions like:
  • What’s the likelihood to open an email based on its subject line?
  • What’s the likelihood to act on special offers once an email is opened?
Once you’ve determined your use case and justified the business value, it’s time to think about different approaches to integrating AI within your business.

Point Solutions

One option is to partner closely with companies that offer point solutions to specific machine learning problems across various domains. This option is great if you want to test the waters and start small before carrying out any large-scale implementation. However, keep in mind that you’ll end up paying another vendor to execute your AI strategy. AI could become an integral part of your business strategy over time, so consider finding a more cost-effective approach that doesn’t involve exporting your data to another platform.

Do-it-yourself

If you want to keep things in-house, you could hire your own internal team of AI experts. This approach is great if AI is a long-term part of your business strategy; however, it takes time to hire great talent in a fiercely competitive market. It’s also expensive to not only hire but also retain premium AI talent.

Hybrid Approach

You could adopt a hybrid approach where you partner with companies that offer point solutions and hire AI experts over time. This is a great approach if you want to start testing AI within your business for a few targeted use cases while building out teams of experts that can replicate results in other areas of your business.

The Ultimate Goal

The ultimate goal is to seamlessly embed AI into your existing business processes and applications. This is the path of least resistance, because you’re using the same technology that you already use every day, just with smarter features that adjust to your preferences over time. Your users can experience the benefits of AI without having to adopt new technology.

retargeting