Is structured programming as we know today going to die soon?

I recall my early days in software programming, way back in late 1990s. Although the new age programming languages were gaining importance, there were still more opportunities in mainframe programming. There was a dilemma for starters like us — whether to learn newer programming language like C/C++/Java/Oracle etc. or explore something like COBOL mainframe programming. Most of us chose the first one and I believe that was a good decision.

Will the next gen programmers face a similar dilemma about the structured programming (traditional programming) and statistical computing (with machine learning)? Are they fundamentally different?

1. Deterministic Vs Probabilistic

Best example of this is IBM blue chess Vs Google AlphaGo. Both the programs ended up beating humans however the approaches were fundamentally different. While IBM blue took a deterministic approach (all possible winning position and possibly coding that with if-then-else kind of programming), it was not possible to do that in Alpha Go due to innumerable possible combinations. It rather took a stochastic/probabilistic approach of winning position and learnt with data over time to perfect it.

Alpha Go could handle a much more dynamic and complex environment than IBM Blue chess. In reality all are environment are immeasurably complex and we can’t enlist all possible actions and outcomes.

2. How vs What

When we meet any of our friend even after many years, we immediately recognize him in seconds. Now if we try answering that — how we recognized him instantly, we would not be able to answer ourselves. Even the nuero-researches answer that very vaguely. Actually we don’t know how it happens in our brain and in reality it does not matter.

In reality our brain works like a blackbox that receives some stimuli and produces some output but not fully explain the process of arriving at outcome. However it (brain) has been solving some of the most complex problems from times immemorial. Structured programming can’t work without the logic (how part) while a neural network in machine learning does not go into the realm of logic.

3. Idealistic vs Natural

We all know the famous quote “there are many slips between the cup and lips”. This is natural — nature has randomness in everything. Every action that an agent may want to take, there would be only a limited possibility of that actually happening. The structured programming can’t handle lack of causality, the outcome is entirely dependent on the input and logic

Input + Logic = Output

While in statistical computing the equation is

Input + Randomness = range of outputs with probabilities

The statistical programming account for these randomness while structured programming does not.

4. Static vs Dynamic

One can buy a robodog that is pre-programmed to walk in certain way or buy a robodog that is provided with very broad architecture of walking — that can experiment, fall and eventually learn walking based on the rewards and penalty fixed. While the first robodog can become productive early on, the performance of the second one will far outclass the first one over a period of time. Also when the environment changes the pre-programmed robodog will fail miserably.

5. Dumb vs intelligent

A structural program is dumb because it does not experiment with itself. The agent always does what he is expected to do from the program. Unless the agent does unpredictable actions, it will never learn the outcome of those actions. This is why a neural network becomes intelligent over a period of time while even a complex structured program remains dumb.

Intelligence has 3 basic components

• Memory
• Computation
• Learnability

While the structured programs have memory and computations — they have no learnability. The neural networks on the other hand keep on adjusting themselves till they become very good at optimizing the goal given to them.

The most important reason is that structured programming perform very badly in key human functions like image processing, speech recognition etc. Whether AI of today will ever reach human like AI (AGI) is a subject of speculation but it is pretty clear that the statistical programming with machine learning is more human friendly, natural and progressive.