AI is what machines do, not how they do it. It is the theory and development of computer systems able to perform tasks that normally require human intelligence.
This broad definition of AI can be a bit misleading when trying to understand the specific capabilities of AI. And all the hype from tech companies and advertisers is also not helping to pinpoint what this technology can actually accomplish today and what is the stuff of the future. To offer a bit of clarity on this subject, two general terms are used in regard to AI: ‘narrow AI’ and ‘general AI’
Narrow AI is designed to perform very specific tasks and is widely used today. Examples of narrow AI include self-driving cars or voice assistants. Though it becomes smarter and better at the tasks over time, it is designed for a pre-determined and pre-defined scope of abilities and is not conscious or emotion driven. As these systems learn, they gradually outperform humans still, however, on a specific scope of tasks.
General AI on the other hand refers to machines that exhibit human intelligence and is the technology of the future. Where Narrow AI is designed for a specific range of tasks, General AI is able to perform any intellectual task that a human being is able to. This generation of AI will be able to solve problems, make judgement under uncertainty, plan, learn, and integrate prior knowledge into decision making. Though General AI is what most people imagine when they hear ‘artificial intelligence’, the reality is that we are very far away from creating true human-like intelligence.
It is easy for decision makers to be unaware of the current state of AI due to hyped press coverage and exaggerations. Often leading to confusion about what AI can actually do today and what is to be expected in a few years or decades. When thinking about AI, it is easy to jump to the examples of the latest consumer applications such as Amazon’s Alexa and Google Duplex. Experiencing the advanced capabilities of these technologies, it seems like AI can do anything and solve any problem they have. However, the reality is that current technologies are still fairly limited.
Narrow AI is undoubtedly a great feat in human innovation and intelligence. It has helped people improve their overall productivity. Companies can either leverage it to replace humans in performing mundane routine work or it can be used to assist employees in their daily tasks to increase efficiency. That being said, this technology can still only be applied to specific tasks. We cannot expect AI to be a fix for everything.
Today we see platforms that promote themselves as AI platforms but that are merely based on corners of AI (e.g. text processing) and then topped with old-fashioned tools to structure data. These platforms are programmed with rules like question-answer trees. Though being well suited for building simple tools that automate processes, the platforms are limited by the need for manual work and have little to do with ‘real AI’.
Without immersing into too much detail, we often see that top management buys into those platforms with the expectation that they can support AI tools across the organization and be future proof. But implementing these platforms is a lengthily and labor-intensive process. The extra human labor that goes into the implementation is inherently contradicting the purpose of AI, which is to reduce human involvement and diminish the workload from day one. Moreover, when real AI tools get to the market, there will no longer be a need to structure data manually.
Instead of seeing increased productivity, top management is rightfully unimpressed seeing the sub-par results of the AI, which has built on top of an existing platform and has been implemented over a lengthily period. At the same time, tech mid management is trapped using a platform, which consumes more resources than it provides.
At times a mismatch between the visions of the technology implemented and the real-life problems it is supposed to solve will lead to confusion and create work processes that are far too complicated. Businesses will aim to implement future-proof solutions, but when it comes to AI, the development is moving at such a pace, that the future- proof solutions are not solely about AI itself. Instead, it is highly dependent on whether or not the infrastructure is compatible with a fast implementation of AI.