AI is here, it is changing the way we work and live, and the rate of progress is only going to speed up.

Artificial Intelligence exploded into public consciousness after the release of ChatGPT in November 2022. But AI has been part of our daily lives for several years now. From the recommendation engines in your Netflix and Amazon accounts, to the maps tools that can optimize for the fastest or shortest route to your destination, to the virtual assistants used by customer service departments, AI is everywhere.

Let’s talk about what AI is, where it’s been, and where it’s going.

Definitions

Artificial Intelligence is a computer system that can perform tasks that normally require human intelligence. That’s a large umbrella, and so it’s helpful to talk about some of the main sub-categories within the field of AI.

1. Good Old Fashioned Artificial Intelligence (GOFAI)
   Agents that mimic human intelligence. These can be rules-based or not. This category includes everything from robots that can navigate mazes, to computers that can play Chess, programs that play Jeopardy!, and even programs that debate humans.
   
2. Machine Learning (ML)
   A subset of AI that uses statistical methods to find patterns in training data and make predictions. Rather than explicitly coding the program line-by-line, ML programs are instructions for how to learn.
   
3. Deep Learning (DL)
   A subset of ML models with a neural network (NN) architecture. NNs are layers of weights and biases capable of learning complex relationships in data.

Most of the advances we’re talking about in 2023 fall in the Machine Learning bucket – let’s dive in!

How Does a Computer Learn?

There are several methods for training a ML model:

1. Supervised learning
   Using labeled training data with a known number of possible categories. For example, a hospital emergency room might use a supervised learning algorithm to predict whether a patient is at risk for a heart attack. Imagine a large table with rows of anonymized patient data. Each column lists a vital sign, and a final column is labeled either “heart attack” or “no heart attack.”
2. Unsupervised learning
   Processing unlabeled training data with the goal of finding patterns in the data. There are an unknown number of categories – the algorithm identifies them through clustering or association. For example, a grocery chain might use an unsupervised learning algorithm to identify purchasing patterns in their sales data.
3. Reinforcement learning
   The system produces its own training data by interacting with its environment. An agent is instructed to optimize a total score. I like to use Pac-Man to help visualize this example. You (the agent) explore all the positions on the map (states), eat as many dots as you can (maximize reward), and avoid the ghosts (minimize punishment). As the one learning how to play, your strategy is to make short-term sacrifices to learn what works and make better decisions in the future.

In all three cases, the ‘learning’ taking place is simply pattern recognition. The underlying mathematics have been around for decades. What’s currently driving AI adoption and development is the availability, affordability, and scalability of computing power today.

What can ML do?

Here are three applications of Machine Learning that have been incredibly successful:

Natural Language Processing (NLP) is the technology behind tools that can understand and generate written language. The most common examples are chatbots like IBM Watson, ChatGPT, Google Bard.

Computer Vision (CV) is the technology behind image recognition. It can be used to classify images and even extract text. For example, processing images of industrial equipment to identify potential mechanical failures, or digitizing the text written in doctors’ notes.

Optimization is the technology behind search, recommendation algorithms, navigation systems, and many business applications. Anytime you’re trying to maximize or minimize something, you’re solving an optimization problem. For example, maximizing profit and minimizing cost. You can also optimize manufacturing processes, resource scheduling, customer loyalty programs, and even missile defence scenarios.

What does the future of AI look like?

Short term, you can expect to see conversational AI-augmentation pop up everywhere. Every application you use at work will include a generative language tool that can produce your first draft. It may be your project plan, your speech, slideshow, or spreadsheet, your marketing campaign, your website copy, your machine learning models, or your product design.

Eventually, AI will likely do everything that humans do – including developing more powerful iterations of artificial intelligence! Once AI can produce better AI, a time often referred to as The Singularity, the technology will surpass human intelligence and we will never catch up. That’s why many smart people are working on AI alignment: the challenge of ensuring AI aligns with human values.

As we launch into the AI-augmented future, it’s important for everyone to be a part of the discussion. Individuals and organizations must learn about the opportunities and risks. That’s the only way to have informed discussions about how we apply AI ethically, safely, and responsibly.