The roots of Machine Learning
Throughout our history as a species we have been inspired many times by nature to carry out technological developments. Artificial Intelligence, a field that has recently gained a lot of traction in the world of research, allows us to solve problems such as predicting the weather tomorrow or detecting objects of interest in images. Despite its complexity, Artificial Intelligence is essentially nothing more than that: an attempt to replicate what we all possess and know as intelligence. So much so that it is often built in the image and likeness of what it tries to imitate.
Finding inspiration in nature
The field of artificial intelligence is very broad and diverse. Throughout its history, it has been necessary to solve a great variety of problems with a great variety of methods and algorithms. More often than not these algorithms are very clearly inspired by some behavior that we can find in living beings or in mechanisms existing in nature.
Swarm or collective intelligence
We have all seen the shapes formed by a group of birds in the sky or a school of fish in the sea, and we have marveled at the coordination they demonstrate. This is the result of what we know as collective intelligence, which we have used as inspiration in the field of artificial intelligence. We find collective intelligence when we have a group of simple agents that interact with their surroundings following simple rules. These agents act independently since there is no centralized intelligence that controls them, but even so the grouping of their behaviors leads to the emergence of a complex collective intelligence.
Multi-agent systems are an example of how the field of artificial intelligence has been inspired by the behavior of animals such as birds or fishes, or microorganisms such as bacteria. They use the idea of collective intelligence to solve complex problems through the interaction between simple agents. Another very different example are Ant Colony algorithms which, as their name suggests, are inspired by the behavior of ants when looking for food. Initially, the ants explore different possible paths and mark them by dropping pheromones. Over time, the optimal path ends up containing more pheromones and all ants follow it. The algorithm in question finds the shortest path to a target following a very similar methodology.
There is a set of algorithms that are inspired by the field of psychology, more specifically in how we usually learn from trial and error or from interactions with our environment. The most obvious example of this is how we usually teach children or train our pets. If we give a dog a reward every time it acts the way we want it to, it will end up associating the behavior with something good and acting that way. This is an example of positive reinforcement, as opposed to negative reinforcement that is based on penalizing unwanted behaviour.
Reinforcement Learning within Artificial Intelligence tries to imitate these behaviors, using rewards and penalties to guide the way an agent acts. This ends up perfecting what is known as the policy, the set of actions to be taken in different situations, in order to achieve a goal. A well-known example of Reinforcement Learning is that of DeepMind 1, a company that developed an algorithm capable of winning the world champion of the strategy board game “Go”.
One of the most popularly known lines of Artificial Intelligence is the one inspired by the process of evolution and natural selection. The algorithms that arise from it are known as genetic algorithms, which deal with the problem of optimization by simulating a population of individuals that compete with each other.
Example of a genetic algorithm, its population (left) and reproduction (right) 2
Each individual is identified by a series of information (equivalent to their DNA, a succession of genes). The algorithm goes through a selection phase in which individuals are evaluated and only the best stay alive (natural selection), a crossover phase in which pairs of individuals generate offspring that contain characteristics of both parents (reproduction), and a mutation phase by which some genes of the offspring are randomly modified. By iterating through this process, the individuals of the population will improve themselves to fulfill their objective.
Imitating the brain
So far, we have seen how Artificial Intelligence has been inspired by mechanisms in nature and the behavior of living beings. However, if we want to imitate human intelligence, why not directly imitate what is responsible for it?
Comparison between a real and an artificial neuron 3
Having its concept born in 1958, Artificial Neural Networks have revolutionized the field of artificial intelligence. As their name suggests, they try to imitate the human brain at the cellular level. In our brain, neurons use dendrites and axons to transmit information from one to another through different levels of abstraction, in order to model complex problems. In Artificial Neural Networks, the basic unit (also known as a neuron) receives a series of inputs on which it operates, and transmits the result to other neurons. Neurons are also organized by levels, and their specific arrangement varies depending on the problem to be modeled. These structures are trained for tasks that humans can perform such as retrieving information from text.
Challenges at IOMED
One of the most important applications of Artificial Intelligence today is Natural Language Processing (NLP), which consists in processing data in an unstructured format such as written texts. This research field is especially useful for the clinical world since it contains large amounts of information in this state that are not being used. After all the advances that have been made in Artificial Intelligence which is so deeply rooted in nature, we can use them for the good of a discipline so connected to biology as medicine. At IOMED we want to push existing technology to the limit in order to make clinical information more accessible and, who knows, maybe innovate again looking for inspiration at the origins of everything.
 Deep Mind; (June 17, 2016). Deep Reinforcement Learning. Retrieved from: https://deepmind.com/blog/article/deep-reinforcement-learning
 Firing Neurons; (July 29, 2019). AI — Taking inspiration from nature. Retrieved from: https://medium.com/firing-neurons/ai-taking-inspiration-from-nature-fea8a690f50f
 Towards Data Science; (September 4, 2018). The differences between Artificial and Biological Neural Networks. Retrieved from: https://towardsdatascience.com/the-differences-between-artificial-and-biological-neural-networks-a8b46db828b7