Artificial Intelligence is something that’s been on the horizon for a long time – probably as long as anyone reading this will be able to remember.
Since its emergence into the public consciousness through science fiction, we have understood with certainty that one day machines will have “intelligence”, and considered the ethical and philosophical implications.
Today, though, there’s a palpable sense that we’re getting close. AI is one of the biggest buzzwords in tech, and fundamental to most predictions how IT will enable change and growth in coming years. Everything is now “smart”, we are consistently told – and doom mongers (including some very smart people!) are predicting that we are rushing headlong towards calamity.
But is this all marketing hype? Are we really any closer to “intelligent” machines than we were 20 years ago, when many of the ideas driving AI today – machine learning and deep learning, for example – already existed – but without the internet, there simply wasn’t the data to exploit them fully?
Well, to answer that question we have to first ask ourselves what this “intelligence”, that we are trying to simulate artificially, actually is.
“Intelligence” is notoriously difficult to define or quantify. Clearly, from the type of tasks that we’re looking forward to seeing AIs do, and the way we are expecting to interact with them as they do them, today’s AI is, generally speaking, in search of human-like intelligence.
Some of the most exciting recent work in AI, such as the development of deep neural networks, is intrinsically based around creating AIs which mimic the function of human, organic brains. But human-like intelligence comes in many forms. We’re all aware of people who appear to be very smart in some ways, but less so in others. Clichéd examples are the bookish professor with poor social skills and limited ‘common sense’, or the charming, persuasive and successful business tycoon with limited academic knowledge or ability. AIs also vary wildly in the form of intelligence they emulate.
IQ tests were formulated as a method of quantifying intelligence, although their validity in this regard is often disputed. Machines can do IQ tests – with around the same ability level as a four-year-old. But there are many other factors which go into “human-like” intelligence that IQ tests do not even touch on.
Emotional intelligence registers how well someone is able to understand and interact with people on an emotional level, or interpret their own feelings. This is sometimes through of as empathy as something which is intuitive but is undoubtedly a mental process, dependent on our brain’s ability to analyze information and infer an insight or solution, so qualifies as “intelligence”. This aspect of our intelligence is thought to be integral to our creative abilities – something else machines will have to master if they are going to develop “human-like” intelligence. Steps are certainly being taken in this direction. IBM’s Watson Music can create music designed to mimic the feelings and emotions conveyed by a different piece of music. AIs have also written poetry and novels.
Athletes and sportspeople rely on a spatial awareness of what is going on around them as well as needing their brain to work quickly and accurately to react to complicated, changing circumstances. Again these are processes carried out in the brain, even though the qualities we most often associate with athletes are physical. AIs have been taught to play old video games using only visual input, showing that they are capable of “learning” how to react to movement and even developing a desire to win.
Our communicative ability – how well we are able to express our ideas and pass the valuable information we interpret from the world around us is another factor of intelligence. Again, machines have made ground here, with recent developments in the AI-related fields of Natural Language Processing and Natural Language Generation, which are bringing them closer to being able to communicate with us in a human-like way.
An AI would have to be able to demonstrate all of these abilities, and probably many more, before it approached what we would consider to be a human-like level of intelligence. Now, undoubtedly, the building blocks are falling into place for this to become a reality, but is an artificial human brain, capable of working at super-speed and with unlimited memory and perfect recollection, what we want or need?
The question has ethical implications, particularly if we bring the controversial topic of consciousness into the equation. From a scientific viewpoint, consciousness is a state that arises when a biological brain interprets the flood of sensory input streaming in from the world around it, leading, somehow, to the conclusion that it exists as an entity.
It’s not well understood at all – but most of us can conceive how this massive flood of images and sounds is interpreted through a biological neuro-network which leads to “thoughts” – and among those thoughts are concepts of individual existence such as “I am a human”, “I exist” and “I am experiencing thoughts”.
So, it’s only a small step of logic to assume that machines will one day – perhaps soon, given how broad the stream of data they are capable of ingesting and processing is becoming – in some way experience this phenomena, too. How long before a machine is capable of saying to us “I too, am experiencing a sense of existence and individuality”? And, when it does, will we have any sound intellectual ground from which to argue that it is isn’t? After all, science has yet to put forward any evidence against the idea that we are entirely mechanistic constructs ourselves. Our brains run on electricity and rely on energy to fuel them.
Oh, and because I realize that religious people might feel alienated by the ideas I’m bringing up here – here’s one for you guys. If God exists, and created us, doesn’t that mean that we’re really nothing more than AIs ourselves? Giving us even less ground to argue with a human-created AI that it isn’t sentient itself?
Today when we talk about consciousness and the possibility that machines will develop sentience, it feels like we are wandering into fringe territories. But it’s potentially a problem that will become very real for us at some time in the future, if we remain in pursuit of ever-more human-like intelligence.
Perhaps fortunately, when we talk about AI today, most of them type we are talking about very specific applications focused on solving a particular problem. It’s doubtful, for example, that we are going to find ourselves in an argument any time soon with the AI that manages energy regulation in Google’s datacenters about whether it is conscious or not. But that could just be because we haven’t given it a mouth to speak with yet, or the sensors it needs to make that deduction. When we do, we might have to prepare for a shock!