FOR ARTIFICIAL INTELLIGENCE (AI) scientists, the idea of being able to show a computer a human brain and just go "do that" must seem like some sort of neural holy grail. But it could be becoming a reality after a team led by Harvard University found a way of using fMRI brain scans to demonstrate the thought process to an AI computer.
fMRI (the 'f' is for functional) images of a volunteer were fed into a machine learning algorithm, 1,200 of them to be precise, each showing the blood flows around the brain in response to viewing particular images, reports New Scientist.
The images were graded according to the strength of the brain wave patterns, and therefore the relative difficulty of interpreting them. As a result, the team were able to "punish" mistakes made in interpreting complex images more than simple ones, in accordance with the risk-reward-pleasure aspects of machine learning that are more established.
Training the machine this way improved accuracy by 10-30 per cent and the team believes that it could allow basic machine learning models to behave and perform more like fancy-pants neural networks of the type that we're seeing from the likes of Google's DeepMind division.
The objective is to make machines think more like us so we trust them more. Not because it makes them infallible, but because it makes their mistakes more relatable, rather than alien and robotic ones that frustrate us.
The next stage will be to study how rats' brains' react to different images, and use that information to further enhance the accuracy of the thought processes at play and to "humanise" them.
"The really exciting stuff is going to come from looking at the fine-grain detail of how individual cells are connected and how they're firing," said lead scientist David Cox at Harvard University.
The work being conducted here could even allow human and synthetic brains to work together, which would be good news for Elon Musk, whose latest company is researching just that. µ
And, er, not much else
To serve, protect, and get incredibly hot and dusty
Symantec links attack to prolific Lazarus hacking group
Chinese firms drive global smartphone growth in first quarter