October 19, 2022 - 728 views|
Companies are increasingly turning to AI technologies—such as machine learning and evolutionary AI—to quickly find creative, unprecedented ways to address complex problems.
Any football fan is familiar with the concept of “a good decision.” One good decision, made in a split second, can win or lose a match. In an instant, players need to read the opposition, predict what will happen, decide on the best action and execute the play.
The same can be said for businesses, which need to make fast decisions in a complex, rapidly-moving environment. Researchers strive to find the most effective cancer treatments; traders develop strategies for financial quantitative trading to maximize returns and minimize risk; botanists seek cyber-agriculture techniques to optimize flavor and increase yield from crops.
Unfortunately, the human mind is not wired for this level of complexity. While psychologists often categorize decisions as being “rational” (based on reason) or “irrational,” people frame their decisions differently, based on their values, preferences and beliefs. Human decision makers are influenced by contextual and situational variables and their own experience. As social beings, they are also vulnerable to desiring approval. Taken together, these factors hinder the human ability to make fast, logical decisions.
However, one good decision that companies are increasingly making is to use AI technologies—such as machine learning and Evolutionary AI—to develop creative, unprecedented ways to address complex problems in a fraction of the time it would take without them.
With a rapidly evolving business environment, it is clear that AI can be used for much more than just automating existing human activity. It can help organizations make far better decisions and address increasingly complex issues in much less time.
Consider decisions that had to be made at the dawn of the COVID-19 crisis. With no vaccinations or therapies available and limited data to guide their efforts, policy makers had to make rapid decisions to protect the public. Varying nonpharmaceutical interventions (NPIs) were implemented at different stages, in different ways and in different contexts. There were no “best practices” to be applied.
To fill that gap, we worked on turning computational modeling using Evolutionary AI into a crucial tool for making informed decisions on preventing the spread of the disease, as well as how to restart the economy.
Previously, most modeling efforts had been based on traditional epidemiological methods to predict the spread of the disease. These models can be accurate and helpful only if key parameters, such as the average number of infectious disease cases transmitted by one infected individual, can be estimated accurately. These models also often neglect to incorporate the decisions made over time to address the pandemic.
Fortunately, as the virus spread, test results, as well as local government and clinical intervention policy data was collected all around the world. We were able to leverage this data, along with actions taken in different regions at different times, to build machine learning models to accurately predict disease spread in the presence of various mitigation policies.
Next, the models moved from prediction to prescription by analyzing which NPI strategies would be most effective and in what context. This was done by creating multiple simulations to not only estimate the disease spread and economic impact, but to also understand the associated trade-offs for each strategy.
As the model “evolved,” new variables were introduced to create a Pareto frontier, an approach that helps determine the optimal combination of the key variables (disease spread and economic impact) without making either worse off.
The demonstrated capabilities of these AI-driven models point to promising applications for business. Through a machine learning technology called “evolutionary surrogate-assisted prescription (ESP),” very large numbers of candidate strategies can be generated and evaluated for any particular scenario. The power of these models is that they are not constrained by past methods or biased by preconceived notions. Like natural selection itself, Evolutionary AI introduces creative, and unanticipated, changes that enable decision makers to assess outcomes that would otherwise not be considered.
We’ve used ESP for several real-world design and decision optimizations. The Massachusetts Institute of Technology was interested in discovering growth recipes for agriculture—basil in particular. However, in a completely controlled hydroponic growth chamber, MIT lacked the historical data to build a sustainable agriculture program. We helped them develop a program based on our Implement/Observe/Model/Suggest approach:
To adopt AI-driven decision making, companies must be willing to let go of time-worn approaches and instead address their core challenges. Such challenges could include continuing legacy issues of siloed data that impede meaningful analysis and action, or accounting for how these actions impact the larger enterprise and trading partners.
But overcoming these challenges is essential. Evolution is a concept that has been successfully optimizing against multiple objectives for four billion years. The ability to apply this approach to decisions is now within businesses’ grasp.
By using evolutionary AI, businesses can build the resilience and agility needed to respond to today’s fast-changing business, regulatory and environmental dynamics, as well as the shifts and changes yet to come.