Dr. Jeffrey Lupker is the CEO of Staccato, which is a platform of generative AI tools for musicians and lyricists which helps them to remain in an uninterrupted state of creative flow as they create. By allowing users to input a few keywords, lyrics or bars of music, Staccato can suggest ideas on what might come next to help overcome writer’s block and inspire new creative directions. Dr. Lupker holds a PhD from The University of Western Ontario’s Don Wright Faculty of Music where he is also on contract faculty in the Music Research and Composition department. Beyond Staccato and his research, Dr. Lupker is an active performer on primarily guitar and keyboards and has played across Canada and USA with singers such as Juno Winner and Canada/America’s Got Talent finalist Stacey Kay and Canada’s Drag Race Winner Priyanka.
Q. Is the hype around AI overblown or are we underestimating its trajectory?
I don’t think the hype is necessarily overblown because I think all of the interesting tools people are building with all the various AI models that have come out recently are very exciting. Some, of course, are gimmicky and it remains to be seen which will survive the initial “boom” of these new products – but, certainly, many are here to stay and will be refined over time. I also don’t think we are underestimating AI’s trajectory because along with this boom of new tools came the understanding that AI has much to offer. While the nature of the rules and regulations regarding how AI tools will be part of our everyday lives and how the relevant companies will be allowed to operate, I think that the recent developments have fast tracked the discussion on how we need to adapt to this new way of life. Nonetheless, I do agree that there is a fair amount of hype regarding the positives of AI and there will need to be many important discussions surrounding the potential future negatives.
Q. What is something about AI that you think most people don’t realize?
AI tools only reflect what any AI system has been trained on and while language models may feel sentient, they aren’t. It may seem scary when language models act like they are aware of emotions and feelings, but that only occurs when the system is given a specific prompt that elicits that sort of response. The system isn’t consciously adding any extra layer of emotion because the more recent versions of AI systems have feelings. What people also need to realize is that the best way to get the most out of generative AI tools is to already have knowledge of the task you are trying to accomplish. For example, using AI to help you program in a specific coding language can greatly increase your productivity, however, that is especially the case if you already have underlying knowledge in that language. That knowledge allows you to catch mistakes, integrate things faster and just have essentially a junior AI-colleague that you’re working with. I would advocate that everyone learn how to write, code, paint, draw or compose music well, and use these new products to enhance your abilities rather than just relying on these tools to do all the work for you.
Q. Have you ever had an experience working with AI that surprised you?
Of course. That happens all the time. As I continue to perfect the AI tools for my own company, Staccato, there are many times where a tweak here and there gets considerably better AI generations when I input some of my own music. There are moments where it is very surprising how well the AI tools can work in some situations but it’s also very exciting at the same time. When I create music with it, I am also often surprised about some of the directions it suggests that I go in. It often suggests new ideas that I would have never thought of and it becomes a very useful creative tool from that standpoint. In fact, at times it feels like your own creativity can become almost limitless when using it. That’s not to say that the tool itself is creative, but rather it makes me more creative by expanding my options.
Q. When you see AI seeming to display characteristics that we could call creativity, what is actually happening?
In line with what I said earlier, I don’t personally think that any AI models are actually inherently creative. As such, people like to give the models more credit than they deserve. If I am using an AI image model to create something and I give it an in-depth prompt about what I want it to do and it comes back with something really interesting, who was the creative one here? Is it me who crafted a creative prompt, the data scientists who curated the training dataset that increased the likelihood of creative images occurring, the machine learning engineers who architected a model that effectively learned all the intricacies from the dataset, or the AI model which only makes high probability decisions based on all of the above? I think ultimately humans are still the ones responsible for the creativity we see when it comes to AI.
Q. How should we think about the difference between what goes on inside AI and the human mind?
This is a little above of my pay grade as I tend to stick to AI and music but it’s certainly the case that the human mind is much more complex than an AI model. The human mind is far from being well understood whereas the AI models that exist and how they function/operate are completely understood (at least by those who designed them). Humans learn in ways that are more flexible than that of AI models which are, essentially, explicitly told how to learn and what they are trying to learn. For example, given a particular task, you select a specific model type that can best learn from the dataset set associated with that task because not all model types are able to learn from that dataset well, if at all. Humans learn in much more broad ways than AI models. Furthermore, humans can learn/experience the world through senses like touch, taste, smell etc. which computers have no ability to do. As I mentioned above, AI isn’t creative, nor is it capable of feelings or consciousness. But again, I really don’t think I’m providing much on an answer to your question because some of these relevant issues delve into neuroscience, cognition and philosophy, areas that I really don’t have much background in.
Q. What about AI keeps you up at night?
That’s easy, how fast the technology changes. You spend a lot of time learning how to create effective models that can do different tasks that you want done but the next thing you know new technology has come out and you need to do something in order to keep up and adapt. Not only is it difficult to keep up with the changing software and abilities of AI but also the with hardware and how people want to interact with AI tools, particularly in terms of the different ways people want to use AI in order to create. Keeping up can be quite stressful but when people are using your AI tools to do something creative, it is also very rewarding.
Q. When you discuss AI with friends and colleagues, what do you like to debate?
In general, most debates revolve around how beneficial AI can be when we integrate it into our daily lives and how society might have to adapt to the changes its usage creates. These discussions are more about the more recent generative AI tools that you can consciously interact with and not the many machine learning algorithms that already exist behind the scenes. One relevant issue is that language models have become quite problematic for schools and universities because students can ‘cheat’ in writing essays or coming up with answers for questions, but the question then arises that, as these tools aren’t going away, should we be testing students’ knowledge in different ways? I am not advocating for a particular way forward here, but these types of situations do lead to interesting discussions concerning how future generations might best learn given the new technology they will inherit. The calculator may be a good analogy here: at times in school you were told you couldn’t use one for a test because you needed to show you had certain skills, but other times you were allowed to use one as the type of questions changed. Essentially, the change led to a new challenge, one is which you were now being asked to show how to best interact with the calculator in order to get to the solution. That the nature of the test changed from asking can you get the answer, to asking if you could get the answer using the new tools we had at our disposal.
