Artificial Intelligence, or AI, is an umbrella term. It's actually a bit of a vague phrase that we all use as a shorthand.
When we use the term “AI”, what we mean is the technology behind it, which is essentially algorithms. Algorithms are advanced or complex mathematical formulas that have been designed to perform or complete a task step by step - really simple, small steps. You need multiple algorithms in order to complete a task and, as a result, you need to network them together. This is why sometimes we hear the phrase “AI network” or “AI model”; it's actually a complex network or connection of different mathematical formulas put together.
Today, we have generations and generations of algorithm advancing every day. As it gets more complex, we add new names. The latest generation of AI models are called “neural networks” because they are designed in a way that imitates the way our brain works. They are also capable of complex tasks that we call “deep learning”; they’re learning a lot of complex patterns and data points from materials that we put in front of them These are also called “training data sets” or “training data”. The AI model can make connections and detect patterns that are brains cannot. So, something that would be invisible to us is visible to the machine because it is equipped with those complex algorithms that we, as humans, have designed and we've powered that with computer technology, as well. For example,
it can analyse big source data - a big corpus of information -and give us patterns within that budget of work. This is what we do when we give it the weather forecast or weather data from yesterday or from the last 10 years and we ask the algorithm to forecast what's going to happen in the next few days.
We can also ask AI technologies to create content, because we're asking the algorithm to train on previous work, images, sound, performances, etc. and then generate new ones, or what we call “synthetic performances”. They’re synthetic performances because they are not technically copies in a way we understand them when we use a photocopier or when we're remixing sound with digital files and we're making a mixtape.
Using sound as an example, AI can learn from recordings what makes a sound, a piece of music or a voice sound the way it does, what constitutes a particular accent, what qualities we find appealing. Having learned that, it is then able to reproduce it by generating new data. So it's completely synthetic, it's not a copy-paste exercise and this really what is new with AI.
This process of learning is also known as data mining. You may hear that phrase every now and then in the media or even on this podcast. So here are the key phrases: AI models/technologies/systems refer to the actual algorithms that are churning the data.
Input data is the material that the algorithms train on in order to imitate or generate something similar.
Ouput data, is the result. It is sometimes called synthetic because it's been generated by an AI, not through an exercise of copy and paste or collage or remixing, but through genuine synthesis generated from scratch by the algorithm, itself.
Importantly, AI models today cannot reflect. They do not know if what they produce is correct, is good, is what we wanted.
It only follows, like a robot incapable of self-reflection, what is being taught to do with the information it's been given. And that's really important to keep in mind because at the moment, AI is not reflective and it is not sentient in the way that you sometimes see in science fiction content about AI. And that brings us on to Synthesis.
Synthetic, synthesis or synthesization is the process of generating content - often sound, images or moving images like film -using AI technology. It is not dissimilar to scanning someone or something, or to make what we call computer-generated content or imaging, (CGI for, example), but the difference is that it's using AI technology.