Towards Augmented Financial Intelligence

Abstract

The development of artificial intelligence (AI) systems is one of the main themes in driving change in the financial industry. It is important to distinguish the different types of interaction between machine and human intelligence since AI is often used as an umbrella term to describe multiple concepts:

- AI 1.0: Machine Intelligence (MI) – algorithms assisting humans by automating processes and aimed toward making intelligent machines (McCarthy, 2004).
- AI 2.0 Intelligence Amplification (IA) – algorithms assisting humans in decision making (cf. new forms of statistics) (Green and Chen, 2019).
- AI 3.0 Augmented Intelligence – humans assisting algorithms through human expert-augmented machine learning (Monarch, 2021).

Most successful intelligent systems are partnerships between algorithms and humans. AI algorithms are good at analyzing large volumes of ‘big’ data and automating processes. Humans are good at insight and prediction. Notably, Superforecasters (Tetlock and Gardner, 2015), individuals that tend to make better predictions than the general public, have specific cognitive and personality traits that cannot directly be replicated by algorithms.

In finance machine intelligence automation includes backtesting, pre-trade analysis, portfolio optimization and smart order routing. Intelligent amplification is mostly used for data analysis, natural language processing (NLP) and sentiment analysis to assists human decision making. Furthermore, due to limited training data and a low signal to noise ratio MI and IA systems often underperform, in applications such as financial forecasting. Therefore, augmented intelligence gathered from Superforecasters, can advance the predictive performance of machine learning.

The focus of this paper is human augmented financial intelligence (AFI), arguably the next major development in financial artificial intelligence. The objective is to capture and automate the insight of Superforecasters and financial professionals using AFI, to enhance the performance of algorithms and then to create an AFI investment process. This entails augmenting the algorithms through human expertise to create a new generation of algorithms. Key challenges of AFI relate to data gathering, expert profile collection, recommendation weighting, model risk assessment and algorithmic automation. The contribution of this paper is to review expert-augmented machine learning and automation of AFI applied to the analysis of financial markets.

Keywords: Augmented Financial Intelligence, Superforecaster, Finance, Financial Markets, Artificial Intelligence, Machine Learning, Machine Intelligence, Human Intelligence, Intelligent Amplification, Augmented Intelligence