Can machine intelligence evolve like organic beings, and what are the implications?

Evolutionary computation delves into this domain by “growing” artificial intelligence through a process known as evolutionary optimisation (EO): simulating natural selection and mutation as seen in biological evolution. In other words, EO optimises algorithms through removing undesirable traits and introducing a few random changes in every generation, allowing the algorithm to grow gradually more fit – that is, smarter.

One of the foremost authorities on the topic is Dr Risto Miikkulainen, Professor of Computer Science at the University of Texas in Austin and Vice President of Research at Sentient Technologies, a company that combines multiple AI disciplines from deep learning to evolutionary optimisation and neuroevolution. Dr Miikkulainen gave a demo and a round table talk for the Algorithmic Inequality track of 2018 SHIFT Business Festival.

Sentient Technologies’ approach is different from most AI today: while typically the process involves collecting big data and modelling it to predict the future, evolutionary optimisation trains AI to solve specific problems in situations where the best solution has not been identified. The right answer is unknown, but there is guidance in the form of a reinforcement signal, or a fitness function. “Evolution is a mechanism for trying to find really good solutions”, Miikkulainen sums up.

As an example of the many uses of EO, let’s look at digital marketing and website design. Designing websites that allow companies to track, predict and influence potential and existing customers is something that everyone doing business is faced with today, and EO could revolutionise the way it is done.

Applying EO to website optimisation can mean improving simple components, like making the sign-in page appealing, but it can also be extended to the entire funnel, optimising the customer journey from an email campaign to a website visit, online shopping and subsequent remarketing. “From the point of view of the method and the algorithm, it can be expanded to the entire experience”, Miikkulainen says.

The uses of evolution in computation are not limited to hard business. Researchers in many fields stand to benefit from modelling environments where theories based on observation and deduction can be put to the test, virtually speaking. “With computational simulations, we can test scenarios and see what happens if we manipulate some of the variables in the environment”, Miikkulainen explains.

In one test, which sought to optimise the growth of basil, EO yielded some interesting new information: “Basil doesn’t need to sleep – require 6 hours of darkness – like the plant biologists thought it would. Instead, EO quickly pushed the amount of light to 24 hours a day.” Sentient Technologies has also used EO to model the complex behaviour and interplay of different emotions seen in hyenas working as a pack to steal a kill from a lion. From business to plants and animal behaviour, evolutionary optimisation can serve as a powerful tool for learning as well as decision-making.

Algorithm-driven decision-making is also known to have its dark side, like we heard from Dr Cathy O’Neil at SHIFT, and when talking about targeted marketing and influencing customer behaviour, the conversation soon turns to ethics. Very often, we are persuaded into making decisions as customers based on the appeal of the ad rather than any quality of the product or service itself.

For example, ever wonder why ads come in bright colours? According to former Google Design Ethicist Tristan Harris, interviewed by Sam Harris, the human brain is hard-wired to like bright colours, because they remind us of ripe fruit. As a result, we may find ourselves drawn to a product simply because we are subconsciously drawn to the well-chosen colour scheme of the ad. With EO harnessed to take our advertising tricks beyond the imagination of our most skilled human experts, will advertisers soon be able to blindside even the most well-informed consumers?

Not any time soon, Miikkulainen thinks, because humans are nothing if not adaptable – and easily bored. When a once-exciting thing becomes ubiquitous, its appeal is weakened. “What we’re doing now, like the bright colours, might not work a few years from now”, Miikkulainen explains.

And this leads us to the real beauty of it all: “However, it’s possible that even then, the same methodologies of finding what works can be used to find something else”, Miikkulainen concludes. So when black becomes the new orange again, chances are the idea didn’t originate in a human brain.