What makes Artificial Intelligence become a driver of innovation

Generative Artificial Intelligence (AI) like the Large Language Models (LLM) represent novel technology that can be disruptive as well as highly innovative. It is worthwhile therefore to think about the innovation potential of AI in general and observe its evolution carefully. A suitable starting point is the simulative nature of LLM’s. They are models of the use of language based on empirical data taken from existing pools of e.g. text from manifold sources. They typically are applied to generate new text the validity of which is not sure without further acknowledgement. Basic feature of AI so is simulation for the purpose of structural or functional investigation. Not only in generative AI, but also in e.g. the application of digital twins for AI based manufacturing, the evaluation of achieved results is a must. There is good reason to accept that for AI systems of any kind, in order to make them enablers of sustainable innovation. Is that all? Or, putting it differently: how can we best use AI to support human intelligence, and vice versa? And what are promising fields of application?

The rationale of combining human and artificial intelligence builds upon basic limitations and strengths:

Human intelligence is lacking the capacity of utilizing massive data:

• Measurement and collection of large amounts of parameters are not practicable.
• The analysis of large data sets is inefficient.
• Algorithmic thinking is slow.  

Artificial intelligence systems have principle limits:

• The formal systems they are based on cannot completely cope with reality.
• Mental capabilities like understanding, meaning, and insight are not available.
• Parameters and metrics depend on ex ante definition and selection.

The architecture of combining human and artificial intelligence rests upon the utilization of powerful data technology and the creative mind of humans:  

Functions on the AI side would be:

• Collection and analysis of massive data.
• Pattern recognition, reasoning, and generalization.
• Machine learning procedures.

Human functions would be:

• Supervision and control of digitization and model design activities.
• Formalization and validation of assumptions and hypotheses.
• Formulation and proof of propositions. 

The resulting concept of  hybrid intelligence is presented and illustrated  in our recent publication[1]  . Further to the two examples described in this paper, from a general perspective there are diverse potential fields of application in science and engineering, including

• pro-active risk governance in the health sector,
• lifecycle enhancement of operation domain design in autonomous vehicles engineering,
• simulation using digital twins in manufacturing,
• smart electric grid management in the public utilities sector.

____________________

Norbert JASTROCH

eMail norbert.jastroch@metcommunications.de

____________________

[1]Jastroch, N.: Sustainable Artificial Intelligence: In Search of Technological Resilience. Springer 2023. https://doi.org/10.1007/978-3-031-25182-5_31 . PLM2022, Grenoble/France