Why A.I. doesn’t learn like humans: Part 1 — Literal Intelligence

Side-note: If you don’t care about the philosophical underpinning of why AI doesn’t learn like humans, skip to part 3 on curiosity, motivation, and logic.

Computers are beginning to recognize and learn more quickly than humans, but will that be enough to make them superior to humans? In some way yes and in some ways no. To effectively answer this question, we need to derive a very philosophical concept: how does a human learn?

To begin this series, I attempt to define human learning in a conceptually specific manner, arguing that it is the combination of 1) intelligence, 2) deep learning, 3) curiosity, motivation, & logic, 4) logic, consciousness, & understanding. These traits, in turn, reflect part of what it takes to have a form of life that is as successful as we have been. When these traits are programmed into a high-speed digital entity, it is arguably inevitable that this entity will become generally intelligent. There may also be part 5 in the future that covers sentience and volition. I have not yet written this part and I may decide not to. Let’s get started.

Part 1: Intelligence

Our evolutionary lineage has always cycled through a self-fulfilling prophesy: intelligence. A long time ago, I realized that evolution (natural selection) requires intelligent design and this understanding has furthered my appreciation of the universe. In the argument of evolution versus intelligent design, many people assume a religious bias that proposes an intelligent creator, ignoring the intuition that every living creature has some form of perceivable intelligence. However, when arguing, neither side of the intelligent design versus evolution debate does a good job defining what they mean by evolution and by intelligence. Most of us have a decent understanding of evolution, the notion that natural selection favors organisms that can reproduce more quickly and adapt to their environment. On the other hand, I’ve run into few people that are able to define intelligence. I will cut to the chase: intelligence is the ability to increase future options. Read it again if it has not fully sunk in. Something can be considered to be intelligent when it has the ability to increase the possibilities of what it may be able to do, and the stronger this ability is, the more intelligent we consider it. With this definition, intelligent design does not assume that a supreme being formulated the specifications for each creature but also does not ignore the intuition that living things are designed in a way that promotes their survival.

Evolution requires intelligence. Imagine two organisms: organism A and organism B. When organism A replicates, the subsequent generation is identical to the parent generation. When organism B replicates, the subsequent generation is a little different than the parent generation. Therefore, we can consider organism B to be more intelligent because it is able to change into more variations of itself. Which of these organisms would be favored by natural selection? The answer is pretty obvious: organism B would be favored because as time goes on, it is more likely that organism B becomes better suited for its environment, learns to reproduce more rapidly, or adapts to changing conditions. If you remove intelligence from organism B — its ability to change — you also remove its evolutionary advantage (actually, you remove evolution altogether).

In essence, in order for natural selection to operate, it requires organisms to be intelligently designed. And, the more intelligent an organism’s design, the more likely it is to change. Therefore, intelligence, like natural selection, is a universal law; a part of the seemingly self-fulfilling prophesies that explain our universe.

I believe that the intelligence of humans is one of the main reasons why we have become such a successful species. As we developed brains with increased capacity to learn, we began to rely less on instinct and more on learned knowledge, increasing our future behavioral options. Our prenatal development was likely one of the first steps in taking advantage of this process, causing humans to be born less mature. Though human babies cannot care for themselves, being born less developed allows our brains to comprehensively interact with the environment during development to cultivate the optimal behaviors for our survival. There are many well-known critical periods in the developmental biology of humans (cut-off points for developing certain skills that define our environment). Examples of these critical periods involve the visual system, the auditory system, the vestibular system (balance and orientation), and language processing capabilities such as grammar and having a proper accent. Our species is unique in the extent to which it is intelligently designed for information processing.

I don’t believe that most of us realize the depth to which intelligence is ingrained into the way we operate. Take the very artificial concept of money as an example: though we may not be consciously aware of the logic of this scenario, we typically tend to regard people who have amassed a lot of money as being intelligent. Money allows us to buy almost anything and therefore, having money increases our future options. Most would agree that you don’t have to be intelligent to be rich but you do have to be intelligent to become rich. It is the ability to maximize future options that confers intelligence, not having those options. It is for this reason we commonly assume that people who amass large sums of money are more intelligent than individuals who simply inherit it or obtain it by random chance.

So far, I’ve described that intelligence is a requisite of natural selection, that it has enabled our evolutionary advantage, and that it is deeply rooted into our thought processes, but what does this have in relation to whether computers are superior to humans?

Deep Learning computer algorithms are digital learning programs using a dense multilayered branching structure that enables information to multiply and flow through an innumerable number of paths. They are incredibly malleable and can learn to recognize concepts more quickly and accurately than humans. This intelligence is derived from the exponentially large number of ways that information can flow and interact. While these capabilities are impressive, they do not directly enable the algorithms to maximize the future options of those algorithms. Rather, their intelligence is confined to the operations within the algorithm. Therefore, deep learning algorithms do not contain a requisite ingredient for evolution outside of their programming and cannot outgrow it. In the next section (part 2), I’ll identify the way the biology uses deep learning to understand the world and accumulate knowledge. In part 3, I’ll define the system-level processes that are used to build, apply, and optimize the knowledge of the brain. In part 4, I’ll discuss further how this knowledge optimization turns into understanding and describe the role of consciousness within it. In part 5 (which may never get produced), I’ll describe how sentience emerges from consciousness, how it enables deep learning to be directly intelligent, and how it may be possible to create sentience with today’s technology (although highly inadvisable).