Learning to Play the Game of Evolution

Cancer and antimicrobial resistance are some of the biggest challenges in medicine that remains far from being solved. One possible reason is that they are fundamentally dynamic problems; it’s a moving target. Our current approaches to solving these problems have always been static, new antibiotics, new cancer treatments. These static solution do work, but they don’t always work, and even if they do, there’s much doubt over whether they will continue to work in the future.

The answer is of course simple: adopt dynamic solutions instead. Certainly, sounds easier said than done. Finding a dynamic solution for cancer of antibiotic resistance is certainty an unsurmountable task. But even so, we should start by framing our mindset to think dynamically.

Unfortunately, many of ‘next-generation’ solutions at the frontiers of medical research is still very static. One example is ‘next-generation’ antibiotic resistance treatments. Rather than looking for a more potent antibiotic, there are scientists looking at using chemicals, electromagnetic waves to destroy biofilms and hence expose the bacteria to be attacked by existing antibiotics. While this is an intriguing and ‘outside-of-the-box’ solution, it still suffers from the unfortunate characteristic of being static. Even if we are able to find a way to destroy biofilms, we can expect the process of evolution and natural selection to give rise to another bacteria that is capable of utilizing alternate methods of protecting itself. Unfortunately, brute force solution finding via random mutations and natural selection is much more efficient than scientists tinkering carefully and deliberately to execute experiments. Applying a single direction selection pressure naturally give rise to greater degrees of natural selection and evolution. These solutions will inevitably invite newer and harder problems to be created.

Instead, we should look to more dynamic solutions. One area of particular interest is the use of probiotic solutions to defend against pathogenic bacteria. This would constitute as a dynamic solution if we could ensure that these probiotic bacteria are able to evolve at a rate faster than that of pathogenic bacteria and yet not become pathogenic towards humanity. A key characteristic in this case is that the selection pressure is applied to the solution and not to the problem. This is a key feature of a dynamic solution. In using probiotic bacteria, we are looking at two main selection pressures:

Select for bacteria that is faster at evolving to kill pathogenic bacteria than pathogenic bacteria can evolve to kill it Select for bacteria that is good at forming a symbiotic relationship with its human host The presence of pathogenic bacteria naturally creates a selection pressure on our probiotic bacteria. This naturally makes our probiotic bacteria more potent and adaptable given the pathogenic bacteria it encounters. This means that even if the pathogenic bacteria evolves with evasive mechanisms, the probiotic bacteria will have to likewise evolve to fight for survival.

The need for nutrients and survival and perhaps the activation of the immune system may create a selection pressure to not damage the human host or to illicit an immune response while still benefiting from the nutrients provided by the host.

The dynamic interplay between selection pressures on the solution creates two constraints in which we optimize for to get the desired result.

In fact this is not too different from the methodologies used in machine learning. Machine learning and AI are dynamic solutions, a big contrast to software of the past which is use-case specific. With machine learning and AI, we realise that we can solve much harder and difficult problems — and in particular we can solve dynamic problems. Digging deeper, we realise that evolution and natural selections are simply just another optimization algorithm, no different from machine learning algorithm that optimizes for human defined parameters. Natural selection optimizes for survival and all the facets that are required for survival. One might say this is in fact and extremely sophisticated optimization function with endless amounts of parameters. Unfortunately, the downfall of evolution is that it’s method of ‘improving’ the next generation is horribly inefficient — namely randomly mutating the genome and hoping that a better organism will result. And for this reason evolution takes a very long time over many many generations for a visible difference to be seen.

We are at the point of human civilization where the problems we face are insurmountable for existing machines to solve. We look for superior machines, machines that are dynamic, adaptable and always evolving. Only time will tell what sorts of machines we will create in the future — human engineered super-bacteria? human engineered super-T-cells? self-reproducing nanorobots? advanced AI bot?

Let’s teach our machines to play the game of evolution.