Photo by the editor
# Understanding just-in-time modeling of the world
This article reviews and summarizes a recently published article entitled “Just in Time” World Modeling Supports Human Planning and Reasoning, which is available to read in full at arXiv.
Using a softer and more accessible tone for a broader audience, we will discuss what simulation-based reasoning is, describe the general just-in-time (JIT) framework presented in the paper with a focus on the orchestration of the mechanisms it uses, and summarize how it behaves and helps improve predictions in the context of supporting human planning and reasoning.
# Understanding simulation-based reasoning
Imagine that you are in the farthest corner of a gloomy, filthy room full of obstacles and you want to determine the exact path to reach the door without collisions. At the same time, suppose you are about to hit a billiard ball and imagine the exact trajectory the ball will follow. These two situations have one thing in common: the ability to project a future situation in your mind without taking any action. This is the so-called simulation-based reasoningand sophisticated AI agents need this skill in a variety of situations.
Simulation-based reasoning is a cognitive tool that we humans constantly employ to make decisions, plan routes, and predict what will happen next in our environment. However, the real world is absurdly sophisticated and full of nuance and detail. Trying to exhaustively calculate all possible eventualities and their consequences can quickly drain our mental resources in a matter of milliseconds. To avoid this, from a biological point of view, we do not create an almost perfect photographic copy of reality, but generate a simplified representation that contains only the truly relevant information.
The scientific community is still trying to answer a central question: How does our brain so quickly and efficiently decide which details to include and which to omit in this mental simulation? This question motivates the JIT framework presented in the target study.
# Study of basic mechanisms
To answer the previously posed question, the researchers present an inventive JIT framework which, unlike classic theories that assume full observability of the environment before planning, proposes building a mental map on an ongoing basis, collecting information only when it is really necessary.
JIT framework proposed in the paper and applied to the navigation problem | Source: Here
The greatest achievement of this model is the way it defines the combination and interconnection of three key mechanisms:
- Simulation: It is based on the principle that our mind begins to plan in advance the course of action or the route we will follow.
- Visual search: As the mental simulation progresses towards the unknown, it sends a signal to our eyes (or perceptions, in the case of agents or AI systems) to explore that particular part of the physical (or digital) environment.
- Representation modification: When an object is detected that could thwart our plan, such as an obstacle, the mind immediately “encodes” this object and adds it to its mental model to take it into account.
In practice, it is a quick and fluid cycle: the brain simulates to a modest degree, then the “eyes” look for obstacles, the mind updates the information and the simulation continues – all in a precisely orchestrated way.
# Framing behavior and its impact on decision making
What is the most fascinating aspect of the JIT model presented in the article? It probably is stunningly productive. The authors tested this by comparing human behavior with computational simulations in two experiments: maze navigation and physical prediction tests, such as guessing where a ball will bounce.
The results showed that the JIT system stores a much smaller number of objects in memory than systems that try to exhaustively process the entire environment from the beginning. However, despite working from a fragmented mental image that captures only a compact part of the full reality, the framework is capable of making high-quality, informed decisions. A profound conclusion can be drawn from this: our mind improves its efficiency and reaction speed not by processing more data, but by being incredibly selective and achieving reliable predictions without excessive cognitive effort.
# Considering future directions
While the JIT framework presented in the study provides an excellent explanation of how humans plan (with the potential implications of pushing the boundaries of AI systems), there are some horizons still to be explored. The study’s trials only considered largely unchanging environments. Therefore, the extension of this model should also take into account very active and even disordered scenarios. Understanding how relevant information is selected when many non-static objects coexist around us may be the next huge challenge for further progress in this fascinating theory of human planning and reasoning and – who knows! — translating this to the world of artificial intelligence.
Ivan Palomares Carrascosa is a thought leader, writer, speaker and advisor in the fields of Artificial Intelligence, Machine Learning, Deep Learning and LLM. Trains and advises others on the employ of artificial intelligence in the real world.