The Paradox of Automation: Why Human Intent Matters in the AI Era

Description: My philosophy on achieving 90/10 operational efficiency. Learn why automation doesn't replace humans, but instead makes human oversight the most critical part of the network stack.

At a recent meeting I used the phrases Process Automation and Automation Solution as part of the explanation of a possible solution for a project. Which I believe sparked some confusion. The project focused on the data gathering, current network state and possible expansion into the network redesign phase. Part of the Automation Solution I suggested involved creating custom programs or scripts to automate some of the processes. After diving into the thought process, an example and providing a map of the end point, they got it. That call was the seed for this article. The denotation of automation seems fairly neutral and broad: the technique of making an apparatus, a process, or a system operate automatically. The connotation of automation is where I found the reason for the confusion on the call. The International Society of Automation (www.isa.org) defines Automation as: the creation and application of technology to monitor and control the production and delivery of products and services.” Which has everything to do with manufacturing process automation, and very little to do with the IT process automation I was discussing. The automation I was discussing on the call centers around mimicking human interaction with computer systems, like updating a cell in an excel spreadsheet or issuing commands to a router and verifying the hardware. Wikipedias entry for Automation also was disheartening, as it again focuses primarily on manufacturing and control systems . I found an interesting gem under Automation Paradox of Automation: The Paradox of Automation says that the more efficient the automated system, the more crucial the human contribution of the operators. Humans are less involved, but their involvement becomes more critical. The gem is vitally important, not only because I have found this to be very true, because it is a critical planning piece. Human QA of the automation output and the corrective feedback into the solution ensures errors are caught early, quickly and resolved. Additionally, errors are not always caused by faulty logic within the automated system, an error or omission in a configuration template, or a version of the template, can inject issues into the process or project. Example: not enabling lldp by default on all switch configuration templates. When it was caught and remedied, more than 800 switches had been deployed, and 300 were affected. Thankfully it was a relatively easy change and the fix could also be automated. {Personal Glossary} [Automation] - the technique of making an apparatus, a process, or a system operate automatically. (With it being applied broadly, rather than specifically to manufacturing.) [Automation Architect] - Professional who designs, implements and executes Automation Solutions. [Automation Solution] - A grouping of automated processes having the purpose of delivering a discrete solution; usually providing a saving of labor, decreasing overall errors and increasing efficiency. [Process Automation] - the act of modeling processes for the purpose of dividing labor amongst computer and human actors. {Automation Solution Example} The example Ill use is a segment of a successfully automated project. The overall scope of the project encompassed the network refresh of eligible equipment at multiple client sites. This could include the router, firewall and switches. Equipment was determined to be eligible based upon the clients business logic. [ Diagram of Integration Process Phases ] The Automation Solution I created was divided into 4 discrete phases: Pre-Install - Covers the collation and transformation of existing site documentation, the site survey and network design. Staging - Automated configuration of site specific equipment. Integration - Here is where most the high-skilled Human intensive work resides. Close Out - Crossing the Ts and dotting the Is of the client deliverable documentation. An important aspect of the phased approach is the Quality Control Review process between each phase. This created a gate for a human QA review of the output, before it goes onto the next phase. Vital to mitigating the Paradox of Automation issue sited earlier. [ Diagram of Pre-Install Phase ] The Pre-Install phase is further divided into multiple phases. Again, each of the phases represent a discrete set of processes, with the ability insert QA reviews between them. Well dive into the Pre-Generate Survey Phase to get a little more detailed. [ Diagram of Pre-Install Phase Detail ] From a high level, we are taking the existing site information and documentation, cleaning and normalizing it, saving it for later (Project dB amp; Project Folders), merging the information into a survey template, and then zipping it all up into a single file for distribution. The squiggly arrows represent the fuzziness of the logic required. In modeling the Clean amp; Normalize Data process, I stepped through, and documented, the individual actions a human would take to: get a copy of the production documentation, query the clients inventory database, remotely access and query the existing site equipment, verify accuracy of the collected information,and transpose the filtered accurate data into a form. Having modeled all of the necessary tasks, they were then mapped to computer or human actors. The final Clean amp; Normalize Data process was implemented in Perl5 on an Apache web server, and was a completely automated process that was initiated by the Project Managers office updating their database with a scheduled site survey date. QA was performed by Network Engineers, and further QA input provided by the technicians performing the site survey. Final Thoughts When designing Automation Solutions the standard goal is to have 80% of the labor performed by computers. The remaining 20% effort being the focus of 100% of the human labor. Realistically, I think a 90/10 split with the 10% going into automation process refinement is advisable. To reiterate, part of the human effort needs to be strict QA gates to catch any errors or oversights in the solution. This also aides in enforcing an Agile view of the process where errors caught during QA are immediately fed back into the Automation Solution for refinement and additional error catching. Overall, I have implemented solutions that decreased labor hours by 50% - 95% for complex processes. Correspondingly, I have witnessed upwards of 75% decrease innbsp;gross process costs. {DO IT NOW - DONT WAIT - BE READY} Businesses have the ability to implement Process Automation now and the standardization of processes will only assist in the long term strategy of becoming a Cognitive Business.