Reasons why we need a human-in-the-loop in AI
AI doesn’t have all the answers – we’ll always need humans in on the design.
AI doesn’t have all the answers.
There is no shortage of doomsayers eager to warn us of the dangers of AI, and how the machines are going to take over leaving humans in their dust. But the people at the coalface are thinking about it differently. They say the future must keep people involved – it is not only better for us, but more efficient. Here are five reasons why.
1. There are limits to how smart AI can be
While machine learning makes it possible to process data on a scale that far eclipses human efforts, there are limits. That's where the 'human-in-the-loop' approach comes into play, as the next wave of automation will see tasks within jobs being split between humans and artificial intelligence (AI).
"The models, at least for many years, will not contain all the information, and therefore will need a human in most cases to make the final decisions and even the intermediate decisions guiding the system," said Professor Maria Garcia De La Banda, a researcher in data science and AI at Monash University's Faculty of Information Technology.
2. It democratises AI
For decades, knowledge of AI has been concentrated in the hands of a small group of data scientists. The human-in-the-loop approach is a way of making the benefits of that knowledge accessible to skilled workers in all kinds of occupations.
It's not limited to desk jobs, either. Tim Dwyer, professor of data visualisation and immersive analytics at Monash, is currently developing immersive display technologies such as augmented reality (AR) that allow people to work with AI systems wherever they are, be it a building site or an operating theatre.
"Imagine the Qantas engineer looking up at the plane and seeing an overlay of stress analysis of the joints," said Professor Dwyer. "Or doctors being able to walk the patient through their diagnosis in the context of their actual medical images, and see it sort of floating in space around them."
3. We can see the bigger picture, and keep a project on track
Uniquely human skills including creative problem-solving and communication, combined with an ability to handle uncertainty, equip us to be the project managers that AI systems can't be.
"What we can't currently do is take a universal AI and plonk it down without context or training in a hospital and have it figure out how to run things efficiently," said Professor Dwyer.
Similarly, Professor Garcia De La Banda contrasts the compartmentalisation of machine learning with our ability to "see several things at the same time, integrate them, and really see the possibilities outside that plan".
4. We understand ethics
We may not always act ethically, or even be clear on how to act ethically in a certain situation, but at least we have some understanding of ethics and how they might apply to our work.
In particular, we need to keep an eye out for bias and anomalies in the data that AI systems are based on.
Amazon's recruitment tool, designed to identify desirable employees, is a cautionary tale. The data used to train the AI system – CVs submitted to the company over the previous 10 years - reflected the under-representation of women in the tech industry. As a result, it systematically favoured male over female candidates.
5. It creates unprecedented training opportunities
A slightly different use of the human-in-the-loop approach is to create highly realistic simulation-based training. Simulations for training purposes have been around for some time, but AI modelling takes them to a whole new level, adding interaction with the real world through the use of physical devices as interfaces.
Trainees can immerse themselves in a simulation of managing a large-scale emergency or responding to a particular military situation. It's also an effective way to develop or test specific skills such as flying an aircraft.