Sometimes there are things that you just know before it happens. Like for example, the fact that your best friend is going to have a burger for dinner or your boss is going to call you for an impromptu chat.
But how does your brain make these predictions?
Our ability to predict certain outcomes is often based on our past experiences stored in our neural networks as information. But exactly how our brains connect series of data and combine them with expectation to predict events in the future is still unknown.
Now NYU neuroscientist David Heeger is offering a new framework to explain how the process might work.
Heeger says: “It has long been recognised that the brain performs a kind of inference, combining sensory information with expectations.
“Those expectations can come from the current context, from memory recall, or as an ongoing prediction over time. This new theory puts all of this together and formalises it mathematically.”
Just like meteorologists use past weather information and current climate to tell you whether it is going to rain or snow next week, Heeger says our brain processes data in a similar way to predict certain events in the future.
He says: “The neural networks in our brains embody a type of model of our surroundings. However, we don’t have a clear understanding of how they operate to make predictions.”
Heeger points out that neural networks used in artificial intelligence use a hierarchical structure – where sensory input comes in at one end and more abstract representations (like memory recall or mental imagery) – are then factored in along the hierarchy.
But this structure, which Heeger describes as “pipeline processing architecture”, does not take into account exactly how the human brain loops in past information to make a prediction.
Heeger says: “It’s possible to run this process the other way around – to take an abstract representation at the top of the hierarchy and run it backwards, from top to bottom through the neural net, to generate something like a sensory prediction or expectation.”
Or, he adds, it can run in a mode that is a combination of the two, in which inferences mix sensory input with prediction.
So basically, Heeger’s model implies our brain uses not just one but several modes to arrive at a conclusion to make a prediction.
According to his hypothesis, neuromodulators (substances released by neuron that help communicate information) like acetylcholine change the brain state to control which mode is operating at each moment in time.
Heeger believes his theory could help scientists who are using AI to replicate human decision-making, saying: “The theory of neural function that I’m outlining aims to fill in some of the significant dynamics that AI is missing.”
Heeger’s research is published in the Proceedings of the National Academy of Sciences (PNAS).
