Bureaucracy Was the Original Artificial Intelligence
It might be difficult to believe today, given the hype around Artificial Intelligence, but popular interest in the topic lay fallow for a long time. Recent developments including Transformers models such as OpenAI’s GPT-3 application have reignited interest in using machines to generate text and images. The meltdown in cryptocurrency means that Silicon Valley is desperate for the Next Big Thing. AI is it.
Just like with cryptocurrency, the flames from this bonfire will attract many types of insects, drawn by greed and envy. Already there have been reports of so-called AI startups that had nothing artificial about them. They were outsourcing to people in other countries and pretending that a machine had generated the results.. Fake it ‘til you make it, indeed.
Or, as the tweet referred to above indicates, scammers exploit the lack of general knowledge about AI with other kinds of simplistic automation, here lampooned as a set of if-then statements.
From a layman’s perspective, I think of AI in three forms: expert systems, machine learning, and deep learning. Let’s focus on expert systems.
The boring type of AI, the kind that still persists under coded names such as robotic process automation, is the branch called expert systems.
Here is a definition of an expert system:
“Main components of an expert system: (i) the knowledge base is the collection of facts and rules which describe the knowledge about the problem domain; (ii) the inference engine is the part of the system that chooses which facts and rules to apply when trying to solve the user’s query; and (iii) the user interface is the part of the system that takes in the user’s query in a readable form and passes it to the inference engine. It then displays the results to the user.”
Sound familiar?
If a bureaucracy is a set of codified, specialized expertise applied with rigid discipline to processing information and making decisions, then bureaucracy was the original expert system. Instead of if-then statements programmed into a computer, bureaucracy is some dude in a cubicle applying rules to data that others have input into forms.
The knowledge base in the computer-based expert system is only as good as the domain expertise of the people who contributed to it. The inference engine is only as good as the problem-solving algorithms that the experts who were the model for its development employ. The user interface is only as clear in its explication of the result as its designers made it.
By extension, the criticism computer scientists levy at expert systems should extend to bureaucracy.
Here’s a start:
“The dependence people place on cases pose a problem for those who treat human cognition as being primarily rule-based. Much work in artificial intelligence, for example is done in ‘expert systems.’ These systems are based on the notion that expert knowledge consists of a collection of rules. By determining the rules an expert in a domain uses, we may then simulate expert behavior in that domain. Not surprisingly, expert systems have run into a significant problem: they are brittle. When faced with a problem which bends the rules, they are unable to cope. They fail because they are not grounded in cases. They are unable to fall back on the details of their experience, find a similar case, and apply it. Likewise, they are unable to use similarities between tough problems and previous experience to update their rules. Their failure to retain cases cripples their ability to learn from their experiences.”
Or another:
“Finally, the following disadvantages of using expert systems can be summarized:
1. Expert systems have superficial knowledge, and a simple task can become computationally expensive.
2. Expert systems require knowledge engineers to input the data, and data acquisition is very hard.
3. The expert system may choose the most inappropriate method to solve the problem.
4. Problems of ethics in the use of any form of AI are very relevant at present.
5. It is a closed world with specific knowledge, in which there is no deep perception of concepts and their interrelationships until an expert provides them.”
More:
“Expert system developments fail for the same reasons as conventional software systems. However, they may also fail because the problem they are dealing with is not or cannot be understood, because they require common sense or knowledge of the limitations of their knowledge, because they cannot be tested, or because they cannot or will not be trusted.
This is a high-level sense for some of the most common criticisms of expert systems, but we can boil them down to handful of keywords:
· Brittle
· Superficial
· Arrogant
· Expensive
· Biased
· Inflexible
· Limited
Human cognition is not rule-based. There lies the rub. We are non-linear creatures.
This is what frustrates people the most about bureaucracy. If a bureaucracy is the codification of specialized knowledge into a set of rules that a system administers broadly, then bureaucracy was the original expert system.
All of this leads to one conclusion.
The Axiom of Bureaucratic Scale: Bureaucracy becomes less effective, the broader the scale of procedures it addresses. The most effective bureaucracy is for narrow, well-defined situations with little to no variability, requiring no human intervention.
The corollary to the Axiom of Bureaucratic Scale may be the following:
Corollary to the Axiom of Bureaucratic Scale: The broader the scope of decisions we ask bureaucracy to make, the more human intervention it requires to compensate for its lack of depth and its inability to learn, even as its indifference to the limits of its expertise may prevent such mediation.
We can add this to our growing list of ways to distinguish and characterize institutions or procedures as bureaucratic, starting with What Is Bureaucracy?