Artificial intelligence is everywhere, including orthodontics, but is it just a tool for treatment planning or our future replacement?
By Steven Martinez
Ever since Hal 9000 turned on the crew of his spaceship in Stanley Kubrick’s 1968 sci-fi opus 2001: A Space Odyssey, we have been awaiting the advent of artificial intelligence (AI) with wary skepticism.
What could it mean for humanity if our computers became too intelligent for us to comprehend or stopped listening to our commands in favor of their secretive ulterior motives? Would AI bring us to the dystopian future of The Terminator or become a helpful, all-knowing assistant like the ship’s computer in Star Trek?
For the longest time, AI was strictly in science fiction, but in 2023, we are starting to see what real, commercial AI is and is capable of. Through communication tools, practice management software, and treatment planning, AI has already come to orthodontics. But what does it mean for software to have AI? Is it just a marketing buzzword, or is there something more to it?
What is AI?
If it seems like we have been suddenly dropped into the era of AI, it’s because we are finally beginning to see the fruit of a long process that began millennia ago with the invention of mathematical algorithms. These ancient algorithms are a set of mathematical instructions that are used to solve complex problems, many of which are still used today to solve problems in engineering, science, or software.
James Jordan, distinguished service professor of healthcare and biotechnology management at Carnegie Mellon University, explains that AI can be broken down into three permutations or stages: software that uses simple if/then decision trees, machine learning, and deep learning.
Something like Google’s search algorithm can be seen as this earlier form of AI that has been iterated over time to improve search results. Machine learning, however, analyzes data to understand a particular subject, and the more data given it, the better it becomes at recognizing what it is searching for. Lastly, deep learning is a more advanced version of machine learning capable of taking in vast amounts of data and then using what it learns to improve its analysis of future data without human intervention. Deep learning aims to mimic how the human brain processes data so that a computer can generate new ideas as it improves.
Today, most commercial AI is a mixture of the first two stages, algorithms and machine learning, says Jordan. As access to large amounts of data has improved through the internet or collected records, AI companies can finally “train” their systems to become valuable assistants.
“Think about making a flowchart going from left to right in a tree diagram. Traditional software can only use the pathways in the tree. When you give it an input, it can only go ABCD,” says Jordan. “What you’re doing with artificial intelligence is asking questions to a database that isn’t plotting ABCD as an output. It’s using artificial intelligence to say that based on the information in this database, here are the potential pathways you could consider. It’s more dynamic and less fixed.”
Using AI as an assistant
SoftSmile is one such company that is bringing AI to orthodontists. The company’s Vision software uses AI to analyze patient data and help doctors plan their treatments.
“What Vision does is it prepares a treatment planning model of how the teeth must be moving in order to achieve a good result or the desired result,” says Khamzat Asabaev, founder and chief executive officer of SoftSmile. “To build this model, which we understand must be precise and efficient, the system has to take into account various data like resistance of plastic, potential collisions, or what would be the best and most efficient way to move a tooth from point A to point B.”
Vision software was trained using large datasets of patient mouth data to learn to create a first draft treatment plan for an orthodontist. Using mathematical algorithms, the software used the data to improve over time, allowing critical parts of the treatment planning process to be automated behind the scenes while orthodontists use the software.
Jordan and Asabaev reiterate that AI is only as good as the data given to it, and more data means more opportunity to finetune the results.
“I think you can have the technology, but without the right validated data sets to run it, then you’re problematic,” says Jordan. “Garbage in, garbage out, right.”
According to Jordan, in the broader medical field, data can be abundant with resources like the National Cancer Database or heart disease database that AI companies can use to improve their AI models. However, he believes that in orthodontics, case data tends to be more siloed than in general medicine, potentially limiting its scope.
To build up its AI, SoftSmile partnered with larger companies that have access to thousands of cases to improve their AI’s capability, says Asabaev.
“SoftSmile was lucky to partner with large corporations who make thousands and thousands of cases in a month, and we do have access to that data through our partners,” says Asabaev. “That’s how we teach our algorithms to get even more precise, to get better, to get more efficient.”
Asabaev estimates that Vision software can cut down treatment planning time from an hour to 3 to 5 minutes, reducing the time it takes by 95%.
Who watches the watchmen
As AI has made its way into our lives, it hasn’t all been smooth sailing. As much fun as you can have asking ChatGPT dumb questions or prompting DALL-E to make an absurd picture, it also brings up more significant questions about what it means to trust an AI in a medical setting or, for some, job security.
For the past few months, a writers’ strike has shut down the film and television industries, and one of the core topics of negotiation surrounds the use of AI and who should get credit for the ideas that AI generates. But, regarding healthcare, the question of how accurate an AI diagnosis can be could have life-altering ramifications.
Time and time again, regulatory bodies and lawmakers tend to be behind the curve when dealing with new technologies, and AI has proven to be the same. It often takes a problem for the government to address it finally.
“I think right now we’re in the Wild West,” says Jordan. “Our dean [at Heinz College] always points out that technology is far faster than the laws and policies regulating it. I think we’re going to see moments in time, like when Uber was using these automatic cars, and somebody got hit. What does that mean? Who pays for it? What’s the rules?”
Asabaev points out that AI can never be perfect, even in an orthodontic setting, with as large a dataset as possible. Just like an autopilot can accomplish 95% of the flying in your typical commercial flight, the experience and expertise of a pilot are still necessary to deal with anything that might be outside the norm. Air travel, like biology, takes place in an inherently chaotic system, and even something as seemingly straightforward as moving teeth involves too many unique parameters to be completely boiled down to a series of simple mathematical calculations.
“In healthcare, you cannot achieve astronomical precision,” says Asabaev. “We cannot say that it will be 1,000% accurate, but what we can claim is that the algorithms or solutions that we introduce are like calculators or autopilot. They help doctors to get to the precision, and we can only calculate the time that we’re saving them.”
Several orthodontic companies, not just SoftSmile, are touting the use of AI to assist doctors and the number seems to be expanding every day. Importantly, in each case the orthodontist retains full control with the AI acting as a tool to improve efficiency. The FDA has already approved SoftSmile’s Vision software, and Asabaev reiterates that it was never intended to replace the expertise of a doctor but to speed up the laborious and more mundane aspects of treatment planning. He says that SoftSmile has plans to eventually use machine learning to help the software adapt to an orthodontist’s preferences for treatment, a tailored solution for each doctor.
What, me worry?
But as AI grows in complexity, discussions of its capabilities tend to become more philosophical.
World No. 1 ranked chess grandmaster Garry Kasparov famously played a series of close chess matches with IBM’s Deep Blue chess computer, beating Deep Blue the first time in 1996 and eventually losing the next year. It marked a milestone in computing where machines exceeded humans in one of the oldest and most complex games. Today, chess computers are far better than the best grandmasters, and it isn’t even competitive. Could AI in orthodontics reach a similar inflection point, where treatment planning software is faster than a doctor and capable of better outcomes?
Asabaev doesn’t believe that to be the case. He points out that while computers may best humans at chess, they are still limited by their knowledge, which is solely based on a superior understanding of the rules. In problems where creativity is critical, human ingenuity is still outside the purview of AI.
“Can ChatGPT write War and Peace? I think yes. Why not? It’s possible,” says Asabaev. “But I will never agree that in the existing moment in time if you put some unorthodox problem in front of a human and a machine, the machine will be better because it still lacks creativity. The difference between Leo Tolstoy and ChatGPT is that you can give just one sentence to Leo, and he will write a book, while ChatGPT will always be restricted by data you had provided to it before.”
Undoubtedly, as AI improves, it will find more ways to make our professional lives more efficient and, in specific tasks, to exceed human capability. As for what that means for society, for good or ill, that may still be ultimately a human problem. Or, like John Connor says in Terminator 2, “There’s no fate but what we make for ourselves.”OP