For a long time, South Korean Liu Sedol was the world champion in the «Baduk» game (Weiqi in China or igo in Japan). This game is especially popular in Asian countries. For several thousands years only people took part in this game. However, in 2016, the AlphaGo neural network from DeepMind has beaten Liu Sedola. This event can be compared only with the victory of the chess computer Deep Blue over Grandmaster Kasparov. And at the end of October 2017, AlphaGo Zero program beated AlphaGo with a score of 100: 0. To create AlphaGo Zero (AGZ) utilizes technology of «Reinforcement Learning» (training with reinforcement).
This means that the developers did not apply the results of correctly played games.
AGZ began from the scratch knowing only the rules of the game. Then the program chose its the next moves based on Go board constraints. Initially making chaotic moves, through 4.9 million played games AlphaGo Zero has developed its skills so that none of the living players in can compete with it anymore.
The company belongs to Google DeepMind which is known not only for successful implementation of algorithms for “Go” and other games. Now the developers are engaged in ethical and economic problems, which are concerned about the future wide spread of artificial intelligence
Zhengyao Jiang (Zhengyao Jiang) writer from the University of Liverpool, wrote a paper on usage of algorithms of «Reinforcement Learning» in cryptocurrency markets. The results are impressive: using historical data for 8 months, the program achieved a 10-fold increase in the size of the portfolio at 30-minute intervals. To achieve this profit it took a little less than two months.
There are also neutral strategies. They use arbitrage, high-frequency operations, errors in estimates and other anomalies in terms of statistics. Thus, you can get income regardless of the direction of the market. And “Reinforcement Learning” allows you to discover those opportunities that are inaccessible from the standpoint of ordinary human logic. And you can not only detect them, but also turn them into code that allows you to make profitable trades.
In this regard, the cryptocurrency market is more interesting than the stock and currency market, as potentially profitable opportunities appear at any time of the day and any day of the week. The cryptocurrency market works 24 hours a day and 7 days a week, and operations in this market can be easily done between different countries without worrying about geographical boundaries. Many operations in this market can be automated using scripts, APIs and smart contracts.
The main problem that faces developers is the accuracy of forecasts. To increase the reliability of the forecast, mathematical and statistical data are used first. They are easiest to get from the crypto-exchange exchanges through the API, and these data are easy to process and analyze. And the information received in non-numerical form (for example, subjective opinion of investors) also turns into a set of various digital parameters that the system can analyze.
With the help of machine learning, regularities are established, which are subsequently used.
Analyzing profitability, the system necessarily takes into account risks. And with the help of the portfolio distribution between the various digital assets, you can achieve a permanent reduction of the overall risk, or make the risk a customizable parameter. Over time, artificial intelligence also learns to determine knowingly losing trends, and exclude them.
CRYPTICS believes that it is important to relate to the academic community, by connecting breakthrough researches with the practical considerations. In particular, Cryptics plans to organize contests and events dedicated to solve AI and data problems appearing when advancing the forecast algorithm, trading platforms and in the overall cryptocurrency sector of the economy. Those contests offer cash prizes for the best solutions, and will involve both closed and open-source results.