AiThority Interview Series With Xin Song, CEO, Bottos
Xin has 13 years of experience in investment, strategy and restructuring of enterprise digitization. He was the head of Droege Group China, an Europe based investment firm with 10 Bbillion EUR AUM. While he worked at Libery Mutual and Accenture, he helped dozens of European and Chinese companies step in to digitization and the Artificial Intelligence space.
BOTTOS is a decentralized data sharing network protocol based on the Blockchain technology. It is also a consensus-based one-stop platform to implement the registration, distribution and transformation of data among different participants within our broad-based AI ecosystem.
Tell us about yourself and the journey to creating Bottos.
My background is in high-tech investment, working for Fortune 500 companies in Europe, Asia and the US I have also done IT consulting.
I was engaged in the IT consulting business more than ten years ago. It was mainly around the development of enterprise-level software applications and the implementation and management of large-scale projects. I completed my MBA in the US ten years ago and then transitioned to the investment field. After I graduated, I worked at Liberty Mutual in Boston for two years. Then, I decided to move back to China, where I worked for the Droege Group for five years and was responsible for it’s investment business in China. The Droege Group is one of the largest family-owned investment companies in Germany.
I have been particularly interested in AI and blockchain technology. In 2017, I met Tingting Wang, who told me about this project called Bottos. I found Bottos to be a great combination of artificial intelligence, blockchain and big data. Blockchain and the AI industry have a very bright future and the team is very solid. I also have a strong passion for entrepreneurship; therefore I decided to make this project happen with several co-founders of Bottos.
Investing is really just about looking at different projects and conducting in-depth research to understand each project in its industry; for example, what kind of products, how the market is, competitors, etc. Basically, the more projects you see, the better sense you have at predicting what kind of projects are more likely to succeed. I have personally screened hundreds of projects. I thought that the Bottos project had great potential: a good market, a strong team, and a novel product which crosses AI, blockchain and big data, so I chose to become a part of it. I am a person with a comprehensive background, and this opportunity seamlessly combined my skills and experiences in technology, management and investment.
Tells us about your team.
I am one of the co-founders, and the other is Tingting Wang, who was the chief marketing officer at NEO, the distributed smart economy network in China. She also founded Rivexo, a robot exoskeleton research and development company. We also have Chao Wang, who worked at Wanxiang Blockchain Lab and Huawei Technologies; he has a tremendous amount of blockchain experience. Zhen Gao is our AI architect; he has published more than 100 papers on AI, robotics, machine learning and automation. We are very proud of our team, which has extensive experience in AI, blockchain and finance. We also have a very strong advisory team and an experienced tech team that works on blockchain development.
Could you explain to us how much time, cost and effort is spent while building a model in preparing the data and training the model? How does the Bottos platform change this convention?
Data is definitely the pain point in any AI project. The cost of data acquisition can take as much as 50% of the budget of an AI project.
That means global giants like Google and Apple are the ones that can afford the data needed for AI development.
We developed the Bottos Data Marketplace to make data more accessible and affordable, so smaller AI companies can get the data they need for their projects.There are three key factors in AI development: computing power, algorithm and data.
Usually, an AI developer finds a workstation with sufficient computing power to start to write an algorithm. Then the developer finds as much data as possible to train the algorithm to be ready for use. The data training stage takes as much as 70% of the AI development time and 50%-60% of the total AI development cost.
Data preparation is much more difficult than you might expect. It usually takes 80% of the data scientist’s time. Here’s a more complete breakdown of time for an AI project:
- Cleaning and organizing data (60%)
- Collecting datasets (19%)
- Building training dataset (3%)
- Mining data for patterns (9%)
- Refining algorithms (4%)
- Other (5%)
In turn, you can also argue and classify that cleaning and organizing data is considered the least enjoyable activity by researchers (57%), followed by collecting datasets (21%) and building training data (10%).
Regarding the costs, this is a question that varies depending on the industry/sector/stage. A good proxy though is obtainable by estimating that the median salary for a data scientist in US is $112k, which means that only in cost of human resources you spend $90k (80% of $112k) per person doing the job. Multiply this for an average team of 4-5 data scientists, and basically you’re already wasting $500k annually in data preparation. Plus the costs of cloud, etc.; it’s a lot of money.
We encourage companies to provide data preparation services through the Bottos Data Market.
What are the conventional avenues to source relevant data? How does a Blockchain platform ensure that the data is not tampered or manipulated?
The conventional avenue to source relevant data goes like this: most data is owned by global giants such as Facebook, Google, Amazon, etc; they sell some of their data through limited agencies, and at very high prices. There are also data brokers, who collect and sell data. Due to the limited sellers on the data supply side, the price is high.
As for security, we use time-stamp and similarity-comparison to detect data manipulation. If the similarity of two data packets are higher than 85%, the one with later time stamp could be suspected as being fake data. Time stamp guarantees all transactions on blockchain platform are immutable and traceable.
Who can leverage the Bottos platform to further their projects?
Any project that has an algorithm and needs data can leverage the Bottos platform. The most common applications would be AI, VR, robotics and IoT.
- AI : We have a community of more than 150,000 people who are willing to create, share, tag and clean data for AI training. This makes the platform more efficient for finding the kind of data that individuals can easily provide, such as data for facial recognition or handwriting recognition. By utilizing the Bottos community, we can create or collect 1,000 pieces of training data over the couse of several hours.
- Robotics: The data market can provide specific videos for machine vision training. For example, security robots need to learn from videos of people damaging buildings. The model market can help avoid repeated training. If a robot learned the skill of picking up a can or grabbing items with different shapes and colors after 3,000 hours of practice, a developer could pass those skills on to other robots.
How is the pricing determined for data asset transaction?
We offer an open marketplace, so the market price is mainly determined by supply and demand. Data requestors (e.g. AI or IoT companies) will post their data requirements and give a price they are willing to pay for qualified data (in terms of BTO tokens). If no one responds to the data requirements, it could mean the price is too low and the buyers might want to increase their offer.
Take us through the process of accessing the Data Market and the Model Market.
The Bottos Data Market will be officially launched on May 31 and will be available on Bottos GitHub page. After a quick and simple registration, it will ready to use. The release date of Model Market has not been decided yet, at this stage we are focusing on optimizing the data market.
How does the Intelligent Storage system work to protect ownership of data, GDPR compliance and privacy of users?
The intelligent storage system is a distributed storage network. It protects data security by slicing one data packet into thousands of smaller packets and stores them in thousand different places around the world. The GDPR gives EU citizens more control, choices and rights over how their data is used and puts forth guidelines for the collection and processing of data for businesses. As a platform, Bottos offers a decentralized data marketplace to facilitate the exchange of data between data owners and data requestors (e.g. AI or IoT companies).
Bottos does not buy the data nor does it have access to the data traded through the platform. The data will be encrypted when the data owner uploads it to the platform. Only the data owner and permitted buyer will have access to the data. We expect that parties will perform their due diligence as they enter into agreements, and we can also provide data verification services. We want to make sure that data that is being sold belongs to the seller, and the transparency and immutability of the Blockchain will also help us achieve that goal.
Your Whitepaper talks about how Bottos will ensure that AI is able to break the oligarchy. Please elaborate.
The cost of data acquisition can take as much as 50% of the budget of an AI project. That means global giants like Google and Apple are the ones that can afford the data needed for AI development. We developed the Bottos Data Marketplace to make data more accessible and affordable, so small and medium-size AI companies can get the data they need for their projects.
How do you see Bottos contributing to future of AI technology?
Bottos aims to accelerate the development of AI by helping small and mid-size AI companies and projects get affordable training data, storage and computing power. Eventually, our model marketplace will help facilitate model commercialization. Our goal is to provide a level playing field for small and mid-size AI companies.
Thank you, Xin! That was fun and hope to see you back on AiThority soon.