It's when machines are able to learn from experience, adapt to new inputs and complete tasks that require intelligence.
What humans would consider simple tasks like making a coffee, understanding figures of speech or pointing to the one dog in a photo full of cats are extremely difficult for computers to achieve. In the same way, what's easy for computers is often hard for humans. Think linear regressions or calculating the exact time it takes to drive from Paris to Rome. This means the most powerful intelligence we’re able to achieve today is that of artificial and human intelligence working in tandem.
Artificial Intelligence is a set of computer science techniques that allows for computer software to learn from experience, adapt to new inputs and complete tasks that resemble human intelligence. The most efficient and popular AI technique today is called Deep Learning. It works using AI's ability to make sense of patterns in data.
On their journey to replicate human intelligence, there are many independent branches of human performance – perception, language processing, planning, spacial motion, etc. – that machines must learn. While each of these branches may one day meet to create something resembling a human, most AI researchers are unconcerned with that result, dedicating their work to growing a single branch and nothing else. In this way, Artificial Intelligence should be thought of in two different categories: Artificial Narrow Intelligence (ANI) and Artificial General Intelligence (AGI).
One of these types of AI is already working and integrated into many different applications, and the other, though very exciting, is still purely theoretical.
ANI - completing specific tasks by seeing patterns in data
Artificial Narrow Intelligence (ANI) is the only form of Artificial Intelligence that humanity has achieved so far. This includes everything from email spam filters to cancer tumor prediction models. AI has proven to be good at performing tasks like playing chess, making purchase suggestions, weather forecasting, almost any specific task for which it has access to plenty of data. Computer vision used in self-driving cars and natural language processing used for real-time translating, though more impressive than ever, are still simply advancing areas of narrow AI.
In essence, narrow AI works within very limited contexts, and though our phones may be packed with many pieces of AI, they’re all operating separately and can’t take on tasks beyond what they’re programmed to do. You can’t expect the same code engine that plays your music to, say, order your pizza. No matter how sophisticated it may be, it’s not truly thinking independently.
AGI – the moonshot for the thinking machine
The goal to unite the many branches of ANI performance into one “thinking” machine is the quest for what’s called Artificial General Intelligence (AGI), and as of now, is still purely theoretical.
This – the singularity – is imagined as a point of no return where exponentially accelerating technology advances beyond human capacities to comprehend or control its outcomes. This could provide the world with unforeseen technologies and, give humans access to immortality as biological and mechanical minds become one. If this sounds a bit fantastical, it’s because it still is, with many years of research and development still required to know if anything close to this scenario could become a reality.