Recent observations (1) from the University of Cambridge that machine learning could identify the amount of pain suffered by sheep from their facial expressions is a good tangent to pursue. It may even have applications in humans, unable to communicate because of recent or permanent loss of auditory and visual functions. Machines have been growing in stature and they seem to trump humans in most routine tasks. But increasingly, they are filling the gaps that humans are unable or untrained to do. There is no turning back from the AI train as it had left the station nearly 3 decades ago. Now cheaper and faster computers are making what could not be done by pure imagination.
On the other hand, humans generally get over-excited about emerging technologies and they believe problems could be solved in the matter of months, if not, days. Often, they have been wrong and many examples are available, in air travel, the internet, human genome based medicines and most recently, machine learning. Humans have been slow leaners, in spite of the massive energy hog they have been endowed with and they are programmed to look to the future rather than the past. That is a good thing but looking too far and over the hills may get them into trouble, something that did not exist for most of their evolution.
Understanding pain from facial expressions is a good step forward but a 67% accuracy (1) is not sufficiently robust for practical applications. Machine learning can easily create models of that accuracy from random and noisy data. Before declaring victory, much work is in store to think about what it could mean.
(1) http://www.sciencemag.org/news/2017/06/artificial-intelligence-learns-spot-pain-sheep
On the other hand, humans generally get over-excited about emerging technologies and they believe problems could be solved in the matter of months, if not, days. Often, they have been wrong and many examples are available, in air travel, the internet, human genome based medicines and most recently, machine learning. Humans have been slow leaners, in spite of the massive energy hog they have been endowed with and they are programmed to look to the future rather than the past. That is a good thing but looking too far and over the hills may get them into trouble, something that did not exist for most of their evolution.
Understanding pain from facial expressions is a good step forward but a 67% accuracy (1) is not sufficiently robust for practical applications. Machine learning can easily create models of that accuracy from random and noisy data. Before declaring victory, much work is in store to think about what it could mean.
(1) http://www.sciencemag.org/news/2017/06/artificial-intelligence-learns-spot-pain-sheep
No comments:
Post a Comment