Googles deepmind is a Google’s division of artificial intelligence research. They believe by making an artificial intelligence interface they can increase our understanding of science and the universe. They want to make this AI available to the world. They feel like this extraordinary AI will improve our relationship with other countries and can improve our intelligence and understating as a species. They are very open with their work. They publish their research, and every piece of research they do is available online and is peer-reviewed.
Deepmind is focused on solving intelligence. So they are working on creating an intelligence that through algorithms will be able to improve the quality of life for humans but also understand the mysteries of the human brain.
In their quest to create this AI they have published many papers on AI available online and through their site.
History Deepmind was started in London in 2010 by Dr Demis Hassabis, Mustafa Suleyman, and Dr Shane Legg. Deepmind was later acquired by Google in 2014 for 500million[i].
Testing Just like any technological product, in-order to guarantee if it's making a positive impact, the creators will put their product through a wide variety of tests to make sure it functions as they'd like it to. For DeepMind, they have gone about their testing in a very unique way. They have put their AI to the test by having it playing games. Shane Legg has been quoted saying “when a machine can learn to play a really wide range of games from perceptual stream input and output and transfer understanding across games” is when human-level machine intelligence is achieved. They are currently testing their AI playing seven different Atari games, Pong, Breakout, Space Invaders, Seaquest, Beamrider, Enduro, and Q bert.
Capability
Alpha Go
In 2015 Deepmind tested one of their AI’s named AlphaGo against Fan Hui, the European Go Champion[ii]. Go is a traditional Chinese game resembling checkers. Deepmind decided to use Go because of the many pieces it is considered one of the most complex games with thousands of possibilities.
Before 2015 previous AI’s were able to compete with Go players on an amateur level but AlphaGo was able to compete at an expert level and won 5 to 0.
Differentiable Neural Computer (DNC)
In 2016, Deepmind introduced a new for of memory augmented computing called differentiable neural computers or DNC for short. They developed this technology to demonstrate that a computer network can learn to use memory in order to answer very complex problems and complete difficult data driven tasks. Some examples of tasks that DNC technology can complete are artificially generated stories, family trees, and maps of city underground train systems. These computers learn how to use memory and how to produce answers by learning to use optimization. When the computer produces an answer that answer is compared to a correct or desired answer and the system uses a controller that is similar to a CPU to develop a memory of what is correct and what is incorrect. Essentially this system will be able to learn on its own as long as good inputs are given.
Benefits In 2016 Deepmind announced they would be collaborating with Moorsfield Eye Hospital. Deepmind will apply their AI to the analysis of eye scans to search for diseases that can lead to blindness.
They also have a research program with the University College London Hospital where they are developing an algorithm that can differentiate the difference between healthy and cancerous tissue in the neck and head.
Deepmind is also being used to improve efficiency of Google. The Deepmind technology has led to breakthrough in the energy efficiency of Google’s data centers, reducing the energy consumed by 40%
Deepmind would be beneficial for doctors as it would be able to help diagnosis problems, so that patience would be able to get the right treatment. Deepmind has many benefits because it could solve a lot of problem that we have in the world today.
*In the world
*In our work
*In our daily lives
Preventing the AI from becoming too “smart” Deepmind has developed a “shut off” button to prevent their AI from becoming too smart and taking over. This will allow humans to turn off the AI whenever they want to. This will insure that the AI does not become too aware and take control of itself. [iv]
Although this requires more human power, because a human would need to supervise the AI at all times it will keep the AI from malfunctioning and causing harm to itself or others.
What are their plans for the future?
Because of the success of AlphaGo playing Go and dominating, Deepmind had turned its sites on Real-Time Strategy games. Deepmind is currently turning its attention to the game StarCraft II. This would be more of a challenge to AlphaGO because unlike the game GO the AI would be unable to see the full ‘board’ but would need to anticipate the moves of the opponents. [v]
Where can you find out more about Deepmind and their research?
Google Deepmind
Googles deepmind is a Google’s division of artificial intelligence research. They believe by making an artificial intelligence interface they can increase our understanding of science and the universe. They want to make this AI available to the world. They feel like this extraordinary AI will improve our relationship with other countries and can improve our intelligence and understating as a species. They are very open with their work. They publish their research, and every piece of research they do is available online and is peer-reviewed.
Deepmind is focused on solving intelligence. So they are working on creating an intelligence that through algorithms will be able to improve the quality of life for humans but also understand the mysteries of the human brain.
In their quest to create this AI they have published many papers on AI available online and through their site.
History
Deepmind was started in London in 2010 by Dr Demis Hassabis, Mustafa Suleyman, and Dr Shane Legg. Deepmind was later acquired by Google in 2014 for 500million[i].
Testing
Just like any technological product, in-order to guarantee if it's making a positive impact, the creators will put their product through a wide variety of tests to make sure it functions as they'd like it to. For DeepMind, they have gone about their testing in a very unique way. They have put their AI to the test by having it playing games. Shane Legg has been quoted saying “when a machine can learn to play a really wide range of games from perceptual stream input and output and transfer understanding across games” is when human-level machine intelligence is achieved. They are currently testing their AI playing seven different Atari games, Pong, Breakout, Space Invaders, Seaquest, Beamrider, Enduro, and Q bert.
Capability
Alpha Go
In 2015 Deepmind tested one of their AI’s named AlphaGo against Fan Hui, the European Go Champion[ii]. Go is a traditional Chinese game resembling checkers. Deepmind decided to use Go because of the many pieces it is considered one of the most complex games with thousands of possibilities.Before 2015 previous AI’s were able to compete with Go players on an amateur level but AlphaGo was able to compete at an expert level and won 5 to 0.
Differentiable Neural Computer (DNC)
In 2016, Deepmind introduced a new for of memory augmented computing called differentiable neural computers or DNC for short. They developed this technology to demonstrate that a computer network can learn to use memory in order to answer very complex problems and complete difficult data driven tasks. Some examples of tasks that DNC technology can complete are artificially generated stories, family trees, and maps of city underground train systems. These computers learn how to use memory and how to produce answers by learning to use optimization. When the computer produces an answer that answer is compared to a correct or desired answer and the system uses a controller that is similar to a CPU to develop a memory of what is correct and what is incorrect. Essentially this system will be able to learn on its own as long as good inputs are given.When
There are details on open-sourcing DeepMind available on the DeepMind web site here is the link https://deepmind.com/blog/open-sourcing-deepmind-lab/
You can also use GitHub if you want to check out the codes used for DeepMind https://github.com/deepmind/lab
Benefits
In 2016 Deepmind announced they would be collaborating with Moorsfield Eye Hospital. Deepmind will apply their AI to the analysis of eye scans to search for diseases that can lead to blindness.
[iii]
They also have a research program with the University College London Hospital where they are developing an algorithm that can differentiate the difference between healthy and cancerous tissue in the neck and head.
Deepmind is also being used to improve efficiency of Google. The Deepmind technology has led to breakthrough in the energy efficiency of Google’s data centers, reducing the energy consumed by 40%
Deepmind would be beneficial for doctors as it would be able to help diagnosis problems, so that patience would be able to get the right treatment. Deepmind has many benefits because it could solve a lot of problem that we have in the world today.
*In the world
*In our work
*In our daily lives
Preventing the AI from becoming too “smart”
Deepmind has developed a “shut off” button to prevent their AI from becoming too smart and taking over. This will allow humans to turn off the AI whenever they want to. This will insure that the AI does not become too aware and take control of itself. [iv]
Although this requires more human power, because a human would need to supervise the AI at all times it will keep the AI from malfunctioning and causing harm to itself or others.
What are their plans for the future?
Because of the success of AlphaGo playing Go and dominating, Deepmind had turned its sites on Real-Time Strategy games. Deepmind is currently turning its attention to the game StarCraft II. This would be more of a challenge to AlphaGO because unlike the game GO the AI would be unable to see the full ‘board’ but would need to anticipate the moves of the opponents. [v]
Where can you find out more about Deepmind and their research?
You can visit https://deepmind.com/ to learn more.
[i] The Company. (n.d.). Retrieved March 18, 2017, from https://deepmind.com/about/
[ii] M. (2017, January 5). Google secretly squared off its AI against leading Go players and it won by a landslide. Retrieved March 18, 2017, from https://thenextweb.com/google/2017/01/05/google-alpha-go-victory/#.tnw_Kcn0UxLh
[iii] Deepmind- Healthcare. (n.d.). Retrieved March 18, 2017, from https://en.wikipedia.org/wiki/DeepMind
[iv] Shead, S. (2016, June 3). Google has developed a 'big red button' that can be used to interrupt artificial intelligence and stop it from causing harm. Retrieved March 19, 2017, from http://www.businessinsider.com/google-deepmind-develops-a-big-red-button-to-stop-dangerous-ais-causing-harm-2016-6
[v] Byford, S. (2016, November 4). Google DeepMind's next gaming challenge: can AI beat StarCraft II? Retrieved March 18, 2017, from http://www.theverge.com/2016/11/4/13518210/deepmind-starcraft-ai-google-blizzard