ChinaScope's CTO Roger Xia recently gave a presentation to a whole group of suits and nerds on how artificial intelligence is currently being applied to Finance in China by ChinaScope. The presentation is in Chinese, so this post will probably alienate a bunch of people. But, who cares, OCR this baby, pop it into the translation engine of your choice and you will get the gist, or just keep on reading my paraphrase of Roger's presentation.
Now, to the average Li, Wang and Chen in China, when it comes to machine learning or AI, they think of voice recognition, driverless cars, facial recognition, human genetic mapping, intelligent home appliances, etc. It seems like not another day goes by is the world around us not reshaped a little bit more by machines. We are marveled by the pace of innovation in our every day lives. Knowledge seems to be something that we humans no longer have a monopoly over. Yet, in the shiny financial citadels that stand in Shanghai, Hong Kong and Beijing, much of the knowledge aggregation and dissemination remain largely a manual endeavor carried out by well educated homo sapiens in even better tailored suits. In the 21st century, this just ain't good enough. Let's see if we can do something about it.
First, there was SAM, named ironically after the lead character in Sean Penn's 2001 movie "I am Sam", and also an acronym for Segment Analytics & Mapping. SAM is a system that identifies individual business lines of companies on a numeric basis, and in turn allows anyone (professional or otherwise) to perform accurate comparable company analysis in seconds. Incidentally, it also corrects the prolonged problem of industry mis-identification that has plague the financial industry like a bad case of herpes.
And then SAM spawns. One of the key progeny of SAM is Industrial Value Chain, which links together the data nodes of SAM to map out the various upstream and downstream relationships of all the business lines of every single company. Yeah, yeah, ok, it's only about 17,000 listed companies for now, but it's growing like veins on Dwayne Johnson's biceps. So, how did we do this? Well, originally it was by hand. Who says machine intelligence starts off as intelligent? Was Michael Jordan born slam dunking? It was a process of human initiation and gradually handing off to the machines, which evolved from weak (e.g. auto-matching) to strong (e.g. deep learning).
Today, ChinaScope uses a fully automated learning system that does not require human data input to transform all digitized disclosure into standardized data formats. This is allowing us to capture all of the information revealed in financial filings, achieving a level of systematic transparency that has never even come close to before. At this point, people are saying, but financial filings have a problem. They are published at the neck breaking rate of once every 3 months, and the juicy stuff is only released once every 6 months. Yeah, they have a lot of information, but come on, time latency, baby!
So, we created KAM, which is short for Knowledge Analytics & Mapping. Whereas SAM is a deterministic data driven system, KAM is a real time non-deterministic natural language driven system. Every day, our machines rummage through tens of thousands of news, research reports, regulatory filings and social media postings to look for a myriad of information dimensions such as people, companies, events, themes, hot topics. The machines then disseminate logical patterns that link them together, and analyze the polarity sentiments of the postings and track trends of themes, and automatically create investable stocks around those themes. Human interference in this process lies only in quality control and the continuous supervised training to achieve higher accuracy. The hall mark of our ability to do this lies in our three year journey of developing our Natural Language Processing technology. And let me tell you, if you think it is hard for a person to learn to read Chinese, try teaching a freakin' hunk of metal. For the nerds who are reading this, the following is just to give you a taste of how we are doing this.
This is a rough look at how we are able to identify entities in documents.
This is how we created polarity sentiments for all the news articles.
And this is how we are able to capture new topics of interest that fall outside the scope of things that are being actively monitored. The idea is to go one step beyond helping people getting from the known unknown to the known known by having machines probing into the realm of the unknown unknown.
So, it is quite incontrovertible that innovation in finance needn't just happen at the transaction end. So much can be done at the knowledge end, and for ChinaScope, we are putting our engineering might behind NLP, Deep Learning, and Image Recognition to build a smarter financial engine. But as we stand right now, if we need to be honest with ourselves, AI is a bit of an exaggeration. I think for the foreseeable future, we will be racing ahead in the land of IA: Intelligence Augmentation. Maybe true AI is for now a flashing dot in the distant sky, but I am cool with that, because as long as we are going toward that, then I am going to be employed for a long time, and chicks dig that.