Explainer: What Do We Mean By 'Artificial Intelligence'?
The AI that has rapidly emerged in everyday technologies may be
smart, but how much control we give it should depend on its safety.
By Michael Rovatsos.
Artificial Intelligence (AI) is about getting machines to do things that we normally associate with human intelligence: problem solving, decision making, learning, communicating, perceiving and acting in the physical world. As a technical field, it emerged in the 1950s in parallel with the advent of digital computing machinery.
Much confusion is caused by failing to distinguish whether AI behaves as if it were intelligent, or whether it does so in a way that is similar to humans – for example, pocket calculators perform calculations humans are incapable of, but they also do it very differently from humans.
The AI we have seen emerge rapidly in everyday technologies over the last few years is very unlike human intelligence. Almost all systems are only good at a single thing. Most are trained on amounts of data no person could ever experience in their lifetime, and they only achieve human-level performance at simple one-step tasks like recognising objects or translating a piece of text. Current AI systems still don't perform very well in areas such as complex reasoning or creative thinking.
Many of today’s systems use a technique known as machine learning. This extracts patterns from observed data to make predictions or judgments in new problem-solving situations.
Imagine you were given a million lists of paint ingredients and the colour they produce when mixed. Neural networks – a currently very popular method which has been around since the 1950s – are a bit like lots of tubes and valves that would mix the paint. Given many examples of what goes in and what comes out, they learn how to make adjustments to produce the right mix.
Recent progress in AI – which has been impressive – has been largely fuelled by the availability of enormous amounts of data, and rapidly increasing processing power. Nonetheless, there are still only a few areas where we’re seeing AI being deployed in everyday life applications: data analytics, language, vision, and robotics.
The “intelligence” that powers them is truly “artificial” and not very human-like. In fact, it is often inscrutable to humans, and many people are concerned about the potential dangers this brings. However, the human race has more often produced problems for itself by creating “dumb” technologies, not smart ones. While there is a complex relationship between automation and AI, it is for us to determine how much control we want to give it – and this will depend on its safety, rather than its intelligence.
