In today’s world, there has been a large amount of talk surrounding artificial intelligence (AI). For years, Hollywood has been exploring potential tragedies that AI may create, and billionaires, such as Elon Musk, have been talking about the benefits they may bring to society.
Even products such as our smartphones and their apps contain some aspect of AI within them. All around the world, AI has become a term that seemingly came out of nowhere but is now just another part of everyday life.
But what is AI, and how does it work? Roberto Souza, an assistant professor at the University of Calgary and principal investigator for the advanced imaging and artificial intelligence labs, (stylized as [AI**2]) says that before we answer those questions, we should make one thing clear: the AI we are seeing isn’t exactly AI.
“The buzz that is happening right now, is all around deep learning technologies which are basically very powerful learning models from big data [sets].”
Artificial intelligence versus deep learning
Deep learning technology works by feeding computer software large amounts of labeled data, or data that has been tagged to provide context to what it is. From there, the software is able to sift through the data it receives, and make predictions on unlabeled data.
Common examples of this include facial/fingerprint recognition software on our smart devices, online chatbots such as ChatGPT, and even Tesla’s self-driving vehicles which contain deep learning software.
However, the term AI has become more of an umbrella term in recent years, which covers terms such as deep learning and machine learning technologies, algorithms, and more. According to IBM, actual artificial intelligence refers to a form of computer software that “leverages computers and machines to mimic the problem-solving and decision-making capabilities of the human mind.” These computers are made depending on the tasks they are asked to perform.
Current forms of AI differ largely from the ones that are portrayed in movies, such as James Cameron’s The Terminator from 1984. In this movie, the computer software is able to adapt, learn and grow, and is generally able to replicate the human mind and all of its nuances.
Concerns over AI
The idea for these types of AI can be seen over 50 years ago, when scientists were experimenting with neural networks which mimicked how the brain and its neurons send signals to each other. This idea of mimicking the human mind and its neural networks is still something that can be found in all technologies related to AI in some form or another.
While these technologies are just a buildup of previous old innovations, a new problem is created by the datasets that get fed into these technologies: online privacy.
Steve Drew, an assistant professor in the Department of Software and Electrical Engineering at the University of Calgary, says that traditional methods of collecting data for these technologies may also include your private information.
“With such [an] amount of data, these companies of course can do the good things which are beneficial for the entire society, but they can also do bad things such as surveillance of the citizen or predicting something people do not want them to predict,” says Drew.
Other companies and web browsers may also ask for some amount of personal information when you create a new account on a website or even your financial information for when you want to purchase something online.
Addressing the matters and finding solutions
The concerns over privacy and the collection of personal data have become such a large issue that governments are now passing legislation, such as the General Data Protection Regulation (GDPR) in Europe, meant to protect the privacy and confidentiality of these users.
Drew’s research, however, showcases that this issue can be combated with a different type of AI learning software: federated learning, a method that allows the software to receive and train data, without collecting any private data.
“The idea is that a company does not necessarily want the input history, but can still learn from it, without linking that personal information to the user. This way the privacy concern is reduced but the machine learning can still continue.”
There are other concerns surrounding the artificial intelligence industry in general, such as how the collected data is obtained, energy efficiency and more. Souza believes that in this case, the pros outweigh the cons.
“I would be [more] excited about how [AI is] gonna impact health care delivery, how it’ll impact robotics, and so many other important fields that’ll make our world better.”
Souza’s research, for instance, looks at ways that’ll help increase the speed at which MRIs are performed, which may also help patients receive those scans faster. Souza believes that by giving the deep learning model labeled data sets of sick versus healthy MRI scans, those scans can overall be performed faster, and detect issues quicker.
While artificial intelligence technologies do have their pros and cons, their integration with our society will still continue. Scientists and politicians are coming up with ways to make it safer overall, but we are a long way away from worrying about the next terminator movie becoming a reality.