The rise of artificial intelligence has brought with it a surge in the popularity of bots and a flood of automated solutions to complex tasks.
One of the main drivers of the surge in bots is the rise in their use in healthcare, where they have become increasingly sophisticated, capable of reading patient medical records, and often perform tasks that are otherwise beyond the capabilities of human experts.
Bots have even been deployed to analyse the behaviour of human doctors.
The bots are now so sophisticated they can even perform tasks they couldn’t even imagine 10 years ago.
In fact, one of the major obstacles to the adoption of robots in healthcare has been their reliance on human doctors to complete the tasks they can’t do on their own.
But it’s not just humans who are at risk.
Bots are also being used in some of the more mundane tasks that most people would not be able to do, such as identifying the difference between two different colours, or performing mathematical equations.
The researchers behind a new study, published in the journal Computers in Human Behavior, said that even though they can be very powerful tools, they also have limitations.
“Our study shows that despite the fact that humans are the primary decision-makers in these systems, the bots may still have some limitations that need to be addressed in the future,” said Dr David Giesbrecht, from the University of California, Berkeley, who led the study.
“The study also provides a new approach for developing machine-learning software that is capable of dealing with the challenges of these tasks.”
The researchers created an artificial intelligence model to help them understand the limitations of their AI system, and they developed an approach to automate some of these challenges.
They trained their AI model on a sample of tasks that humans typically would perform in a real healthcare setting.
The task was to identify which of two colours on a colour-coded sheet of paper was different to which one was the same.
The model was trained on this sample of data, and the results showed that it was very difficult for it to identify the two different coloured sheets of paper.
So it had to rely on human judgement, as opposed to a machine, which is what our approach is based on.
The study showed that a machine could only identify the difference of two different colour sheets of coloured paper.
The next step was to train it on a more complex task.
This time, the task was simple: to name the colour of a single letter in a text document.
The AI model had to correctly identify the letters, but could only do so by learning the spelling and reading each letter individually, and by taking into account the spelling of each letter.
However, the researchers found that even when trained to recognise letters that were different to each other, it could only correctly identify two different letters, so the task had to be re-trained.
The final step was for the AI to predict the colour to be used by the text document, and it would then read out the letters it identified, and compare them with the actual letter.
The process took just 10 minutes.
“A lot of our work has been about creating a machine that is able to handle some of those tasks, but still has the capacity to be able make sense of complex text documents,” Dr Giesbrocht said.
“We’re still in the early stages of learning how to make that machine do that, and this is the first step towards doing that.”
This work has found its way into a number of other projects, including a robot designed to help doctors perform a surgery, and a robot that can recognise patients in a hospital.
One possible way that AI might help doctors is by providing a machine with information about the patient, which could be relevant to their diagnosis, and provide guidance for their treatment.
“These systems are still relatively new in the world, but they’re definitely making a big impact,” Dr Jochen Meyer, from MIT Media Lab, said.
The research was funded by the US National Science Foundation, the University the University at Buffalo, and Microsoft Research.
The paper will be published in Computers, Human Behaviour, next week.
Follow @BBCNewsEnts for more stories from Australia
