Some researchers were still able to conduct their tests in early 2020, even as cities all across the world started to lock down in response to COVID-19.
Like everyone else, they had been forbidden access to their labs, but they managed to access “cloud laboratories” and submit their trials from a distance, relying on automated instruments and robotic arms to carry out their commands.
As software, robotics, and artificial intelligence (AI) have combined to bring the idea of “work-from-home” to scientific experimentation, what was once a charming convenience during a crisis has become a commonplace reality. Commercial cloud labs have already started to reverse conventional scientific procedures globally, to the extent that samples go through robotic channels rather than researchers hopping between their instruments.
Self-driven labs go one step farther in this regard. These autonomous labs can go beyond simply carrying out commands to actively creating them by integrating AI right into the system. In addition to identifying novel experiments and executing them with robotic infrastructure, these intelligent automated systems can also analyze the data and make decisions based on feedback. It is possible to condense the lengthy cycle of experimentation into an ongoing feedback loop during this phase.
All of this will have the immediate effect of drastically accelerating the pace of scientific advancement. The ability to condense a year of human study into a few weeks or days allows for the parallel exploration of hundreds of experimental permutations. Failure is inexpensive in such an environment, and it is not only feasible but unavoidable to find through constant iteration.
These capabilities have the potential to significantly alter the economics of scientific endeavors in domains including materials research, protein engineering, and medication development.
But as we have learned time and time again, every effort to lessen friction frequently results in unexpected effects. By speeding up scientific research, are we unintentionally putting ourselves at risk for problems that we have not had to worry about up to this point?
It is equally easy to discover chemical and biological agents that can cause illness as it is to identify the remedy for a sickness using any AI system. I described MegaSyn, a machine-learning system created to find previously undiscovered substances with a high likelihood of healing illnesses, in a recent piece.
The system ended up producing a list of incredibly deadly substances that were not only more potent than the most toxic chemical agents known to us, but also practically untraceable because many of them had not yet been discovered. This was because it was removing from a long list of suitable molecules those that had toxic side effects.
Despite how horrifying this may sound, Megasyn only detects potentially harmful materials. Someone would need to take those theoretical formulas and turn them into real items in order to use this knowledge to truly create dangerous biological compounds.
In addition to having access to a fully furnished laboratory, this calls for staff members who possess the knowledge and moral ambivalence to use the facility regardless of the repercussions. The advent of autonomous laboratories will soon remove this obstacle.
This risk is not fictitious. The majority of biological AI systems have little regulations. A lot of them are open-source. Few include significant protections. Even though they can conduct extremely potent tests, today’s cloud labs operate in a regulatory gray area.
It will be difficult for legal frameworks like the Biological Weapons Convention to adjust to this new AI reality because they were created for a time when the only ways to produce biological chemicals were through physical facilities and human-controlled research.
Nevertheless, autonomous cloud labs provide us with previously unheard-of avenues for clinical research. If handled properly, this could enhance our capacity to create medicines that save lives and allow for mass customization of care. Despite the possible risks, there are many reasons to try to figure out a safe way to do this.
We must immediately update our treaties and alter our laws if we hope to attain this delicate balance. However, we can not end there. We must incorporate accountability into automated laboratory systems from the very beginning. AI agents must be able to identify, audit, and trace experiments they design, carry out, and improve back to human decision-makers.
Because cloud labs make research resistant to physical disturbance, they have made remote science possible. By doing this, they have also eliminated a lot of the frictions that were, without our knowledge, protecting us. AI’s quick development has not only sped up this process but also made it possible for science to advance significantly.
There is typically a brief period of time between the development of a new technology and society’s realization of the potential risks. During this time, it is permitted to operate unchecked and without official authorization. Given how quickly AI is developing, that window is far more important than most of us know.
Furthermore, it is imperative that we maintain the highest level of security, especially in the case of self-driven laboratories. We must ensure that it is never too wide for disastrous consequences, in addition to swiftly closing it given the potential hazards.
