The Azaries SPARK platform.

Human Artificial Intelligence (HAI), a World First Technology in the UK.

Azaries Spark, augments, improves and adds value to AI and offers a 'no code' entry to the AI world discovering patterns in your data to improve decisions and knowledge.        

Improving Conventional AI Methods with 'Inductive Logic Programming' (ILP).

By incorporating ILP into conventional AI approaches like neural networks and statistical models, their capabilities can be significantly enhanced through the addition of symbolic thinking and interpretability.

Neural networks frequently receive criticism for their opaqueness, as the reasoning behind their judgements can be obscure. ILP overcomes this constraint by enabling the retrieval of symbolic rules from neural networks.

ILP can incorporate structured priors or restrictions, resulting in models that are more precise and applicable across different scenarios. The combination of human experience and machine learning boosts the ability of artificial intelligence to address intricate, practical problems that require both statistical analysis and logical reasoning skills.

ILP serves as a vital connection between symbolic and sub-symbolic AI methodologies. It provides a distinct blend of interpretability and strong pattern recognition, successfully closing the divide between human-like reasoning and the efficiency of machine learning.

This integration facilitates the development of advanced, comprehensible, and reliable AI systems in different fields.

What is Inductive Logic Programming (ILP).

HAI improves traditional AI technology by using human involvement.

It starts with a set of examples plus any background knowledge.

SPARK then deciphers which are the most applicable notions, patterns and data.

Finally, it proves which ones are relevant and useful in the context of the application. Then creates rules for predictions.

This process is continuous and improves results using human learning feedback. In addition, it learns from the new data and results ensuring constant improvement.

What is One shot learning?

The SPARK platform excels when learning from minimal examples, metrics and information. An important step forward in high volume, fast moving data. It pinpoints immediate improvements.

'Meta-Interpretive Learning' (MIL) embedded in SPARK learns accurate models from as few as one or two examples.

When compared to traditional methods, for example, large language models and generative AI, which require a significantly higher number of examples for realistic performance, SPARK excels.

The net improvement of MIL reduced required training examples by up to 90% while maintaining or even improving accuracy.

The Details:

Accuracy: In tasks like string transformation and game strategies, MIL achieved accuracies of 95-100% with minimal examples.

Generalisation: Demonstrated strong generalisation, effectively handling unseen cases with immediate logical rules.

The Benefit: Progressive for applications where data collection is fast moving, expensive or impractical.