Is AI Ready to Compete with the Human Workforce?
Part 1: The complexity hurdle
The world we live in is complex, on the edge of chaotic even. Our natural intelligence has adapted to this complexity surprisingly well. Our brains use highly effective heuristics to deal with the overload of signals coming in. They do this by ignoring most of what is going on and only responding to signals that need an immediate response.
Our natural intelligence is also highly versatile. We are capable of dealing with a huge variety of tasks and circumstances. We can even deal with completely artificial tasks no natural instinct could ever prepare us for. We can drive a car through congested traffic. Or sit in a concert hall listening to other people making music together. Our brains are adaptable and experience continuously changes our strategies, decisions and behaviours.
One of our brains most impressive accomplishments is how well it manages to navigate this complex, unpredictable and often lethal world of ours. Our brain’s ability to be creative and think outside the box is especially impressive. It can make sense of completely new situations it never encountered before. It can imagine possibilities and then make them real. Our brains do not just survive in a complex reality, they actually create whole new realities in the process.
Recent advances in AI have shown that it can outperform humans in specific tasks. But there is no sign as yet that AI is reaching a human level of versatility and adaptability. Application of AI seen so far are highly specialised. They are developed for a narrow domain – with a limited and often controlled range of stimuli and responses. They are unable to respond to variations and inputs falling outside their pre-determined domain.
A self-driving car may be able to solve a range of traffic-related problems but it cannot recognise its passenger is suffering from a life-threatening medical condition such as an injury or a heart attack. Even an untrained human driver would notice when their passenger became unwell and respond to that unexpected event. An autonomous factory may be able to deal with many variables but it would not recognise a deliberate act of sabotage or a natural disaster unless it has been specifically designed to do so.
What makes the human brain so adaptable is not just how it learns and adapt to known and predictable situations but how it makes sense of and responds to the unpredictable ones. We recognise when things deviate from what we expected. We notice when our learned behaviours do not match the situation we are actually in. We can mentally jump out of the domain we are currently in and change our thinking and behaviours to match whatever new domain we are in. We would not, for instance, continue cooking dinner when we notice the kitchen is on fire. Instead, we would move into fire-danger mode and either try to fight the fire, rescue the other people or flee the building. The current generation of AI would still be trying to finish the meal as the house burns down around it.
I do not claim that for AI to succeed we need to understand how natural intelligence is produced by our brains, or try to emulate our brains in every detail. There may be many other ways in which intelligence can be produced. But the sheer complexity of our brains does show that intelligence is much more complex than most models account for. Until we have models for AI that are at least as complex as the brain, AI will be be limited in how it can function in the complex reality we are exposing it to.
Within limited, reasonably predictable and well-understood domains AI is already demonstrating its value. It will continue to expand its reach and usefulness. But it will take a long time before fully autonomous AI can operate unsupervised in our complex human world. Self-driving cars may be more reliable than human drivers – until something unexpected happens. We need a human brain to recognise when things are no longer falling within the known parameters. Then we need a human to take over, assess the new situation and make some sensible decisions. Until then AI can complement, even augment, many human activities. Competing with humans in the workplace, however, is a whole different ball game.