Eureka! 💡

The Microsoft Research Lab in Cambridge, England (Microsoft Research Lab – Cambridge, 2023) has teams that focus on expanding the capability and efficiency of machine intelligence. As such, they have examples of several projects ranging from the code to drive the AI of Non-Playable Characters (NPC) in video games that collaborate with humans via Project Malmo or Project Paidia, to ways to find your best match mate for an online game with TrueMatch, to the code to build your own machine learning adaption for whatever datasets you may have from single projects up through enterprise-scale using the infer.net framework available on GitHub.

While each of these example have different end goals, digging into what makes them tick I start to see common references to the hierarchy of AI.

Machine intelligence (MI) is an umbrella term for artificial intelligence (AI) and tangential technologies according to Shield.ai (Barngrover, 2018). Shield AI uses MI to control autonomous aerial vehicles (AAV) for defensive purposes, but the theory and the technology behind the AI are the same regardless of the application. Just like our brains work the same way whether we are getting dressed in the morning or driving to a restaurant, we simply change the game plan and related variables and let the intelligence of our brain do the rest.

Having this universal architecture is, in my opinion, the key.

One of the clearest definitions of this universal architecture is presented by none other than IBM. In describing how AI, machine learning (ML), deep learning (DL), and neural networks (NN) relate to each other, comes this brilliant gem.

“The easiest way to think about artificial intelligence, machine learning, deep learning and neural networks is to think of them as a series of AI systems from largest to smallest, each encompassing the next.

Artificial intelligence is the overarching system. Machine learning is a subset of AI. Deep learning is a subfield of machine learning, and neural networks make up the backbone of deep learning algorithms.” 
(IBM Data and AI Team, 2023)

Drilling down a bit on the IBM site came my true eureka moment for the day when the graphic depiction of a deep neural network.

(IBM Data and AI Team, 2023)

That should look very familiar to those who have seen an example of a parallel distributed processing (PDP) connectionist network. This PDP model of how the brain works is a perfect match for this deep NN model.

(Goldstein, 2019)

For cyber-physical control systems, we can take this model one step further, in that the NN can receive its input from physical sensors with fuzzy logic (FL) capabilities. FL is a precision digital system, but it has more than the traditional 0 and 1 levels. A fuzzy digital system has a range of levels with 0 and 1 as the ends of the range (e.g., 0.75 is 75% of 1). This allows for a digital cyber-physical system that “wants” values of 0 or 1 (off or on, false or true, etc.) to function in a non-digital world (e.g., faster, brighter, hotter, etc.) (Driankov, 1993).

This now creates a fully manmade replica of the major physiological data processing systems in a human from our sensory organs (e.g., eyes for sight/cameras for computer vision, fingers for touch/pressure sensors for tactile input). Once the raw data is input from these electromechanical sensors, that data is then fed up the FL > NN > DL > AI > MI hierarchy in a multiple-input and multiple-output (MIMO) system.

Now that we have a hardware system that can mimic our wetware (Farrelly, 2022) system and an MI model that mimics our brains… we can roll up our sleeves and learn something about ourselves and our physical or virtual automaton helpers. And our digital twins are the perfect playground for simulation-based training (Causer et al., 2014).

References

Barngrover, C. (2018, September 13). What is Machine Intelligence? Shield AI: https://shield.ai/what-machine-intelligence/

Causer, J., Barach, P., & Williams, A. M. (2014). Expertise in medicine: using the expert performance approach to improve simulation training. Medical Education, 48(2), 103-221. https://doi.org/10.1111/medu.12306

Driankov, D. (1993, January). An Introduction to Fuzzy Control. https://doi.org/10.1007/978-3-662-11131-4

Farrelly, J. (2022, May 26). What is Wetware and How Does it Impact Cybersecurity? Electric.ai: https://www.electric.ai/blog/what-is-wetware

Goldstein, E. B. (2019). Cognitive Psychology: Connecting Mind, Research, and Everday Experience (5th ed.). Boston: Cengage Learning.

IBM Data and AI Team. (2023, July 6). AI vs. Machine Learning vs. Deep Learning vs. Neural Networks: What’s the difference? ibm.com: https://www.ibm.com/blog/ai-vs-machine-learning-vs-deep-learning-vs-neural-networks/

Microsoft Research Lab – Cambridge. (2023). Machine Intelligence. Microsoft.com: https://www.microsoft.com/en-us/research/theme/machine-intelligence/