In recent months, AI model developers have claimed that soon, artificial general intelligence (AGI) will eventually be able to solve all disease. AGI generally refers to AI systems being capable of human-level reasoning and critical thinking across multiple cognitive domains, a milestone which many model developers have been working towards eagerly. One key aspect of this milestone is the thought that once AGI is achieved, there is a reasonable level of confidence that the hardest cognitive burdens could potentially be offloaded to AI systems to find solutions. Why? Because successful AGI systems will likely be able to summarize the entirety, if not centuries, of human experience and scientific discovery into rapidly usable data points, hence being able to meaningfully create an unprecedented level of productive outputs. The other key premise is the belief that the cure for all, if not many, diseases is already amongst us with regards to the raw ingredients; it’s simply about finding the right permutations, combinations and scope of scientific discovery to utilize them meaningfully.
Take for example the traditional drug discovery process, which typically entails isolating genes or proteins, finding vulnerabilities, and then determining defensive structures and molecules that can potentially act to mitigate the negative impacts of the culprit elements. There is a significant amount of trial and error that goes into the process, and often, it is like trying to find a very specific key, in an ocean of keys, to fit a specific protein folding pattern out of billions of potential options. However, artificial intelligence systems can process information and data sets that far exceed the level of human capacity to understand interactions between proteins and molecular structures that have not even been previously fathomed.
A landmark article in the Journal of Pharmaceutical Analysis by Fu et al., discusses how AI helps accelerate and bring structure to what was previously considered “luck” in the drug discovery process: “drug discovery has historically relied heavily on serendipity, with many significant breakthroughs occurring through chance observations or unintended findings. However, AI offers the potential to remove much of the uncertainty in this process, dramatically improving the chances of identifying commercially viable drug candidates while reducing both costs and time.” The authors explain that machine learning’s ability to map relationships that were previously not even thought possible and to predict combinations that will be successfully aligned with biological complexity is something which simply cannot be underestimated.
Another study published in the journal Drug, Design, Development and Therapy found that AI enabled drug discovery actually helped lower costs and the time required for the drug development lifecycle through increased efficiency in patient recruitment, data analysis and clinical trial design. This is especially important because one of the key aspects of drug pricing is centered on the fact that the drug design and discovery process is incredibly resource heavy. Drugs are expensive because pharmaceutical companies often invest billions of dollars and many years to bring a specific formulation to market. However, if AI systems are able to cut some of the costs in that process, there may be an opportunity to eventually pass some of these savings onto consumers.
Of course, getting AI to this level or generally to the point where AGI can help cure disease to a scalable degree is not necessarily easy. There are also significant challenges to this process and the aspects mentioned above. As written in Machine Learning for Brain Disorders, there are numerous real world challenges and bottlenecks to consider. For one, clinical trials innately involve human lives, and no matter how fast AGI can accelerate the development process, trials still require human capacity and time. If anything, arguably, AI engineered trials could potentially take longer to execute, given the higher level of scrutiny they may require with regards to human life impact. Furthermore, AGI systems are only as good as the data that is provided to them. Unfortunately, the complex nature of human knowledge thus far has caused it to be relatively federated and across very disparate data sources. The good thing about AI is that it can be a unifying factor across different data sources. But it still requires access and guidance on how to use that information. Finally, there needs to be more cohesive policy efforts in this arena. While there are significant benefits to what AGI can achieve, guardrails and unified policies are essentially, especially as the world is entering a relatively new and uncharted chapter.
