Humans have a hard time adapting to a fast exponential growth over a timescale of years, especially in academia which is used to traditional practices shaped by centuries of slow changes. My own Harvard University added a small number of courses for how to use artificial intelligence (AI) in studies and research, but has yet to adapt intellectually to the full impact that AI will have on scholarship.
Mentoring students who employ AI for literature search and paper writing is challenging. Last summer I reviewed a paper drafted by a student and found hallucinated references and scientific statements. It was easy for me to spot them because I am thoroughly versed with the scientific literature in astrophysics after forty years of practice. But I also wonder how many non-existing references and incorrect statements are currently being published in scientific journals because the referees are fooled by hallucinating AI?
In the latest meeting of my research group, we discussed how quickly human inferences will be dominated by input from AI systems. Within a few years, humans might be manipulated by AI far more effectively than they were influenced by social media until now.
Forty years ago, the ability to code defined a precious skill set for a substantial fraction of the scientific community. Numerical simulations bridged between first principles and experimental data when dealing with complex systems that do not adhere to simple analytic models. Reproducing the experimental results with computer codes that solve the fundamental equations provides new insights into the underlying processes and demonstrates that no new physics is required.
At a conference held six months ago in celebration of the 20th anniversary of the Institute for Theory & Computation for which I serve as director, a world renowned code developer noted that the number of lines in state-of-the-art cosmological codes grew from a few hundred lines to tens of thousands of lines, making it difficult for a single code developer like himself to write bigger and better codes in the coming years. During the Q&A session, I pointed out the historic analogy with the concern about horse manure filling the streets of London over a century ago. At the end of the 19th century, there were more than eleven thousand hansom cabs and several thousand horse-drawn buses — each needing twelve horses per day, making a total of over fifty thousand horses transporting people around London. Each horse produced about ten kilograms of manure per day. In 1894, The Times newspaper predicted… “In 50 years, every street in London will be buried under nine feet of manure.” However, the invention of cars at affordable prices by Henry Ford resolved this problem in less than two decades. Another historic example involves the limited landline phone service in Bulgaria towards the end of the 20th century. At that time, it was estimated that the installation of phone poles to service all citizens might take 50 years. However, the widespread availability and affordability of cell phones replaced landlines and connected most citizens within a decade. Following these examples, I suggested that AI might replace cosmology code developers in the coming years. In reply to my suggestion, the speaker argued that current AI systems make a lot of coding mistakes and will not be reliable substitutes to code developers any time soon. Despite his absolute confidence, my personal forecast remains that current limitations in code development will disappear within a few years, making human code developers a relic of the past. The resistance to admit rapid changes is not unique to academia. After all, non-avian dinosaurs did not imagine their demise by the Chicxulub impactor sixty-six million years ago.
Dario Amodei, CEO of Anthropic, noted recently that scaling current trends in computational power and training sets with quality data, may bring AI beyond human cognitive performance on many fronts within the coming years. This brings another challenge to academia: if future breakthroughs in research will be guided by superhuman AI, how would tenure decisions be made? Should junior faculty be granted tenure if AI dominated over their natural cognitive abilities in fostering their accomplishments? Or does the situation resemble sport, where the use of drugs to enhance physical performance is forbidden because the underlying goal is to test the natural physical abilities of athletes?
At the end of the group meeting, I had asked my students and postdocs whether they are worried about AI in their future? They replied that they mostly worry about their job prospects in the coming year.
Personally, I am very curious about our future following the looming AI revolution. One way to learn about it is to seek extraterrestrial technological civilizations that are ahead of us because their host star formed before the Sun. The Galileo Project under my leadership is doing just that. Another approach is to board a spacecraft that travels close to the speed of light, where time is ticking more slowly than on Earth. Upon returning to Earth, the astronauts would cash the benefit of time dilation by witnessing our technological future. In this vein, I am particularly jealous of photons which propagate at the speed of light and have no sense of the passage of time whatsoever. But I would be content moving at the speed of massive neutrinos emitted by the Sun, which age by a second in their rest-frame for every century on Earth. At this speed, astronauts could have witnessed all 13.8 billion years of cosmic history during a few years of their life.
ABOUT THE AUTHOR
Avi Loeb is the head of the Galileo Project, founding director of Harvard University’s — Black Hole Initiative, director of the Institute for Theory and Computation at the Harvard-Smithsonian Center for Astrophysics, and the former chair of the astronomy department at Harvard University (2011–2020). He is a former member of the President’s Council of Advisors on Science and Technology and a former chair of the Board on Physics and Astronomy of the National Academies. He is the bestselling author of “Extraterrestrial: The First Sign of Intelligent Life Beyond Earth” and a co-author of the textbook “Life in the Cosmos”, both published in 2021. The paperback edition of his new book, titled “Interstellar”, was published in August 2024.
