Solving Complex Challenges With AI That Evolves As It Learns

Jerry Smith

Organizations typically suffer two common constraints when it comes to growing effectively: a lack of resources and flawed decision-making. But when both are present, it can be a bellwether of an extinction-level event. It’s one thing to lack the resources to adequately lead and deliver tomorrow’s business value today. But when sub-optimal decisions are made across the enterprise, that’s a disaster.

But the reality is that humans can actually be very bad at making effective decisions, even at the leadership level. It’s not that we don’t try but that the real world has out-evolved us. Life has become extremely complex and continues to accelerate at an exponential pace that exceeds human capacity and capability.

The gap is widening between the complexity of the world and our problem-solving abilities. With the explosion in data (know-your-customer, products, operations, support, marketing, interconnected things), no human being can keep up, even with the support of advanced analytics and reporting.

Don’t believe this? OK. Right now – this very second – why didn’t that customer buy your product or service? If you don’t know, you should.

The human condition

In addition to being overwhelmed by too much data, humans are blinded by their own biases when making decisions, or we’re tasked to make decisions outside of our individual expertise. I’m not just talking about “comfort zone” here. Our individual viewpoints can skew our decisions to what we believe to be true, regardless of other viable options.

Our decision-making process is also subject to fatigue. When we lack adequate sleep, we have a 1.62 times higher risk of being injured. We make mistakes. We rush to judgment to get things done. These decision-oriented challenges are limited by our human intelligence.

But there is another way, a journey based on leveraging intelligence that scales on the backs of computers: evolutionary computation.

Beyond human limitations of decision-making

Evolutionary computation informs business decisions by providing optimal solutions to incredibly complex problems. Currently, AI is still based on a singular point of view or perspective; it doesn’t take into account other good ideas. It doesn’t know how to create new ideas from older ideas or a history of ideas.

To create new levels of value, AI practitioners must evolve how they marshal the following:

  • Data: The raw bites and bytes or material used to train AI algorithms.
  • Data sciences: What we know about data.
  • Machine learning: How data can continuously teach/steer AI algorithms and influence business change.

Evolutionary computing is like having thousands, or even millions, of people brainstorming, prototyping, and testing ideas. It takes what works and builds out a new version, discarding those with sub-par results. By doing so, it discovers possibilities that lie outside the knowledge of any one person or team of individuals. It is truly evolutionary (pun intended). But not all evolutionary computing is implemented alike.

Differentiators among evolutionary computing systems

Through years of experience, we’ve found that there are three components driving business value through evolutionary computing: causality, prediction, and prescription. All three operate in a loop – a natural, continuous lifecycle within a business.

  • Causality: After we collect a whole lot of data, we need to really understand what is causal within that data. What are the problems? What is influencing engagement and a decision to buy? If data is the debris of human activity, then the goal is to extrapolate causal data to understand what impacts human behaviors that, in turn, affect business outcomes. And, the first part of that involves figuring out and then focusing on those behaviors (reflected in data) that drive true business value.
  • Prediction: Once we have that causal data, the real evolutionary magic happens. We organically create and evolve digital surrogates (simulations) that represent the business objective: evolutionary neural networks. These surrogates think and act like their real-world counterparts – for example, your customers, employees, products, or even a fraudster. These surrogates are developed through techniques such as genetic construction, characteristic mutation, and feature crossover that are applied to massively large populations. The best outcomes are used to create the next generation, and that process repeats itself rapidly.
  • Prescription: Once a digital representation (surrogate) of how business value is generated, we can look for actions that maximize value and generate prescriptive solutions that will literally bend the business performance curve.

Evolutionary computation identifies optimal actions that maximize the business value of the surrogate models. It actually looks through all potential possibilities to find those activities that offer the best chance of achieving our desired objectives. Millions of simulations are constructed and evaluated (e.g., genetics, mutation, and crossover) in order to create an optimized set of actions. These are real, executable activities that can change the behavior of people and generate new data that can be collected – rebooting the evolutionary computing lifecycle all over again. Wash, rinse, and repeat.

Real-world applications

Consider the complex range of compositions of a promising drug that must be evaluated before it’s approved for human consumption. Pharmaceutical companies must test the drug’s effects on patients whose biological makeup is as varied as the human genome. Mapping genomic permutations to optimal treatments is dependent on the examination of thousands of outcomes that are extrapolated from a range of potential therapies. Such trials entail risk and are exceptionally expensive to undertake.

Evolutionary computation can help pharma companies simultaneously evaluate various compounds across genomic variations to identify the most effective treatments before they are ever tested in a clinical trial.

Businesses will increasingly turn to evolutionary computation to make better decisions. And when they do, they’ll make a quantum leap because they’ll be applying an advanced intelligence system to mitigate key limitations in human performance and scale – not to mention the unethical or biased use of AI. When that happens, it will signal a new evolution not only in AI, but also in the way businesses make decisions.

Click here to learn more about evolutionary AI.

This article originally appeared on Digitally Cognizant and is republished by permission.

Cognizant is an SAP global partner.

Jerry Smith

About Jerry Smith

Jerry A. Smith is vice president of Data Sciences at Cognizant. He is a practicing data scientist with a passion for realizing business value through enterprise data sciences services. Prior to Cognizant, Jerry was the North American chief data scientist for Capgemini. He has a PhD, masters, and bachelor of science in computer science, with theoretical and practical experience in artificial intelligence. He can be reached at