With so many potential use-cases, Artificial Intelligence’s value may still seem somewhat illusory. Its current and future abilities are vast and wide, so determining how it can actually help you and your business can be difficult.
What is AI, anyway? When we refer to AI today, we are actually referring to “weak” or “narrow” AI. “Narrow AI” refers to a computational system that replicates the way our brains work, but focuses on one particular task (e.g. analyzing then organizing images based on some given criteria). While this name might imply AI isn’t anything remarkable, that would be woefully underestimating just how powerful the brain is.
Yet AI isn’t here to replace humans. We’re still a far way from Artificial General Intelligence, i.e. computational systems that replicate the workings of the human brain to do any intellectual activity humans can do, becoming actualized. There are however fields of AI already impacting businesses. With the immense amount of content being created and shared online, managing, organizing, moderating and understanding that content has become pivotal for almost every industry. The field of computer vision has therefore become vital as it allows computers to “see,” which in turn lets companies process enormous amounts of visual content with far less human effort and time required.
As you determine where AI will best fit into your business, we’ve created The 2023 Ultimate Artificial Intelligence Glossary provides non-technical professionals who aren’t yet familiar with the ins and outs of AI with a comprehensive referral resource as they learn more about this emerging technology.
In this post, I’ve selected and outlined 5 of these key AI definitions to ground you as you explore the wondrous, seemingly limitless world of Artificial Intelligence.
1. Artificial Neural Network (ANN)
If someone were to show you a picture of a dog, so long as you have previously learnt what dogs look like, you’d be able to recognize it with ease. This seemingly simple process is actually the end result of billions of neurons receiving, processing and transmitting that information within our brain. Artificial Neural Networks mimic this brain process. They consist of a network of interconnected, layered processing elements that work together to power artificial intelligence, like computer vision.
2. Machine Learning
Another subset of AI that relies on ANNs to work is Machine Learning. These are the algorithms that learn patterns from the existing data and use these patterns to make predictions or decisions with new data. Consider that same picture of the dog. Without machine learning algorithms, computers would need to be told a dog is in the image every single time it received it as an input. With machine learning algorithms, computers can not only see what is in the image (“computer vision”), but learn the concept of a “dog”, allowing them to then recognize dogs in entirely different images that are inputted in the future.
3. Model
How does the computer process images to make predictions? Using models. These are processing blocks that take inputs, such as images or videos, and return predicted concepts. For models to work, they have to first be trained. What do you want the model to see? Here at Clarifai, we have several pre-built models including our General Model and more focused models that recognize concepts related to specific things like “Weddings” or “Travel”. Depending on which model you use, different concepts may be predicted, based on what the model was trained to see.
4. Custom Model
Models can also be customized to fit your specific needs. Custom models are small ANNs that take inputs particular to a user and return predicted concepts, based on what the user has trained the model to see in their inputs. Our CEO Matt Zeiler, for instance, once trained a custom model that could recognize his dog Roly from other dogs. Your custom model may include images or videos of your particular products.
5. Custom Training
Custom training is simply the process of teaching a model to make certain predictions. While models only need a few inputs to learn your concept, having sufficient training data is the key to making your model as accurate as possible. Much like the human brain, the more examples of images or videos with your desired concepts the model receives, the better the model will become at identifying these concepts.
