Robots are brilliant at following instructions. They’re so good at it, in fact, that it seems most of us will eventually find ourselves replaced by automation although that might not necessarily be a bad thing. For example, if you told a robot to go and put the kettle on, it would do exactly that. What it might fail to do, though, is make you a nice cup of tea. Most robots are programmed to complete tasks, but aren’t so great at understanding context. With context, a simple command can have an entirely different meaning. An inability to apply context is a severe barrier to the integration of robots into our everyday lives. However, a team of researchers at MIT Computer Science and Artificial Intelligence Laboratory may have found the answer.
Context is everything. . .
This summer, a team at MIT’s Computer Science and Artificial Intelligence Laboratory released a paper detailing the development of ‘ComText’, an AI systems which, hence the name, puts ‘commands in context’. The project’s aim is to address the reasoning gap present in most robots. Rohan Paul, leader of the MIT team, explains this using the example of a futuristic tea party. He asks the robot to pass him his cup, and because the robot has prior knowledge about the ownership of that cup, it knows which one to select. This past knowledge, therefore, can impact the way that robots respond to both present and future commands. So, going back to the original example of boiling the kettle, a robot would be able to remember that the last time few times it ‘put the kettle on’ for its owner, it was then instructed to make a soy latte with vanilla syrup. Therefore, with context, putting the kettle on changes in meaning. Apply this to any setting that involves human to bot interaction, and the merit becomes clear. But how will this impact the development of robotics, and which industries will it affect?
How will contextual understanding disrupt robotics?
Equip robots with the ability to comprehend context, and the dialogue between machines and humans will become increasingly fluid. At the moment, robots are designed to complete carefully worded commands, which limits their usability in real world, interactive scenarios. If bots are less likely to get things wrong or misunderstand colloquial commands, humans are less likely to get frustrated with them. This is an important enabler for the wider acceptance and adoption of robotics. In the workplace, for instance, cobots would be able to carry out tasks alongside human employees rather than simply taking orders. It also represents a huge step towards the implementation of robots in the care and hospitality industries, allowing a bot to apply contextual knowledge to the requests of their owners or customers. According to Shane Wall, Chief Technology Officer at HP, robot intelligence will exceed human intelligence by 2029. At this point, he believes that we will begin to experience the technological singularity. AI systems can already ‘remember’ things in the sense that they use existing data to inform insights, but an AI that can compile masses of human data might not come to the best conclusion about the integrity of our race. That’s where the singularity becomes really problematic, and avoiding this means that robots must either really like us, or not know too much about us. If Shane Wall and a host of other prominent technologists are right about the imminence of superior robot intelligence, then this is something that developers need to decide on today. On the flip side, giving robots this ability could also help to avoid malpractices when they become commonplace. If robots can understand that they are being instructed to carry out a potentially harmful task, they can be programmed to refuse to do it.
Using the past to enhance an understanding of the future is something that, in light of years of international conflict, humans aren’t very good at. One day, perhaps AI enabled robots could be better at this than us. Either way, by 2029, developers want to be as certain as possible that these clever machines like – or at least sympathise – with humanity. Now is the time to ask if we really want robots to remember things. Although it increases their usefulness, allowing bots to recall any type of information gives them more power over us. Is it worth the risk?
Does your business ‘employ’ advanced robots? Will robot intelligence surpass human intelligence by 2029? Which industries will benefit most from contextually aware robots? Share your thoughts and opinions.