Information overload may be the curse of the digital age but it’s definitely the enemy of sound decision-making, which is the daily job of money managers around the globe. Even those investment managers with enormous resources and professional networks are struggling not only to keep up but to discern reality from crowd psychology, and make accurate predictions that outperform benchmark indices on a risk-adjusted basis.
Fortunately for asset managers and their clients, there’s a solution: Augmenting human capabilities with artificial intelligence (AI) to surface actionable insights buried in the nuances of language hidden in large volumes of constantly changing data.
At Accrete.AI, our bias-free, contextually adapted investment tools empower investors to spend less time searching for insights and more time making informed decisions. Our products are helping analysts and portfolio managers detect changes in nuance across key topics mentioned in earnings calls, find actionable M&A rumors, parse confusing Fed-Speak and identify supply chain risk before others.
We apply Watson services with our proprietary AI in three key ways:
- Discover, ingest and normalize massive amounts of social media, earnings calls, Fed chatter, company filings, news, and other data investors rely on, in real-time.
- Create a training foundation that reduces bias and increases contextual awareness so that our solutions continuously learn and perform accurately in a constantly and rapidly changing environment.
- Help customers analyze and evaluate critical data and its likely impact on market performance, with less effort and greater confidence.
The digital revolution has changed the nature of volatility
There’s no question that successful money managers must build, or accrete, knowledge over time. But, today’s savvy investment managers would need to read millions of documents daily across a variety of domains to even begin to keep up. The real question for today’s investors is: Are you able to accurately account for the new and idiosyncratic nature of volatility resulting from digitally amplified crowd behavior? For most, the answer is no.
That’s why, through our partnership with IBM, we’re embedding Watson technologies into intelligent systems that read, understand and learn at scale across a variety of market domains, continuously compounding knowledge and connecting dots. They’re literally always working to make sense of the digital world and offload cognitive bandwidth so that investors can account for unforeseen risk and take advantage of new opportunities.
Not all AI is created the same
Artificially intelligent systems are only useful to investors in a practical way if they’re smart enough to adapt to changing conditions and solve real-world problems. In theory, observers might think that general artificial intelligence would be able to understand fundamental concepts such as Federal Reserve hawkishness or what an actionable M&A rumor looks like. The reality is that machines can only understand context in information with our help. On the other hand, human beings are ill-equipped to constantly teach machines every nuance of human language because in the digital age nuance evolves far faster than humans can keep pace with. Also, if biased humans teach machines, then the machines simply learn and apply those biased perspectives at scale.
AI used in investing must be trained using methods that can reduce or even eliminate bias, and that enable machines to learn new contexts in order to achieve sustained, high levels of accuracy, and enable investors to meet their clients’ performance expectations. That’s what we do.
The first step toward bias-free contextual awareness is to train your AI to a specific domain
At Accrete.AI, we combine collective human intelligence with supervised and unsupervised machine learning to train machines to be bias-free and contextually adaptive at scale. IBM Watson Knowledge Studio and Watson Discovery Service play an important role in the supervised learning process.
To help investors solve real-world market problems, we start by defining domain-specific pain points related to information overload.
Then, we train machines to interpret different contexts, for example, where the same word could be used with potentially different meanings. Consider two headlines that read “Apple’s Jobs Rising” and “Jobs sliding at Apple.”
The machine would need to discern whether the word “Jobs” refers to employment opportunities at Apple, or to Apple’s founder Steve Jobs. Those different meanings could have vastly different implications in a market context. This step begins to address meaning and relevance, but it doesn’t fully address bias in context.
To do that, we source and vet domain experts from academia and industry, spanning a variety of perspectives. These formally engaged domain experts identify important entities and relationships in training data we provide to them. Our domain experts don’t know each other and work independently, so when a majority of experts in a particular domain agree upon the importance of a concept, that concept can be characterized as a bias-free ground truth. Then we build ontologies from kernels of ground truth to teach Watson how to speak the vocabulary of a specific domain in a bias-free way, thus seeding our neural networks with trustworthy contextual awareness.
A constant collaboration among humans and machines
Like a child learning to ride a bicycle using training wheels, our solutions learn over time and need less support. It’s a constant trade-off between supervised and unsupervised learning depending on the amount of available training data and the scope of the domain-specific problem we’re trying to solve. For example, there are only so many ways people reference actionable M&A rumors in social media, and there’s lots of training data available, so the scope of the problem is relatively limited. In contrast, while there’s enough training data covering nuances in earnings calls across each company in the S&P 500 going back 10 years, actually mapping linguistic nuances to key topics is much larger in scope, making it necessary to leave the training wheels on longer.
A scalable platform with unlimited extensibility
IBM is only one of a handful of cloud companies with the computational resources necessary to crunch the mountains of unstructured data that investment managers face every day. That’s why we’ve built our smart investment tools on IBM’s infrastructure, and why we’re making them available on IBM’s Financial Solutions Cloud Hub. Accrete.AI leverages expert networks, shallow learning and deep learning techniques to build smart investment tools that solve real-world problems and help investors generate alpha in the digital age.