The Ottawa Chapter of the Canadian Association of Journalists recently organized a discussion on the integration of Generative AI in the newsroom. The event’s speaker was Florent Daudens, the director of national and international news gathering and deployment at CBC/Radio-Canada, who teaches digital journalism at the University of Montreal. Some of the attendees included journalists that reported on Parliament Hill and a university professor of journalism, and it contributed to an evening of interesting exchanges on what the future holds for generative AI both in newsrooms as well as the world at large.
Distinguishing between different elements of AI and Generative AI.
The acronym “AI”, short for artificial intelligence, has been referenced as a blanket term for anything that has to do with technology and advanced computations. However, AI is far more complex depending on its capabilities, methodologies, and applications than a general blanket term could ever provide context around. There are many different subsets of AI, but there are a handful of specific subsets that are necessary to establish a baseline understanding of what AI is all about.
Machine Learning (ML) is a subset of AI that learns from data and identifies patterns to make decisions. It requires almost no human intervention. ML is best understood as a non-stop learning force that has adaptive and predictive capabilities in hopes of continuously getting better.
Deep Learning (DL) is a subset of ML that is best understood as a the “brain” or neural network that interprets large amounts of data. The strength of an AI’s DL abilities determines how high its level of accuracy will end up being when it comes to making sense of data.
Computer Vision (CV) is a subset of AI that enables machines to digitally process interpretations of the real world. CV is best understood as the “eyes” that make sense of visuals, being used for things like object detection, classification, and recognition.
Natural Language Processing (NLP) is a subset of AI that enables machines to be able to understand and interpret language, but also to recreate it. NLP is what makes it possible for AI-powered assistants like Siri, Alexa and Grok to engage in dialogue with users, to understand questions and to be able to generate text.
Robotics is where AI intersects with the physical and it is best understood as the place where AI is given a physical form to carry out tasks that would otherwise require humans. Robotics may occur with a single AI-powered robot, or it can occur in a coordinated effort of robots, and the tasks that a robot can perform are specific to its design.
Generative AI (Gen AI) is a category of AI that utilizes DL and exists to generate new material like photos and videos from text. GAI is best understood as workforce behind creating content from text-to-image or text-to-video prompts, it gives life to words, and can be indistinguishable from content created by humans.
Seeing AI as it currently is.
Today’s AI systems have been trained by people, and every person carries with them conscious, unconscious, and subconscious biases, including programmers. As a result, today’s AI is likely to have aspects to it that exhibit these biases, depending on who, how and where these AI systems are trained. Despite that, different forms of AI are making major breakthroughs across different field of science, with a seemingly vast potential of what AI can go on to help humankind achieve, it is not without limitations.
Gen AI often seems magical in its powers to create, but that end product may also end up including a high level of bias based on a variety of different factors including a person’s location. One example that was highlighted by Daudens was that a person asking a Gen AI to create images of homes in Beijing who is located in North America was likely to show traditional homes from centuries ago, while the same ask from a person in Beijing was likely to show skyscrapers. Then there was the potential to misuse Gen AI to create deceptive photos, videos and more.
Generative AI in the newsroom and beyond.
During the Q&A interaction with Daudens, some statistics were discussed relating to the usage of Gen AI in the newsroom. It was mentioned that there have been different forms of data that have been collected that indicate around 30% of newsrooms around the world are leveraging Gen AI for written material, but that it was far more likely that upwards of 50% of newsrooms were engaging with Gen AI – largely unknowingly.
In Canada, some newsrooms have already pledged to be transparent about their usage of Gen AI, but the threat posed by Gen AI has less to do with newsrooms in Canada and more with “a war on perception”. In the next few years, half of the world’s population will be voting in different elections and this period will likely go down as one of AI-powered chaos.
How will deepfake videos that look and sound like a presidential candidate making outrageous remarks get discredited as being fake, and at what speed and ability will these deepfake videos be able to get deplatformed? Or, what happens when some platforms decide against removing deceptive content?
The dark side of Generative AI.
At the start of the year, I wrote about the vulnerabilities associated with financial companies and financial institutions that were allowing users to verify their identity over a phone or tablet. It appeared to be the only article highlighting the issue at the time, until 404Media published a major article in February about a site called “OnlyFake” that was churning out fake IDs that were bypassing the security systems of financial companies and institutions. Shortly after 404Media’s article came out, the website was deactivated, but it was too late. The source code behind generating the fake documents seems to have been copied, and now there are an endless number of sites that offer the same service, and it is impossible to shut them all down.
What makes Gen AI document generators such a threat is that they allow for users to pick and choose the nationality of their passport or the city of their driver license. This option allows for cyber criminals to create profile of people that have never existed, and “ghosts” are moving money around the world. And if the cyber criminals are really good, they will make it so that their identities are untraceable. But that might pale in comparison to the power of being able to generate videos of people with audio that replicates their voices, whether it be for the purpose of interfering in an election or to carry out targeted crimes on vulnerable people.
All of the progression around different forms of AI and their capabilities has opened a can of worms that no government or police force is truly equipped to handle. Out of all the possible solutions, the only sure-fire way for the issue to get addressed is to open many different cans of worms, including further regulating how people use the internet and what kind of network activity should or should not be private, as well as how much autonomy should social media companies have when it comes to determining the content that gets posted online. All of it is uncharted territory that needs to be charted far sooner than today’s social norms might be prepared for.
