Defining Artificial Intelligence: Narrow versus General Artificial Intelligence

Narrow vs General Artificial Intelligence

At its most basic, artificial intelligence is simply a system that manifests intelligence or if we would likely to get a little creepy, a system (software or software plus hardware) with some “consciousness”. But AI is deeper and more complex than that.

In general technological writings or pop-tech writings such as on blogs and newspapers, you are likely to encounter the two terms “artificial general intelligence” and “narrow artificial intelligence” a lot. Although it is very rare to find the words used interchangeably (unless the writer absolutely doesn’t know what they are talking about), they are often thrown around with little precision.

Both general and narrow artificial intelligence are used to refer to types of artificial intelligence machines that can either be real or hypothetical. But what do these terms generally mean? Can you draw a distinction between the two? Let’s delve into the differences between general and narrow artificial intelligence.

What is General Artificial Intelligence?

To begin with, the general artificial intelligence has not even been realized and it might probably take decades before we can create a general or strong artificial intelligence. When we talk about general artificial intelligence, we are therefore imagining a hypothetical AI machine that will have human-level AI and be totally or semi-autonomous.

A general artificial intelligence will have a problem-solving capacity that will make it possible for the machine to self-learn various tasks in multiple domains. So the general AI will have the core abilities that will give it human-level intelligence: a problem solving ability and a cognitive ability (ability to learn autonomously). You will need a machine that is able to “think” abstractly, perceive things the same way a human would, and juggle between various unrelated tasks. Humans are able to work out problems at a general level and come up with ideas that have no precedence. This is a very complex kind of thinking and with the currently AI technology and human ability, it is not possible to design a machine that can replicate that at this time.

The general artificial intelligence or strong AI is a machine or software piece that is capable of interacting with its environment in a way that a human mind would. It is so complex that it has remained largely elusive so far in spite of decades of research. By pursuing strong AI, we are trying to build machines that will mimic humans or probably be super-intelligent “beings” that are totally autonomous and self-motivated.

Narrow Artificial Intelligence

Unlike strong or general AI, the narrow artificial intelligence is within the realizable realm. In fact, when people talk about the existing artificial intelligence technology, they are generally talking about the narrow artificial intelligence.

The narrow artificial intelligence refers to an AI that performs tasks that normally require human intelligence. However, it does these tasks within a very narrow or specific domain. For example, a narrow artificial intelligence can be designed to play chess and the AI will only play chess but will not learn any new game or even new strategic chess moves unless it is “trained”.

However, narrow AI is not a specific definition. It’s more of a moving target. While there are AI implementations that certainly lie within the “narrow” confines, some forms of AI are able to do much more than just specific narrowly defined tasks. Case in point was the AlphaGo Zero AI that mastered superhuman playing skills in the Go board game by just knowing the rules and the mechanics of the game and playing against itself over and over again without any human input or constraints. The software made use of reinforcement learning in which a system is rewarded for hitting certain milestones without showing it how to reach the goals. The case of AlphaGo Zero shows the limits of using terms such as “narrow AI” in describing AI systems. It can sometimes be imprecise. Here was a case of a “narrow” artificial intelligence system that was learning and creating new algorithm without any data set.

Many of the current artificial intelligence systems that seem so advanced such as weather forecast systems, sales ordering systems, natural language processing systems, image recognition systems, translation systems, speech recognition systems and even self-driving car technologies all lie within the spectrum of narrow artificial intelligence.

The defining feature of narrow artificial intelligence is that it can only take tasks within its scope. While it can learn, it will only do so within a narrowly defined scope. For example, an artificial intelligence system that does speech processing cannot learn how to do language translation. However, it is possible to “cascade” several narrow AIs to create a more complex coordination of multiple narrow AIs. That is often the case with the self-driving vehicle systems.

Most developments in artificial intelligence in the next decade or so will be in the realm of narrow artificial intelligence. It is narrow AI that will ultimately replace routine human jobs in industry.

The complexity or sophistication of narrow or “weak” AI also shows why General AI that will be similar to “human-level” intelligence is still a long way from realization.