The state of machine learning in AI

Posted on Wednesday, April 24, 2019

The rise of artificial intelligence (AI) has been a rapid one. Its significance is akin to the invention of the computer. It is here to stay and it will very likely improve our lives in the future and in many aspects, it already has. A University group is already underway looking into AI.

While the concept for today’s AIs has been around since the 1940s, it is only until recently that the needed computational power became readily available. AI is a broad term and covers everything up to fully autonomous programs that do not even exist yet.

One main area of interest in today’s AI revolves around machine learning, which in essence, is the ability of a computer program to discern patterns in large quantities of data. This can take many forms. For example, image recognition happens after feeding the algorithm millions of known images so that it “learns” what an image represents. Once the algorithm has learned what it needed to learn, it can interpret images that it has never seen before. AI also helps researchers find patterns in massive dataset and healthcare providers diagnose certain conditions. It can understand commands through speech recognition. Do Siri, Google, and Alexa ring any bells? AI even enables self-driving cars to recognize traffic signs and applications to translate restaurant menus.

In fact, translation is an area where AI is heavily used. Although there are many subtleties in the meaning of words used in certain instances as well as often-complex grammar rules, we can still train an AI to translate by feeding it translated texts. Google Translate is the perfect example of an AI used in translation. Still, it is not always perfect as it picks up the desirable patterns as well as the not so desirable ones. For example, translating “the receptionist” and “the doctor” to French, using Google Translate, will always make the former female and the latter male. This is a bias in the data, which it learned from. However, as this problem lies in the input data and not the algorithm itself, this can be overcome.

AI is a powerful tool. It is already well integrated to many aspects of our daily lives. While it keeps spam out of our inboxes, unlocks our phone with FaceID, and lets Netflix recommend shows for our next binge-watching all-nighter, it also lets Facebook and Google make us the target of their tailored advertisement … that some people consider a privacy infringement.

The state of machine learning in AI uOttawa


Back to top