On the Strengths of Artificial Neural Networks: Versatility, Scalability

In my previous post, I discussed the weakness of artificial neural networks (compared to their biological counterpart:) Energy efficiency.

In this post, I will discuss their strengths, namely: Versatility, Scalability, and Inorganicness.

First, on versatility. Artificial neural networks, unlike biological organisms, are not biased towards any one architecture. They are not genetically or epigenetically programmed or preconditioned in a way that is out of our control (or our legal purview, as the case may be--at least for the present.)

Next, on scalability: Though the hardware that supports ANNs may be inefficient, it is, nevertheless, scalable. We can assemble and apply, in theory, as many or as few GPUs or processing cores as we want to a given task or application. We can make small robots or enormous data centers. And that brings us to the third strength: Inorganicness.

As I touched on in the paragraph on versatility, ANNs have no biological imperative; they have not "evolved" to be suitable for any particular task. This can be a strength or a weakness, depending on the situation. But they are not susceptible to the same adversarial attacks as humans are. They cannot become addicted to drugs. Their supporting hardware may be more robust to various environmental conditions: lack of oxygen, temperature variations, etc. -- making them useful in hostile environments like space. They come without the trappings (or the strengths) inherent to organic biology: The need to sleep, to breathe, to eat (and eat well if high performance is desired,) etcetera. Take it for what you will; it is what it is.

Clearly AI is an interesting field; an enormous amount of hype surrounds the topic at the moment. It is important with it, as with anything, to see both sides of the issue, the pros and the cons, so as not to get swept up in a monotonal frenzy -- a one-sided view of the situation which fails to consider possible alternatives. AI can thrive where human workers are weak. But for some tasks, humans will continue to excel for some time to come -- and other technologies are worth considering.

Consider, for example, the "security guard" pattern, in which a guard sits in front of a collection of monitors. What are his strengths and weaknesses in this scenario? A strength is that he can efficiently monitor many views in times of low demand, when an anomalous event is unlikely. A weakness is that if he has no way of calling for help, he may be quickly overwhelmed if multiple anomalous events occur immediately. On-demand scaling is clearly important for the guard. So it is for any computing service where demand can change rapidly. 

But without giving a cloud computing seminar, let us consider also the case of the air traffic controller, whose job it is to monitor and manage many trajectories.

We can also consider the human waiter, as at a restaurant. Or the chess expert who plays multiple games at once, or the rubik's cuber who solves many cubes in parallel. 

All of these individuals multitask, to varying degrees. But this capability is only developed through expertise within their given domain. A beginner cannot perform this multitasking as well as an expert, because the beginner is not optimized for it.

Clearly there is more to be explored here, but the pen tireth for tonight.