Introduction

When thinking about Artificial Intelligence (AI), what most people in the field have in mind is machine learning models and heuristics to solve problems that are complex enough that their solution cannot be written using an exact set of steps but should be learned from data or from interaction with the environment. In this narrow view, AI is a collection of procedures that can be seen as a smart extension of functions (in classical computer programs) that we can employ to make our life easier. However, contemplating the term “artificial intelligence” one can appreciate that this term holds a deeper meaning. For instance, it may refer to Intelligent creatures or subjects that live in some universe and want to survive using their intelligence and skills. Depending on the universe rules and physics they live in, these creatures may develop complicated skills such as collaboration and communication.

In the last few years, with the emerging interest in deep learning and reinforcement learning, there has been a great effort to develop environments that can be used to demonstrate the ability of AI. Nevertheless, most of these simulations are a game environment, such as OpenAI gym, deepmind lab, OpenAI Universe, and more. However, the limitation of game environments is that the actions of the agents are limited and its goal is too strict. While it is nice to show how artificial agent can master games, it is more interesting to see how artificial creatures can develop complex behavior, social behavior, and survival skills.

In addition to the thrill of developing intelligent creators, project-origin allows you to observe the behavior and destiny of these creatures and maybe learn something about ourselves. Answering these questions, by demonstrating how complicated skills can be developed by artificial agents in a simulation would shed some light and take us closer toward understanding the mechanics and the true nature of our intelligence. It may even reveal some insights about more abstract nature and skills that animal and humans possess, like socializing, communication and even love.

To make these ideas more tangible consider the following examples:

• Assume that the physics of the simulation dictates that two creatures should be in a specific place and perform a specific operation simultaneously in order to both get reward, would they learn to do so?
• If doing an action in a specific age or a specific time cycle of the universe, could result in a great reward, would make them “wait” to this age to do the action?

Scientific Motivation

project-origin can be employed to get us closer to answering intriguing questions on intelligence, on the behavior of intelligent creatures and maybe help us ** hypothesize on the reason underlying nature rules**. In the intelligence realm, first, it is a realistic environment for training intelligent creatures with various skills. Second, it may provide us with a deeper understanding of the development of complex skills from basic actions. Third, it can show us how intelligence and evolution, are two different mechanisms that affect each other and “cooperate” to build intelligence. In biology complex behavior is explained by evolutionary concepts such as nature selection, however, in intelligence studies, behaviour is explained by the concept of promoting individual current and future reward. However, we think that studying both of these aspects together and the affect one has on the other, may make our understanding much deeper. Intriguing and counter intuitive behaviour may be explained only when both nature and nurture are considered.

Below are several example questions embodying these aspects:

• Can we create intelligent creatures that can develop survival skills such as collaboration and communication? Namely, being put in a challenging environment, can the state of the art AI, using trial and error and evolution, learn how to survive?
• Can they learn to look for food?
• Can it figure out complex behavior, from the procedural and social viewpoints, if that is required to their survival?
• Would they learn using their actions (moving, mating, fighting and producing sound) to flourish?
• Would counter intuitive behavior like altruism, like parenting would develop? What environment (physics) would promote such behaviour?
• Would they “understand” the effect of these actions on their own survival and on the environment?
• Would they develop non-obvious behavior showing their understanding of the effect of time and age?
• How fast can creatures adapt to changes in the environment? Such change may include epidemic, dearth, introduction of enemies, poison and change in the enemy intelligence/fit?

In addition, this project can be used as a tool by researchers in different disciples of science to give possible explanations to a wide range of phenomena in nature and help us reveal the essentiality of specific rules. For instance, consider the question of the need for minimal reproduction age in organisms biology. One answer may be that this mechanism ensures that ancestors are physically and mentally mature enough to be able to take care of their offsprings. However, this is an insufficient explanation when it comes to germs or other organisms that do not take care of their offsprings but their biology implements this mechanism. Using the simulation we noted that if no maturity age is implemented in the universe, a population explosion consisting of stupid creatures may happen. This statistically can happen because natural selection can push them to reproduce frequently and immediately after birth, allowing their race to flourish without showing any tracts of intelligent behavior. However, this phenomenon rarely happens in universes enforcing maturity age. This suggests that the minimal reproduction age is implemented in nature not only to make sure that ancestors are mature enough to be able to take care of their offsprings but also that the maturity age is an essential element of the developing intelligent and fit creatures. Namely, one should prove that it has some survival abilities before passing down its genes. This is a nice example showing how natural selection and intelligence are basically two mechanisms that work together to produce fit creatures.

Talking with Dr. Joseph Fanous (a.k.a “iljoe”) - a pharmaceutical scientist and a close friend of mine; he mentioned that such simulation, not only can help giving alternative an interesting answers to long standing questions in biology but can also assist answering questions raised by contemporary researches. For instance in his field of study - bacterial sciences and antibiotics development, such simulation can help giving answers about the underlying mechanisms of bacterial developments including:

• Why germs which develop more slowly show better survivals skills than their fast growing counterparts?
• How persistors develop? Persistors are germ individuals that develop superior survival abilities in a hostile environment (e.g., an environment with antibiotics).
• Explain why pathogens behave differently in lab bacterial culture vs. in organismic hosts? How they learn to use different aspects of the environment to survive when they are in environments with different hostile agents. For instance, in lab bacterial culture the germs mostly eat the glucose provided in the Petric plate but in the organic host, they eat feces of their friends.

Widening the horizons of Intelligence research

In the current implementation of project-origin, there are aspects that in other game environment is not possible or not relevant, for instance, the combination of Evolution and Learning for development and survival in the race. In the current implementation of project-origin DNA can affect some biological aspects of the creature, for example, changing its sight range. While all implementations of agents playing games, has a constant input state size, in project-origin different policy graphs can be implemented with a different number of input parameters or even different inner structure of state input or action set output and parameters. In addition, the neural network train and improve using backpropagation during the lifetime of the creature. This allows investigating how both evolution and intelligence contribute to the creation of highly skilled creatures. See Nature versus nurture.