Nightmare Machine is being taught how to scare us
CSIRO researchers are using visions of the undead and the human face to teach a machine what terrifies us most.
Nightmare Machine is an algorithm-based piece of artificial intelligence, or AI, created by a team of researchers at CSIRO and the Massachusetts Institute of Technology (MIT) that spontaneously generates zombie faces out of human ones and transforms images of places into visions of the inferno.
Dr Manuel Cebrian Ramos, a research scientist at Data61, the CSIRO’s digital and data innovation group, and his colleagues fed 200,000 images of normal human faces into the machine’s neural network to teach it to recognise faces.
The algorithm was then able to generate faces at random according to what it had learnt.
They then added a single zombie face, giving it slightly more weight in the neural network than the others to turn human faces into zombies.
Dr Cebrian said feedback from people who have answered an online survey gauging how scary they found the newly transformed faces and places would soon to added to the machine to enhance the images it generates.
“Once we are able to incorporate close to a million reviews of what’s scaring people, you can actually feed that back into the algorithm and the algorithm can generate even scarier images,” he said.
Machines ‘are becoming more human’
However, Dr Cebrian said the Nightmare Machine is not simply a seasonal novelty, it also has practical implications.
“Our main goal is obviously not to scare people – this is just a Halloween fun goofy project,” he said.
200,000 pictures of faces were fed into the machine’s neural network to create the creepy images. Photo: CSIRO
“What we are very interested in is how to instil particular emotion in people – so can we feel positive emotions, like warmth friendliness, a machine telling the human ‘work with me, trust me’.
“In order for these machines to not scare humans, you need to teach them first what scares human, this is what we do with children.
“The fact that we can tell machines, ‘Hey, this is really scary, don’t do any of this’, that might be a practical application for the Nightmare Machine.”
For some people it is not just the images produced by the machine that they find terrifying, the fact that the machine is capable of learning and thinking creatively is equally off-putting.
“As machines start to look more human, start to act more human, start to do things that we consider to be human, our sense of uniqueness goes away.”
Data 61 research scientist Dr Manuel Cebrian
“Now as we know the machines are getting closer and closer to being like us, that is really scary for humans, we’ve known this for a while.”
‘It’s not Terminator we need to worry about’
While it may be a while before the machines rise up to annihilate us, there are more mundane concerns to be had as well.
A photo posted by Nightmare Machine (@nightmare_machine) on
Toby Walsh, professor of AI at the University of NSW and group leader at Data61, said the things people should actually be worried about are not really things like Terminator.
“There’s much more immediate threats, I think we’re already seeing that today with things like autonomous cars,” he said.
“We’ve started to realise that we’re giving responsibility, autonomy to machines and they may be making life or death decisions.”
University of NSW AI Professor and Data 61 group leader Toby Walsh
“Recently there was a story about how a program was being used to decide sentencing in the United States, it was biased against black people.
“I think we’ll start to realise the consequences of giving too much responsibility to machines.”
– ABC
