If you were to Google ‘artificial general intelligence’, or AGI, you would find an array of different definitions – many of which contain so much jargon, even AI experts may be left somewhat puzzled. Although put simply, the term refers to the ability of a machine to perform any task that a human can.
At this point, you might think that we have already achieved this milestone. After all, AI-powered technologies like Amazon’s Alexa have become virtual assistants in our daily lives, performing menial tasks like reading emails and finding recipes, that we would otherwise complete ourselves. Thanks to feats in natural language processing (NLP), we’ve also learnt to interact with virtual assistants as we would do with our peers. Indeed, IBM’s Watson now even understands idioms after a ‘sentiment’ update, paving the way for more nuanced conversations between man and machine.
There are many other fields where specialised artificial intelligence is replicating human-like reasoning and cognition. Deep learning algorithms used by social media platforms like Facebook, for instance, are becoming increasingly more adept at recognising objects and people. They can even offer detailed characteristics of the subjects, including their expressions. This is not unlike the way humans are able to perceive images.
That said, there is still a limit to what such technologies can accomplish. It takes a lot more than just performing specific tasks to a higher standard than humans to qualify as AGI.
Are we on the cusp of achieving artificial general intelligence?
It certainly might feel this way, but in reality we have a long way to go before we achieve true artificial general intelligence: a world where machines can replicate human functionalities. So far, we’ve built systems that excel at certain human tasks like data processing. However, we are yet to create programmes that are generally broad and adaptable like human intelligence. After all, the vast majority of machine learning (ML) methods employed today rely on human input and direction, and unsupervised learning remains a distant pipedream.
While creating algorithms that can replicate the complex computational abilities of the human brain is theoretically possible, there is a significant challenge standing in our way: we have a very limited understanding of how the brain works. So, by such logic, until humans gain a comprehensive understanding of the brain, it will be impossible to programme AI to perfectly replicate human behaviours, or indeed intelligence.
At the same time, there are multiple dimensions of intelligence; it’s not just about IQ and the ability to process and analyse data. As an example, humans possess a so-called ‘social’ dimension of intelligence, or, in other words, the ability to effectively navigate social interaction and read the emotions of fellow humans. Given the intrinsic complexity of emotions, programming AI to truly understand these, and how to appropriately react to them, would likely take decades – if not centuries.
There are no easy answers in a field as complex as AI. However, we have no doubt made great progress in the last decade in developing this technology; achieving artificial general intelligence certainly isn’t impossible. It will, however, take time and patience as we gain a deeper understanding of the limits and potential of AI.