Machines, Artificial Intelligence, & the Workforce

House Budget: Machines, Artificial Intelligence, & the Workforce: Recovering & Readying Our Economy for the Future, September 10, 2020

As a graduate assistant in 1985, I was given the task of investigating how AI would impact public administration. In those days, one could call MIT and get Marvin Minsky on the phone. This was also the topic of my Presidential Management Intern application essay and one of first topics I investigated as a PMI in the Air Force Cost Center. In those days, the World Wide Web was an DARPA project with no graphics, which I got a look at during my first temporary duty assignment to Space Division to shop for a library computer database for Center. I did not deliver the baby, but I was in the birthing room.

I was also assigned to be the Air Force Comptroller Liaison to the Cost Management System conceptual design for the factory of the future, as facilitated by Computer Aided Manufacturing – International, which is now known as Consortium for Advanced Manufacturing – International. The design product from our work was the book Cost Accounting for Today’s Advanced Manufacturing. 

The next phase of the project was a Systems Design. The latter was completed by the four original working groups based on the design, but was not totally integrated. Other duties kept me off the project, but I was able to get a copy of the documents, which I used as my Intern Research Project – an Air Force Users Guide. The Guide both detailed the CMS and cross walk it to Cost Management Control Systems Criteria. It was not classified and I have since published it on Amazon.

As far as I know, the systems design never went anywhere. The product was not fulling integrated between the committee silos and, because of regulatory and competition concerns, although it did evolve into Activity Accounting.

Expert systems, accounting systems and machine learning are as good or bad as the information that goes into them and depend as much upon the talents and objectives of the end users and developers. They can work beautifully as an online reference, provided the user is both talented and trained.  Management also must be able to use the results of the system.

Systems can make work easier, but as I learned both as a congressional intern with a rotation in the correspondence section and a frequent user from the other side, they are no substitute for good human decision-making. Automation can make things worse in that they can regularize error. In most cases, congressional offices receive simple viewpoints or requests rather than detailed analysis. When detailed analysis comes, the response is often the kind of form letter we all know and love. 

Having worked as a staff member or contractor doing agency responses to “congressionals,” I suspect that if agencies returned the same level of responsiveness to Congress that it provides to citizens with detailed proposals, there would be hearings and budget cuts. 

The fault is not the current automated correspondence system, it is the degree of freedom that legislative correspondents have in both answering responses and forwarding them to members and legislative staff for both input and response. I am quite certain that at some time in the future, a researcher will look at the Committee Print of these comments in the record and appreciate the irony intended.

Technology can only provide what the human system will allow, whether in manufacturing, sales or government service. The limitations of culture, particularly capitalism, limit what workers are allowed to do or the rewards they receive from expanding their responsibilities. 

Several quality improvement technologies have been tried. Dozens more will be tried in the future, but until compensation structures are flattened with increased shop floor responsibility, QC programs will simply be seen as the flavor of the week.

One technology which will greatly impact the workplace of the future is actual employee ownership and control, rather than simply training employees to understand financial statements. Upon request, I can provide more detail on how these would work.

Technology applications can certainly help manage production and distribution in our current crisis as some sections of the country open up while others must shut down. Cases, which will be followed by deaths, are beginning to explode in the Midwest. Not shutting down the entire nation, however, is a scientific and political decision – one that was managed incorrectly the last time it was attempted. It turns out that a virus which spreads by being sneezed on has a significant geographic component. The last shutdown was based on fear. Let us try science with the next.

Adding more funds to household budgets will be important, but adding too much money to a smaller labor supply will still have bad effects. Our current models are not attuned to estimate this possibility. Until the possibility of hyperstagflation is raised, it cannot be modeled. It needs to be.

The model for SARSCov2 is still for two weeks from exposure to major symptoms, largely because nasal symptoms are not included as part of disease progression prior to any asymptomatic transmission period. The experience of those who have had severe symptoms who have not required hospital intervention is of a four-week period from likely exposure to SARS or fatigue. An AI model may detect this, but only if the measurement parameters are correctly set. 

Garbage In, Garbage Out. 

The best prediction model is to take a per capita death rate somewhere between those experienced in New York and Massachusetts and apply it to every state. Until the CDC correctly specifies the symptoms of this pandemic, there is no way to stop its spread. Appealing to solidarity in public mask wearing will not prevent peer to peer transmission if everyone believes that their sneezes are allergies and not COVID-19. Because the public image of the virus is that of a plague, people do not want to believe they are sick. The problem is bad medicine, not fraternity parties.