Last year, a new kind of car was seen driving on the roads of Monmouth County, New Jersey. It was an experimental car developed by researchers at chip-maker Nvidia. In many ways it looked just like the self-driving cars being developed by Google, Tesla, and General Motors. What set it apart is that it didn’t follow a single instruction for a driver, engineer, or programmer. It relied entirely on its in-built algorithm that had taught itself how to drive by watching humans do it.
This is an amazing achievement that highlights the tremendous strides being taken in the field of Artificial Intelligence. It also highlights one of the key problems: since the AI taught itself to drive, there is no way to know exactly what kind of choices it will make in the future. If this car happens to drive its passengers into a tree or off a bridge, there will be no way to know why it made that choice, or find a way to fix it.
If we are going to have self-learning AIs, we also need a way to structure their education so we know what they have leaned and can figure out why they act in the ways they do. If we don’t do this, there is no way to anticipate the dangerous decisions the AI may make.
Some of the roles AI are expected to take a larger role in in the future include deciding who gets approved for loans, gets hired or fired from a job, they will take a larger role in making medical diagnoses and administering treatment, conducting operations on the battlefield, and overseeing world-wide trade. It will be increasing crucial that what the AI knows is not contained in a black box.
What we need are large datasets of hyper-accurate, labeled information. These datasets can then be presented to developing AI in sandboxed (isolated) environments where they can practice, make mistakes, and learn the skill the are expected to use in the real world. With these large, well-understood datasets, it will also be possible to go back and figure out exactly what the AI was thinking when it made a specific decision and make alterations to ensure that it is coming to the right decision.
Where do you get that type huge, labeled, hyper-accurate datasets?
Neuromation is a new project designed to address exactly this issue. Since the only current way of creating the large datasets is to do so manually, which effectively means it can’t be done, Neuromation is building a large, distributed network of computers that will use their spare processing power to create datasets on demand.
Why would people give their valuable computing power to this project?
Neuromation is currently in the middle of a token sale that is distributing the ‘grease’ that will allow the whole distributed system to work. The Neurotoken (NTK) will be bought or earned by those training AI to pay the network to create their custom dataset. Those who create the sets will be able to use the tokens to pay for their own data needs or sold to others who need them. An NTK economy will spring up around the massively growing field of AI learning allowing the token to accrue and retain value. NTK will, of course, also be tradeable for cryptocurrencies or fiat.
Why should I trust Neuromation?
There are a lot of ICOs right now, and investors should always be careful and do their due diligence before putting any money into project. But there are also ways to see if a project is actually trying to build something useful or are simply putting up a front to see if they can get funding.
One of those ways is to look at the story they tell. Is the story designed to hook into the emotions, or something that may be a little obscure and difficult to get your head around? The second case is the one that deserves another look. The idea of creating datasets for the purpose of AI learning is not something that occurs to people wondering how to create an ICO. Rather, it’s something that people who are working on the problem have come up with and have realized that blockchain technology is something that may actually, finally, provide a viable solution.
One other achievement of note is that the Neuromation project third place at this year’s international d10e conference on distributed systems, garnering a 100,000 dollar prize. This is their second win at this conference.
Key information about the project:
Website – //neuromation.cryptonomos.com/
Category – ERC20 (click to find out what this means)
Legal Name – Neuromation.OU – Estonia based company
Social Media/Marketing Info
Bitcoin Talk – //bitcointalk.org/index.php?topic=2280785
Telegram – //t.me/Cryptonomos_ICOs
Pre-sale starts: October 25th, 2017 – GMT
Pre-sale ends: When ICO begins
The main sale starts: Nov 28th 2017, GMT
ICO stops: 24 hours after secret cap is reached, or Jan 1st
NeuroTokens Minted: 1,000,000,000
Issued in ICO: 700,000,000
Reserve: 300,000,000 (for liquidity)