AI Go-playing computer beats all of humanity
DeepMind, a Google project, has an AI player called AlphaGo. The normal approach for the AI to learn the game was to watch thousands of hours of humans playing humans to learn strategies and techniques. The newest AI player, AlphaGo Zero, didn’t bother with watching humans play, it simply started playing Go against itself. A mere three days later, having honed its skills over 4.9 million games in this time AlphaGo Zero beat a predecessor version 100-0. This predecessor had trained for several months and had previously beaten the 18-times (human) world champion.
Advanced AI like this can be applied computer learning and problem solving, not just game playing. It can help solve scientific and complex problems such as protein folding (a team is already working to apply this method to this), climate science, drug discovery and quantum chemistry.
However, it currently relies on the fact that the self-taught AI has to learn based on the rules it is given. Simple rules are ok, but tasks quickly become more challenging and then impossible, the more rules and complexity there is, such as driving.
Source: New Scientist Magazine, 21st October 2017
Genes linked to Obsessive Compulsive Disorder
A team working at MIT and Harvard, led by Hyun Ji Noh, have identified four genes that are linked to OCD. The genes are active in parts of the brain (the striatum, thalamus and cortex regions) that control learning and decisions. In people with OCD, the messages passing between these regions can become corrupted, suggesting it can be harder for sufferers to tell whether a situation is safe or risky.
A genetic link also suggests that it is possible that you are more likely to develop the disorder if a close relative has it, though of course it is possible to develop it if no family member has it and it doesn’t mean that you would develop it if a parent has it.
Understanding this genetic link adds to the hope that better treatments can be developed. For example, by developing a drug that reverses the effect by targeting the linked genes or the pathway it regulates.
Source: New Scientist Magazine, 21st October 2017, Nature Communications DOI: 10.1038/s41467-017-00831-x
Scientist of the Month
Eva Crane (1912 – 2007)
Originally a lecturer in physics at Sheffield University, with a BSc in mathematics, an MSc in quantum mechanics and a PhD in nuclear physics, her resulting lifetime work and interest in bees began when she and her husband were gifted a beehive for their wedding, where the giver hoped it would help supplement their wartime sugar ration.
She was shy and self-effacing and wrote over 300 papers, articles and books, contributing widely to the subject. She founded the Bee Research Association in 1949, an organisation with international aims, but only became the International Bee Research Association in 1976. She also formed the Eva Crane Trust, with aims to advance the understanding of bees and beekeeping by the collection, collation and dissemination of science and research worldwide.
Night Sky this month
13th Nov: Conjunction of Venus and Jupiter
Can be seen between 6:15 and 7:15 am (before sunrise) on the eastern horizon. The planets will appear only 0.3 degrees apart, so less than the diameter of the full Moon (0.5 degrees).
17th/18th Nov: Peak of the Leonids Meteor Shower
Has a rate of 15 meteors per hour. Meteors appear to radiate from the constellation of Leo, but can appear anywhere in the sky. Best viewed after midnight, from a dark location. This meteor shower results from the debris trail left by the comet Temple-Tuttle, discovered in 1865.
18th Nov: New Moon
The new Moon will help with viewing the Leonids meteor shower.