Hi everyone,
today I wanted to talk about DeepMind. It is an Artificial Intelligence company that was created in 2010 in the UK by some very brilliant minds such as Mustapha Suleyman, Demis Hassabis and Shane Legg. It was doing so well that the company was acquired by Google in 2014. Information about DeepMind Inception can be found here.
Despite the fact that it is quite young, DeepMind is now a major company and institution, publishing among the highest quality scientific papers and hiring some of the best minds around the world. DeepMind became very well known to the general public when they created an AI who could beat Lee Sedol(best player) at AlphaGo. They were later awarded the very prestigious Marvin Minsky Medal, which is probably the highest kind of recognition you can get in AI.
Recently, DeepMind has been pushing its research in the medical field. Actually, they are working in collaboration with some of the most prestigious Universities in the world such as the University of Oxford. For instance, they recently published a paper that shows that they now have the AI technology to detect about 50 different eye diseases, which is clearly unprecedented. To achieve that kind of prowess, they had trained their AI with about 14,884 eye scans. Another thing for instance that they are also working on is human psychology. It is a joined effort between their AI team, their Neuroscience team and Harvard University. In this paper, they try to understand in which situations humans would try to invest in mental efforts and in which situation they would not. Clearly, DeepMind just like OpenAI's ultimate goal is to try to understand/create the concept of AGI (Artificial General Intelligence), a concept in which the AI would have an intelligence equal to a human (or even greater). It also seems that DeepMind is really trying to have a positive impact in the health sector and bring good innovations. Another project of DeepMind that really strikes me is their project related to grid cells, that I will talk more in detail below.
A new study found that an AI was “more accurate than dermatologists at identifying malignant melanomas”. We need to continue to assessing how systems like this perform over a longer period, but this is very promising. (Mustapha Suleyman, co-founder of DeepMind, Twitter 2018)
Mustafa Suleyman, co-founder of DeepMind
The work of DeepMind on Grid Cells
Raia Hadsell is a research scientist at DeepMind and works on robotics and neural networks. She is very well known in the AI community, for instance she got her Ph.D with Yann LeCunn as an advisor, who is considered to be the father of an AI algorithm called "convolution neural network" and the head of AI research at Facebook. Raia focus a lot on an AI algorithm called Reinforcement learning, the same algorithm that is used for instance by OpenAI to train AI at Dota 2 games. She recently had a podcast with google cloud platform podcast this year where she discusses grid cells, reinforcement learning and spatial navigation in both robots and humans. More on the podcast here if you are curious.
1) What is Reinforcement learning?
Reinforcement learning is an AI algorithm where you train a robot in a virtual environment by letting him be free to completely explore any action and basically try whatever he wants to complete a task. The way you interact with the robot is by setting up a "policy" for him that sets up the kind of rewards it would get for a specific action. I wrote an entire post on this that you can find here if you want more information.
2) What are grid cells?
Grid cells are neurons anterior of the brain that can measure how far we are walking. They help measure both directions and distance. Thus they enable any helping individual to be able to trace back his steps, even in a completely new different route. Grid cells were found by Edvard Moser's group. This is super important as for instance they found that the health of grid cells was associated with the probability of someone to have Alzheimer's disease or not. You can find more information about it by listening to the podcast link that I posted above.
autocorrelation plot of grid cells
The big deal
Raia Hadshell and her team found by training the AI algorithm based on Reinforcement learning and Recurrent Neural Network (another type of algorithm that allows AI to have memory) on navigation, the neural network would actually fire like human grid cells!!!! This nice blog written by DeepMind explains it more in details. This is a major discovery for both robots and humans. For robots, for instance they were able to make the robot learn how to take shortcuts in navigation (in a previously unknown environment). For humans, this might in the future help better understand how the human brain works especially with dementia diseases such as Alzheimer.
Conclusion
DeeepMind is a huge organization that is really making impact in the robotics industry but also in health, in gaming and so on. They have the among the brightest minds and tremendous amount of resource. These days, they make a lot of their findings opensource such as the papers, blogs and some code (if I am not mistaken). In my opinion, they are definitely going to have an tremendous impact in many sectors in the future.
Disclaimer: I tried to objectively write what DeepMind is doing on this post. Note that I did not take any stand (or at least I think I didn't and I apologize to anyone who thinks I did) and only tried to share the information as it is. I am passionate about AI and also passionate about Blockchain, Internet and Freedom at the same time.
Pictures
Picture 1 was taken from https://goo.gl/goxj5W
Picture 2 was taken from https://goo.gl/kxiB1N
Picture 3 was taken from https://goo.gl/dK2Hgo
Picture 4 was taken from https://goo.gl/5jM8J9
Picture 5 was taken from https://goo.gl/iBM7SU
juv79505 sincerely thank you for reading this article. Please feel free to comment below in case you have ideas, questions, suggestions or simply want to criticize this article. Stay tuned for more articles on health, environment, artificial intelligence, video games, technology in general, books, graphic novels, geography, history, sports and much more!