2020 will be a record year in retirement numbers for industrial organizations across DACH and the US.
It all started with the Baby Boom after 2nd World War: A whole generation surpassing the age of 67 until 2028.
Age Source: CIA World Factbook
Industrial manufacturers face a challenge having to replace up to 50% of their organization within the next 8 years: This means up to 1 years of know-how loss every day per 200-500 employees employed if you’re working for an organization in manufacturing.
Note: We've carefully estimated based on the information we have
In combination with the lack of new workers to hire, the estimated economical damage sums over USD 3 Trillion just in the US in the next 8 years according to a study by Deloitte and the Manufacturing Institute in 2018. By taking income per employee difference, amount of workers employed and calculating, our combined outcome for DACH and US summed up damage lies at USD 6.16 Trillion.
One's attempt to make changes on a macro level requires an understanding on a granular level, so let's try to go deeper.
Post-retirement programs of part-time jobs for the ex-employees are planned to reduce the incoming disappearance of operating certain areas of ones organization. Temporarily such a post-retirement program will reduce the damage done. And that’s great: smart video-call integration to on-site operations allows to combine operational competence with the physical energy of youth and a charming way for intergenerational collaboration. Still, these programs are comparable to stretching gasoline with ethanol while fossil fuels are declining: It’s part of the plan, but not the solution (in this case, chemically a solution, but you get the point).
The chart shows program to retain an employee over the age of retirement, assuming the commitment decreases over time based on personal well-being and capabilities. Departments integrating such programs require an onboarding to the new process system which would reduce available know-how in the immediate period before the day of exit.
Instead of a hard exit date, retirees can support part time and increase their pension money, which also decreases age poverty. At the same time organizations have a longer window of opportunity to work with the most valuable experts.
In most industrial operational real-world cases communication orbits around being able to see what the other one sees.
As of today, multiple services for live human-human communication between help centers to onsite personal exist, the basic being a simple phone call to head-worn remote video calls. Mainly used for operative support scenarios, the main value provided through those who know it better but aren't physically onsite, giving them the best possible ways to see what others see.
These remote services are often combined with VPN data access to the problematic assets to give the maximum availability of data for quality decision making and problem solving, for example in the support department of a machine manufacturer.
Assumably these systems will be further evolved to suit experts in their transition phase, as they are doing right in this moment.
Human Resources will have to transfer know-how from retirees to existing work forces.
With an infinite budget such transfer question between two individuals would be no question for discussion: manual teaching and thousands of classroom events. However it’s always the cold hard truth of calculating the business cases:
→ Organizations decrease revenue by pulling workers out of day-to-day work for a longer period of time for training. These opportunity costs are often the even more expensive number than just the salary of trainers, trainees and infrastructure
→ The effect of know-how transfer depends on the method. Schooling teams for 1 hour with no significant results in long-term memorization (retention) is fatal
The Problem with Know-How is, it’s not just a rough description of doing something, it’s also not the surface level knowledge inside of a manual.
An expert car mechanic and his trainee would both be able to replace the same part, maybe even in a similar time, but it’s the expert who would need only 1 minute to diagnose the problem while the trainee would be clueless and not able to tell you what’s wrong, or even worse just fixing the consequence of a problem with deeper roots.
I’ve experienced this during my latest road trip during the United States where I bought a FORD Explorer 2000 which had a coolant leak through a crack in the intake manifold. After the 4th facepalming replacement of an ignition coil I pulled into a shop in Kansas where the (pardon my French) Red-neckish-appearing guy comes over, he wasn’t even the mechanic, just a parts-vendor happening to stop by and hang out for a couple of minutes.
So he stood at the opened hood, wrinkled his nose and went something like: „Is that coolant I’m smelling? Yep, that’s coolant. That’s definitely coolant. Look guys, your problem ain’t the ignition coil, you got a coolant leak probably caused by a crack in the intake, happens often with this model. You gotta replace the intake, otherwise you’ll end up replacing them over and over.“
The cause, but not the root
This true expert for FORD engines (I guess they break more often, so it’s a more constant education) identified the cause after 2000 miles on the road. Semi-knowledgeable people with a good will aren’t going to overcome situations where the knowledge and experience just doesn’t exist, especially not in situations where a gas turbine shut-down means having no electricity thousands of residents.
In applied medicine, doctors at hospitals and decades of experience apply a method they call „differential diagnosis“ which is simplified an experience based „intuition“ applied when simple „decision trees“ don’t apply.
In software engineering, developers spend days and days trying to fix non-existent code, often with baby steps of identifying why things are not working. Over time they learn where to apply the debugger as a logical analysis and interestingly also tap into collective intelligence and swarm intelligence via platforms like stackoverflow to save time and unnecessary memorization, a comparable method of „decision trees“ and „intuition“ as the doctor or car mechanic applies to conclude the right actions.
In data science, deep learning engineers can develop a sense to predict of how a yet-to-be-trained neural network would behave and find ways how to do small steps and changes in order to analyze results faster.
The list of professions with similar stories is endless – probably it’s every profession. In fact we can even talk about two principles related to describing know-how.
An elaboration on „decision trees“ and „intuition“ with models of Artificial Intelligence systems will provide more clarity.
This Senseo Coffee machine has an extremely simplified functionality for standard actions and situations. The hand book says „if this light blinks repeatedly red, then you have to refill the water“ if it’s red continuously, then go to page 6 because something else is broken. Hand books for a coffee machine are surface level decision trees to you can use to learn preventively how to react when a situation appears or to use during the situation as an operational guide. It’s to be pointed out the value of such knowledge would apply both for
→ Preventive Learning
→ Operational Guidance
The handbook then also proceeds to give instructions on what to do step by step to reach the state of a healthily functional coffee machine.
When comparing this coffee machine to a human individual, a Tesla Model 3 or large industrial piece of equipment like a Computer Tomography Scanner, the increased complexity of the latter systems results in a key main difference in their handbooks.
→ The knowledge of problems and their indicators for Industrial equipment is mostly unknown at the day of release from the production line: Their handbooks are neglectible because what’s inside them covers a fraction of what a mechanic expert of his machines applies in intelligent ways to identify and get it back running.
→ The human organism is on a level of complexity where general patterns can be documented and used as scientific backing, but with environmental, genetic, epigenetic and general relationship between physical and mental coherences, diagnosis and therapy is a continuously changing ongoing process with an increased man-machine relationship to deliver better outcomes.
For describable cases (often the simpler ones), an „if-then“ way of thinking can be applied. This leads to a process chart or a decision tree.
To describe the know-how of a piece of machinery in a single model, a network between attributes and their relationships is established. This is also the same way neural networks and their weights look after a training process. During the training phase (computing over time), neural networks are repeatedly trying to „solve“ and get feedback on how „close“ they are to being „right“ or „on the right way“. The 3 factors on how good a network can be is the amount of data and the quality of the description of what’s „right“ and the computing power applied (and the design of the network).
On this scale, AI training and human training is more-less the same, since AI has taken all its inspiration from nature and the laws of science which apply to our organisms (why would a computer file of 0s and 1s would be called „neural“ after all).
So when you ask an experienced person on how accomplish a challenge, you will end up hearing at least one „hard rule“ – an extract function derived from a massive conglomerate of intertwined neural networks. Even the most complicated organisms will have high level indicators to allow either for diagnosis „blood pressure, heart beat, lung function, strange noise, iris reaction to light, oil leaks, heat output and in the worst case, extreme silence“ and high level rules to achieve strong improvements in a low effort applied, like „hydration, adding oil, change in diet…)
In the end a map of know-how would look like a combination between hard rules and complex understanding represented through neural networks, all of them in a relationship to each other either through a meta-decision-tree or neural network. As of 2020 many AI researchers assume the closest thing to an Artificial General Intelligence (AGI) could have such a design… a direction towards „what would a human intelligence be described through software“.
If this is the model for Know-How, how would I be able to get it from one person into a medium or stored knowledge and allow it to be accessed through other people. In time, budget and in a way a human would adopt and not repel?
At the same time, more and more enterprises are evolving internal academies based on latest training technologies to teach faster, in higher quality and with a multiple leverage to trainers and instructors:
→ Central learning management systems control all learning and development on an individual level
→ E-Learning platforms allow for fast content distribution and tools to create and distribute content fast
→ VR learning for remote training of dangerous situations or with equipment of high opportunity costs like trains, planes, reactors or situations expensive or costly to replicate like evacuation scenarios
→ Augmented Reality direct learning situations for 90% memorization and continous improvements during operational use in the „70“ part of the 70/20/10 process of learning.
→ We need tools where both the simulation and the expert are intertwined and generate a new medium.
These tools require full efficiency
What tools are
→ We need tools where this knowledge is
Last Edited on Feb 1, 16:20 PM. Published by Daniel Seiler