Our inability to multitask, at least more complex tasks, has been a great source of humor over the years. However, the fact of the matter is that we humans spend a vast amount of our lives multitasking. From talking while we eat together to driving while listening to music, we regularly engage in lots of forms of multitasking.

The truth is that we need to, thanks to our language abilities and our ability to live in enormous social groups, our brains are constantly processing huge amounts of data. This includes everything from understanding speech, reading, talking, moving, and undertaking all kinds of complex tasks like using tools, etc.

All this requires a very sophisticated brain that is capable of learning a huge range of tasks, storing immense amounts of data, and processing the data that we are constantly being fed from the world around us.

But what are the implications of this when trying to develop and train artificial intelligence systems to better understand us? Well, as Bill Murray’s character in the film Caddyshack (1980) hilariously points out, “in order to conquer the animal, I have to learn to think like an animal. And, whenever possible, to look like one.”

Well, while most of the AI systems that we are developing now don’t need to look like us, the need for AI and ML systems to ‘think’ like us is imperative if they are going to master the jobs that we ask them to do. Central to this is the ability of AI to multitask.

Teaching AI to Multitask

In a recent article, MIT Technology review announced that a team at Meta AI had developed an algorithm that allows data scientists to train an AI neural network to recognize images, text, or speech. While these tasks are already being undertaken by different AI algorithms, what is special about this breakthrough is that, for the first time, it is a single algorithm that can be used for all these tasks.

Known as Data2vec, the algorithm “unifies the learning process” while being able to accomplish all the tasks to the same level of success. It works in two stages, much like a student and a teacher. To do this, it utilizes two neural networks, where the first, the teacher, is trained in the normal way to understand images, sounds, speech, etc., thereby allowing it to identify meaning from these data sets.

However, the ingenious part comes when the second neural network, the student, is then trained to predict the “internal representations” of the teacher meaning that it tries to predict what the teacher will see when it hears a particular word or sees a specific image, etc.

In essence, what this does is to allow the entire AI system to understand the world in a deeper way by increasing its perspective, and thereby, increasing its understanding of the ‘why’ behind the tasks it is undertaking.

This breakthrough in approaches to self-supervised learning has enormous implications for all current applications of artificial intelligence and machine learning systems.

While we don’t have the time to go into every example, the impact of this advancement on the movie-making process gives a good idea of just how big a splash AI systems with the ability to multitask will make.

AI-assisted Moviemaking Assisted by Multitasking.

The most obvious advantage that AI and ML systems will gain from the ability to multitask is a boost in the accuracy of the results that they are able to achieve.

This will be the direct result of the system’s ability to gain a better perspective of understanding s to why certain trends come about. So for example, while the ‘teacher’ will be able to pick the most suitable actors for a certain role, the ‘student’ will be able to learn a deeper, or more intuitive, understanding of why those actors might be better for a role than others.

By combining this deeper level of insight with the AI system’s ability to master multitasking on a range of disciplines, which in the case of AI-assisted filmmaking, might include analysis of the script and how an actor delivers their performance, the camera shots chosen to relate this performance, etc., the AI system would be able to undertake a far more accurate appraisal of how audiences will react to these elements to give more accurate insights to directors and producers.

Such systems would be able to much better understand why, for example, simply casting Robert De Niro in a film, despite it having a bad script or being poorly shot, etc., would not guarantee that the movie would be commercially successful. Indeed, such a level of understanding would allow AI systems to delineate between what makes a commercially successful film and what makes a critically acclaimed one, something current AI systems are unable to do.

The second important advancement that will result from the development and implementation of AI systems that are capable of multitasking is the rapid development of powerful new tools to help filmmakers and producers.

Without the ability to combine different disciplines such as voice and image recognition, current AI tools are limited in the range of functions that they can offer.

The implementation of AI multitasking technology will allow for the first tools that can accurately understand an actor’s performance, or indeed, the misse-en-scène of a film as a whole. An AI system that had multidisciplinary capability would be able to analyze elements ranging from how and what an actor was saying, their facial expressions and movements, and even elements such as what they were wearing, all to gain an accurate assessment of their performance.

By cross-referencing these insights with historical data, AI-assisted movie-making systems would be able to give insights relating from how an actor could improve their performance, for example, be scarier or more animated, to how much specific audiences would react to the character being performed in that way. Such insights would also help to improve current tools such as the suggested actor for a role feature.

This exciting new approach to AI development is filled with promise. The possibilities are endless, we await the development of these systems further to see just what new powers they will give to artificial intelligence systems and to AI-assisted filmmaking tools.