Bureaucracy Reduction with Mediocre AI
A friend recently opened a new bank account for an existing textile business which had been profitable for 150 years and had no plans to change. The bank required a detailed business plan, so he asked ChatGPT to create one and handed it in without reading. The bank accepted it.
In the past, the effort required to create any type of business plan would ensure a degree of effort on the part of the applicant, meaning that the bank could become lazy in terms of due diligence (and not read it thoroughly, on the assumption that the effort required to deliver a plausible one was enough) and still not be exposed. Now, the onus is on the bank to actually read the plan to see if it is good and accurate, given it is easy for AI to produce a mediocre one with no effort. Going further, in the past, delivery of a plausible business plan would be regardless of whether it was the right ask in the first place. In the case of a 150 year old profitable business it isn’t and so now that the onus is on the bank to do proper due diligence the responsibility is also to have to ask for different documents. The complexity of dealing with bureaucracy has shifted toward the form creator rather than the form filler.
Until now, the asymmetry in terms of effort between those who create and administer bureaucracy and those who have to fulfill its demands has created a massive increase in inefficiency. This is obvious in government, particularly under statute based legal systems where, for example, the French employment law handbook is three times the length of the bible.
Bureaucratic asymmetry has been amplified by computers, which have enabled simple things like easy mass production of forms, leading to process overhead within corporations. Avoiding this bloat which is a function of size, is what often gives startups an advantage, even with less cash and sources. Middle layers in large organizations are often incentivised to create process over end results, where process metrics themselves are considered results.
The effects of bureaucracy bloat are most visible in regulated industries that require significant ass covering, particularly where litigation is high, creating an arms race. Litigation is higher in common law based systems like the US, so the bureaucratic overhead of a statute based system like France’s is merely diverted elsewhere.
Bureaucracy increases are also greatest in public sector industries or ecosystems (e.g. defense, education) where competition is lacking or where providers are a closed shop. In these instanced there are no alternative systems as reference points which reveal the inefficiency.
Generative AI potentially reverses the trend of bureaucratic bloat. Most importantly it does it automatically, due to incentive re-aligment and it does it now, because it doesn’t require really good AI, just sufficiently good.
Although generative AI creates tools which can also be used to create forms and process, just as computers always have, the difference is that AI based systems can fulfill the demands of these processes more efficiently than computers have been able to until now, which turns the tables on those that create process.
The incentive realignment is that once the tables have turned and a significant part (if not all) or the burden of effort in administering bureaucracy is now passed to the bureaucrats there is less of an incentive to create unnecessary forms to fill or documents to provide. Within a corporation, too much of a demand for, say, performance review documentation and people will just game it, passing most of the burden to an LLM. This will be self governing and, most importantly, with existing AI systems, since poor quality and unnecessary bureaucracy can be satisfied with mediocre content. Bloated internal middle-management systems are about to drown in AI generated spam.
The beauty of such systems is that they don’t need a corporate AI strategy or implementation. They will spread as soon as people have access to an AI chatbot, just as they have for student homework and essay assignments. Mediocre AI will spread within corporations like maggots on a wound, nibbling away at the diseased flesh of process for the sake of process and making corporations more efficient.
Generative AI changes the world by making organizations more productive. Most importantly, however, AGI (Artificial General Intelligence) is not needed, current mediocre output from generative AI is sufficient. Call it AMI (Artificial Mediocre Intelligence)! What we have now, AMI, reduces bureaucracy and increases organizational efficiency.
Thanks to Andrew Côté, who shared the graphs above.
