Scientific AI

The Emergence of ‘Objective-Driven’ AI

Artificial intelligence offers multiple methodologies for problem-solving, with ‘objective-driven AI’ emerging as a stronger approach for real-time process optimization applications in the Mining industry.

Scientific AI is the name we have given to our unique and innovative implementation of objective-driven AI.

Unlike the exhaustive training requirement of reinforcement learning (RL) or the potential inaccuracies of large language models (LLMs), objective-driven AI bases decisions on a comprehensive understanding of the world. 

Objective-Driven AI platform in action
Neural Network Logo

Enabling Human-like Planning & Reasoning

Why this New Approach is so Important for Industrial Process Control

Unique to this approach is its adaptability; objectives can be adjusted in real-time without the need for exhaustive model retraining with new data.

In dynamic environments like mining operations, objective-driven AI can mitigate some of the pitfalls of other modern AI approaches and enhance efficiency and safety by offering a combination of first principles and data-driven, flexible decision-making that closely resembles human planning and evaluation.

Understanding Objective-Driven (or Scientific) AI

Objective-driven AI operates on a simple premise: give the system a set of objectives and let it decide the best way to achieve them based on a model of the world.

The key components of this system include:

  1. Perception (a data-driven understanding of the operating environment)
  2. An Actor (proposing possible actions)
  3. A World Model (AI’s understanding of how the world works based on self-supervised learning)
  4. A Critic (evaluating potential actions against set objectives)
Objective Driven AI - An Overview

Objective-driven AI brings a nuanced approach to machine decision-making, echoing human-like reasoning. Through understanding, action proposal, and meticulous outcome evaluation, it bypasses the extensive training of RL and the potential pitfalls of LLMs. 

Objective-Driven (Scientific AI) in Action

Thickener Optimization

Within the realm of plant operations, the real-time Thickener Optimizer Application exemplifies the capabilities of the objective-driven AI paradigm.

At the core of the application is the world model which consists of a Bayesian deep neural network supported by virtual sensors based on first principle calculations. The thickener optimizer proposes actions in the form of future control settings, and it evaluates the predicted outcomes of those actions, considering multiple aspects and guard rails of the thickener operation.

The simplified flow diagram below visualizes the thickener optimization process.

Objective Driven AI - Thickener Example
  1. Raw measurements are encoded into the perceived state of the thickener by the perception module
  2. The actor module suggests possible action sequences, which, in this case, correspond to control values that could be applied over the near future
  3. The world model predicts the future states when following the suggested action sequences
  4. The critic module evaluates the predicted outcomes and selects the one with the largest expected benefits

This method, while specific to thickener control, illustrates the foundational principles of objective-driven AI: a model discerning its environment and optimizing actions in alignment with defined objectives. The integration of an explainable and real-time configurable optimization method in the thickener application furthermore reinforces the transparency often advocated for in comprehensive objective-driven AI systems.

While the prediction horizon is typically one hour, the whole optimization process is repeated every single minute to keep replanning for the best course of action in a changing environment. This makes objective-driven AI systems truly explainable and trustworthy. 


Objective-driven AI brings a nuanced approach to machine decision-making, echoing human-like reasoning.

Going back to our example, in the realm of industrial process control, mining, and the specific application of thickeners, objective-driven AI stands out in its efficiency, safety, and adaptability.

Its design enables operations to evolve seamlessly, embracing changes in objectives, and ensuring decisions are rooted in a comprehensive understanding of the world.

Discover how Scientific AI Optimization can be applied to your operation

The Application Portfolio

Our process optimization apps can be deployed on a specific process bottleneck or expanded across the entire value chain.

They are powered by our Scientific AI Decision Intelligence Platform,

Stockpile and Inventory Optimization

Stockpile & Inventory Optimization

Grinding Optimization App

Grinding Optimization

Thickener Circuit Optimization

Thickener Optimization

Flotation Circuit

Flotation Optimization

Solvent Extraction Optimization App

Solvent Extraction (SX) Optimization

Leaching Optimization App

Leaching Optimization