Can AI Predict The Future?
Can AI Predict The Future?
Predicting the future is a fool's errand. Could artificial intelligence help?
Join the DZone community and get the full member experience.Join For Free
It's well known, not least by me, that predicting the future is a fool's errand. The huge range of variables that need to be taken into account makes accurate predictions impossible. Could artificial intelligence help?
That was the question posed by a couple of papers published in the journals Physical Review Letters, and the papers highlight how machine learning can be used to accurately predict the future evolution of fundamentally chaotic systems, even out to quite distant horizons.
The work, from researchers at the University of Maryland, saw an algorithm trained to learn the Kuramoto-Sivashinsky equation that is used to underpin many chaotic systems. It's commonly used in the study of turbulence and similar systems.
Seeing Into the Future
After being trained on historical data, the algorithm was able to accurately predict how the system would evolve out to eight "Lyapunov times," which is around eight times more than existing methods allow.
Suffice to say, the algorithms don't know the first thing about the equations they're working with, only the data that's generated. This marks a promising shift from existing methods that tend to require a detailed knowledge of the system itself.
They suggest that this approach is more effective because traditional methods require you to measure conditions of a system in a specific moment in time and use that data to produce a model that is then evolved forward. To accurately gain predictions eight times further ahead would require initial measurements some 100,000,000 times more accurate.
The team believes it could have a wide range of applications, from monitoring for heart attacks to predicting rogue waves at sea.
So, how close are we to such practical applications? The simple answer is a lot closer than we were. As with so many applications of AI today, the researchers believe that combining it with human forecasters could provide the best immediate impact.
"What we should do is use the good knowledge that we have where we have it," they say, "and if we have ignorance, we should use the machine learning to fill in the gaps where the ignorance resides."
It's some fascinating work, and with the reputation of expert forecasters taking a bit of a battering, any assistance that technology can provide must surely be welcomed.
Published at DZone with permission of Adi Gaskell , DZone MVB. See the original article here.
Opinions expressed by DZone contributors are their own.