The duty invites you to return to the side roads of university life. A proposal that is both scholarly and intimate, to be picked up all summer like a postcard. Today, the season continues with a reflection on artificial intelligence (AI) which is here for good. You might as well tame it.
“The aliens arrived, but we didn’t notice it because they speak our language. » This is how Geoff Hinton, nicknamed the “grandfather of artificial intelligence”, describes the arrival of generative AI systems like ChatGPT. These systems are extraterrestrials according to him, because their way of understanding and acting in the world is fundamentally different from ours, like the heptapod creatures in the film The arrival by Denis Villeneuve. But unlike these, we do not have to hire linguists to learn to communicate with them: these artificial intelligences were created to converse with us.
And like the Martians from the cult series The invaderswhich could only be recognized by their inability to bend one of their little fingers, these artificial intelligences will invade us and become more like us day by day. The capabilities of systems like ChatGPT depend on what are called large language models, abstract and complex representations of natural languages developed solely from texts gleaned from the Internet.
But they’ve already begun to be coupled with large visual and audio models to give them powerful forms of vision and hearing, and companies like Boston Dynamics are starting to integrate them into robots. It’s no longer science fiction to think that we’ll soon live in a society where we can interact with our machines as we interact with each other.
But what to do when you are invaded by aliens who do not want to leave? The usual attitude, because it does not require any decision, advises waiting: let’s see what happens and make decisions when abuses and negative effects appear, as we did for the current generation of large Internet companies. This would be a failure to understand the disruptive potential of these technologies that herald significant systemic changes in all the major systems that make up our societies. What to do, then?
We first need to better understand how generative AI systems do what they do. It may seem strange to recommend the study of systems that we ourselves have built. But, in fact, we did not understand these systems well: they are immense and are expected to grow exponentially for a long time to come. They have hundreds of billions of parameters, each like a small rheostat individually adjusted by an automatic process based on a massive set of data whose possible interactions no one knows.
This makes these AI systems very opaque, almost black boxes, manipulating information in ways that still elude us and thus producing all sorts of unexpected behaviors. If the existential risks arising from this uncertainty remain low, far lower than those announced by climate change, the economic upheavals are certain and pressing, the only doubt concerning their magnitude.
This is why it is also essential to regulate them and plan their deployment in a thoughtful and consensual manner. To do this, it is necessary to develop adequate digital literacy among those called upon to implement their introduction in the various work environments of the economy.
I will not address the issue of regulation here, as Canada is in the process of introducing its own regime through the Artificial Intelligence and Data Bill, currently being studied in the House of Commons, to focus on the issue of planning, which is closer to each and every one of us. Indeed, while our societies are in the process of regulating the AI industry, they have not yet developed within the population the knowledge necessary to thoughtfully implement AI systems in the environments that constitute it: the world of work and the health, justice and education systems.
This lack of literacy is currently reflected in debates full of catastrophic or naive remarks.
Angelists dream, for example, that the introduction of AI systems will free workers from boring and repetitive tasks, allowing them to concentrate on projects requiring reflection, evaluation and creation, thus promoting human flourishing. However, the first echoes we hear from environments where we have begun to replace routine tasks with complex and creative work rather suggest that the so-called boring and repetitive tasks are those where workers can relax during their working day. work, while complex and creative tasks generate a significant cognitive load, difficult to sustain for an entire day, as well as increased responsibilities for the results expected by the employer, leading to higher stress.
Conversely, catastrophists fear, for example, that the arrival of AI systems heralds the end of postsecondary education as we know it. These systems promote, according to them, laziness among the student population. While we must certainly find ways to prevent their introduction from transforming post-secondary institutions into factories where a checkbook and a subscription to the most recent version of ChatGPT are enough to obtain a diploma, we are already meeting students who use these systems judiciously, demonstrating a deep knowledge of how to interact intelligently and creatively with these systems, thus giving us a glimpse of the possibilities of assisted human intelligence.
For the summer, I invite the readers of the Duty to develop their knowledge of generative AI and account planners, to provide training on these systems in their respective environments and to promote informed discussions on them.