After over millennia of slow progress from the wheel to the printing press, mankind has finally made it to the ‘Information Age’, an era where data is an invaluable resource, and leveraging it equates power. In such a setting, it is only natural to be wary of the technologies that enable the use of these vast pools of data – Artificial Intelligence, to be precise. At first glance, it is perfectly understandable to believe that AI may one day go rogue and take over the world, or at the very least replace certain professions, leaving multitudes of people unemployed. AI is, after all, an attempt to produce ‘super intelligent humans’, or agents which perform the same tasks as humans, but at a far more capable level.
This is where the misnomer of AI begins. Artificial Intelligence broadly falls into two categories: Artificial General Intelligence (AGI), a lofty feat of developing human or super-human like capabilities, and Artificial Specific Intelligence (ASI), which focuses on an AI agent performing a very specific task like playing chess. As of today, AGI is purely a myth. The closest that researchers have come to achieving AGI, is developing an AI algorithm which defeated the world champion of the ancient board game ‘Go’; and even this narrow application algorithm falls into the category of AGI purely due to the fact that the algorithm learned to perform new moves that had never been used by humans before, showcasing an element of ‘creativity’. Despite the impressiveness of this feat, it seems rather hard to believe that a board game playing AI agent can gain sentience and take over the world, does it not?
On this note exists another misconception about AI: an AI agent can never become ‘sentient’ and develop emotions. The concept of synthetically creating an emotional understanding is an impossible task AI researchers have never even considered working on. The field is called Artificial Intelligence and not Artificial Emotions, after all.
The AI we see in today’s world is almost entirely ASI: from Google searches which can predict the next word you type, to your phone which unlocks on recognizing your face. All these applications of AI are focused towards incredibly narrow domains. Current AI agents suffer from the limitation of being unable to ‘generalize out of distribution’. In layman’s terms, this means that an AI agent that has been taught to perform a specific task, like a camera app that identifies the faces in a picture, is completely incapable of doing anything else. If one were to attempt the agent to perform something else, like completing your Google search, the agent is very likely to break down or forget the original task it was taught.
In such a setting, it seems highly implausible that AI agents would be capable of replacing existing human jobs. They are currently designed to augment human performance instead, either adding new features such as impressive photo filters, or helping humans perform more efficiently. A wonderful example of this is in the medical domain. An AI agent can be trained to look at a picture of a patient’s eye, and detect certain traits which might show the onset of diabetes. Since the agent picks up on certain patterns in the image that a human is incapable of visually identifying, doctors can now detect and begin early treatment for patients who would have otherwise suffered.
The current uses of AI far outnumber the small glimpses mentioned here, but one thing is for certain – while powerful, AI is merely a tool to help humanity take longer strides on the road to progress, just like the many human inventions that preceded AI. AI is the new pen to a writer, more powerful and capable than ever before, but it could never replace the writer.
By Dr. Kuiljeit Uppaal,
World’s First Image Scientist & Impact Strategist