Last week on our webinar, we had three experts – Risto Miikkulainen, Babak Hodjat, and Amir Banifatemi — in the field of what’s called evolutionary AI who painted a vivid picture of the innovative technology and its potential to reshape decision making across industries.
Miikkulainen, VP of AI research at Cognizant and professor of computer science at the University of Texas at Austin, and a pioneer in the field of evolutionary AI, explains that evolutionary AI takes its cues from biology, much like neural networks. However, unlike most machine learning techniques that rely on supervised learning with clearly defined targets, evolutionary AI thrives in scenarios with weaker feedback. It can navigate complex problem spaces without predetermined answers.
While traditional AI might excel at recognising patterns – like identifying a dog in an image – Miikkulainen says stopping at just prediction is akin to “leaving money on the table”. Instead, evolutionary AI goes a step further: it doesn’t just predict, it suggests actions. By simulating a population of potential solutions and allowing them to “evolve” towards better outcomes, evolutionary AI can uncover novel strategies that even human experts might overlook. As Hodjat, CTO of AI at Cognizant, puts it, evolutionary AI is more aligned with the way we go about making decisions. In the real world, decisions often involve balancing multiple, sometimes conflicting objectives. Evolutionary AI excels here, says Hodjat, offering a range of strategies that balance various outcomes, allowing decision-makers to choose the most suitable approach.
“Look at the pandemic, you want to reduce the number of cases, but at the same time you don’t want to have a major economic impact on people’s livelihoods. Evolutionary AI is very good at coming up with strategies that give us a balance of multiple outcomes and lets us then choose which balance of outcomes is more interesting to us.”
Hodjat describes a two-step process: first, using AI to create “predictors” that forecast outcomes based on different decisions. Then, evolutionary algorithms search the vast space of possible strategies to find optimal solutions. This approach, he argues, mirrors human decision making processes, allowing for scenario testing and collaborative refinement between AI and human experts.
Banifatemi, technology and innovation strategist for AI Commons, a nonprofit that seeks to bring the benefits of AI to everyone, high lights evolutionary AI’s creative potential. Unlike traditional AI that derives solutions from existing knowledge, evolutionary algorithms, he says, can create entirely new solutions. This creativity, coupled with the ability to optimise for multiple objectives simultaneously, makes evolutionary AI a powerful tool for tackling complex, multifaceted problems.
Banifatemi also points out a crucial advantage: transparency. While many AI models operate as “black boxes,” the solutions generated by evolutionary AI are often more explainable. This transparency is vital in fields where understanding the reasoning behind a decision is as important as the decision itself.
Real-world use cases
Miikkulainen describes an agricultural experiment where evolutionary AI discovered that basil thrives under 24-hour light – a counterintuitive finding that challenged biologists’ assumptions. The experiment involved planting about 300 basil plants and providing them with different amounts of light, water, nutrients, temperature, etc. “Most of them were recipes that the biologists came up with – the usual suspects. But we also had some exploratory examples. And so, we now had 300 points about how the plant grows under different conditions. And then we built a predictive model. We could then come up with crazy recipes and ask the model what would happen in different scenarios. It so happened that we had initially set up a limitation of six hours of darkness. But then we said, let’s open it up to try other solutions. And to the biologists’ big surprise, the model found basil thrives when lights were on 24 hours. This is the kind of solution that will be difficult for humans to discover because they have all kinds of assumptions and evolutionary AI does not,” Miikkulainen says.
Banifatemi says evolutionary AI is also showing a lot of promise in pharmaceutical research, particularly for rare diseases where traditional approaches fall short. “For rare diseases, from a pharmaceutical perspective, a number of drug candidates need to be identified. And while the cost of drug research is huge, the market is not that large. With evolutionary AI you can find a cluster of potential molecules for a given set of constraints: like for toxicity, or for a particular genetic background. This accelerates drug research,” he says.
The story of evolutionary AI is still unfolding. As researchers and companies like Cognizant continue to invest in and develop this technology, more innovative applications will emerge – and some could come from Cognizant’s India offices. “Cognizant has a very strong presence in India and we have colleagues in India that are well versed in using evolutionary AI,” says Hodjat.
LINK |
HOME