Artificial Intelligence (AI) is a broad term that is used with increasing frequency across all industries, yet it remains nebulous and is often misused or mischaracterized in its applications. In the simplest of definitions, any device that can perceive the environment around it and has the ability to take action in order to maximize its possibilities of success can be said to have at least a degree of Artificial Intelligence.
We see AI everywhere in our day-to-day lives, ranging from facial recognition software on our smartphones (with success defined unlocking your phone) to driver assistance features in our cars (with success being defined as avoiding hazards).
At its heart, Artificial Intelligence is math. This math is brought to life in the machines in the world around us through computer code in a way that allows us to harness its power and direct it towards practical solutions. The ability of these machines to learn is central to what makes technology intelligent. Deciding what actions to take based on constantly changing data from the surrounding environment, rather than by selecting from a limited set of pre-defined alternatives, is what sets AI apart from simple machine automation.
General Principles of AI
There are four general principles of AI that are critical for it to truly exhibit intelligence:
Data Quality Matters
We’ve all heard the expression “garbage in, garbage out” and that’s especially true for AI. The comprehension, actions and learning performed by AI are only as good as the data that it senses. Unfortunately, the big data revolution of the past decade focused strongly on the quantity of the data amassed, rather than the quality of the data. Poor quality data leads to inaccurate and unreliable results and provides the AI with a warped view of the reality of the environment around it. For high-consequence decisions, this can lead to disastrous results.
Evolution of Modern AI
The concept of AI has been around for decades, and its potential has been the stuff of science fiction. As computational horsepower and cost-effectiveness that modern computing offers have grown, AI has now moved to the forefront and is putting us at the brink of a fourth industrial revolution. In this, we will see the fusion of technologies and a blurring the lines between the physical and digital realms.
As noted, AI includes basic mathematical functions and as a result the term can be potentially misleading. Simplistic methodologies can be classified as AI, and while important, are only the foundation of what’s truly required for intelligent machines.
Machine learning, or ML, is a subset of AI builds on Basic AI in that it references the study of mathematical algorithms and statistical models used use perform a given task or function without the provision of specific instructions or actions. ML relies on its comprehension of data provided to recognize patterns and uses algorithms to construct predictive mathematical models. This allows a machine to make forecasts or generate decisions without being specifically programmed to do so.
Deep Learning, or DL, is centered on multi-phase, multi-layer approaches to optimize the ML algorithms used to interpret the machine’s environment and make accurate decisions. Multiple layers are each parameterized separately, and lower-level models used to generate predictions for higher-level models.
Reinforcement learning, or RL, is a further expansion of ML where an intelligent machine takes appropriate actions to maximize reward in a particular situation. RL creates an adversarial environment where multiple analytic methodologies exist in parallel. In this predictive intelligence arena, the optimal analytic methodology wins and receives a higher weighting within the iteration of analytic methodologies. As such, an intelligent machine using RL is bound to learn from its experience and constantly adjust its actions to achieve the best possible outcome.
Complete Intelligence Process – CI Futures
Enterprise cost (spend) and revenue forecasting present unique challenges that extend beyond traditional regression analysis. Complete Intelligence incorporates our extensive global economic and markets database with AI and machine learning algorithms for progressively increasing intelligence and more confident planning decisions.
We have designed our integrated global model to ensure that actions in one market, country or sector of the economy are reflected elsewhere in markets, industries and the global economy. These functions include, but are not limited to the following data:
- International Trade
- Economic Indicators
- Commodity Prices
- Equity Indices
Data are sourced from national statistical agencies, multilateral banks, multilateral government bodies, and other publicly available sources. Our data frame includes historical data going back to 2010.
We begin our process with the screening, repair and validation of data with multi-layered, multi-dimensional analyses to understand how data should act to create predictive intelligence scenarios for prices, costs, budgets and revenues. We test economic, industry, market and endogenous factors to compare behaviors of historical data sets within our proprietary artificial intelligence (continuous learning) model.
We then employ our ensemble methodology, utilizing ~15 billion data items and testing more than 1 million functions against the historical data series in each forecasted line item. Our platform generates an environment that is a proxy for the global economy, industries, and market events (currency movements, commodity prices, etc.).
The ensemble methodology utilizes more than 30 different approaches to account for both fundamental and technical movement of an asset/element. Thousands of scenarios are run to identify optimal error over the forecast window. Accuracy is accounted for in the process with each successive iteration, resulting in robust forecasts for each individual line item, responding to market context and improving performance over time.