The Alignment Problem of Universities with AI
The alignment problem is often phrased as the fundamental challenge of ensuring that artificial intelligence (AI) acts in accordance with human goals and values, without adopting harmful biases or pursuing goals with unintended consequences.
This is often viewed as a one-side alignment, anchoring AI to the intentions of humans. However, in any relationship between intelligent entities, both sides have to adjust to the qualities of the other. Here, I focus on a particular segment of society that must adjust to the new world-order in the age of AI: academia.
Given the unprecedented performance of the latest AI system, it is time for universities to reassess their goals and structure. This is one example of the reverse alignment problem of humans with AI.
Universities are engaged in education and research and both will have to be transformed.
Traditionally, classroom education was focused on teachers delivering to students a body of knowledge accumulated in different disciplines, as summarized in textbooks. However, much of this knowledge is now accessible through AI systems that were trained on these textbooks and research papers. When I asked my Harvard College student about the radioactive energy output of carbon-dioxide ice, he found the answer within seconds by asking Perplexity which provided the answer with references to primary sources.
The future role of classroom education will concentrate on teaching students how to think. This is particularly relevant in the age of AI, when students are tempted to get answers to all questions from artificial oracles. Avoiding the need to walk by taking public transportation leads to muscle loss. Similarly, the availability of AI agents for accomplishing tasks that previously required thinking will inevitably lead to brain power loss. As reported here, I noticed in recent months that students get dumber unless they exercise their natural brain, made of flesh and blood.
The classroom must play a critical role in enhancing the cognitive abilities and critical thinking of students. The urgent goal of education would be to compensate for cognitive laziness, an unavoidable casualty triggered by excessive use of AI.
But what about research? Traditionally, the progress of research in the natural sciences relied on the employment of a large cohort of research assistants in the form of students and postdocs. Six years ago, I tasked a student in my research group to go through a NASA catalog of meteors and find if there are rare interstellar objects in it. The student reported back within a week and it took us a month to write the related paper. Last month, I gave my postdoc the same task regarding an updated version of this catalog. While we were still discussing the task, my postdoc asked the most advanced version of ChatGPT to do the job. Within minutes, the AI system produced a new code that analyzed the full dataset, recovered the results of my student from six years ago and gave new results regarding the latest data. Another AI agent also made the related plots and had a paper ready for us to submit for publication by the time we finished our conversation.
Of course, experimental work still requires humans and cannot be substituted by AI agents or robots as of now. At any event, the primary reason for working with students and postdocs will remain to groom them as the next generation of researchers. There is already less of a need for large cohort research analysts as a workforce required for accomplishing tasks that can now be fulfilled faster with AI agents. This means that the size of graduate student classes will need to be trimmed down to just fulfil the specific need of society for replenishing a steady cohort of human researchers. The researchers will ask the questions that AI systems and future experiments will help to solve.
This forecast applies to the natural sciences, but what about the humanities? As it turns out, there is a version of “humanities of the future” which is essential in the age of AI. Much of the societal disruption of AI has to do with the interface between humans and the machine. This interface involves new ethical and legal challenges that are related to humans and are therefore within the territory of the humanities.
Related questions include: How should the legal system treat fake AI-generated content, akin to identity theft and misinformation? What limits should be placed on AI surveillance and data privacy? Do corporations that train AI systems on specific content have legal responsibility to byproduct crimes in the same way that parents are liable to their underage kids? How should society put guardrails on emotional dependency and manipulation by AI companions that could potentially lead to human death or suicide?
The “humanities of. the future” should aim to focus on resolving these ethical and legal challenges, rather than focus on teaching the work of ancient thinkers who had no access to computers.
Over the past century, academia has made adjustments to societal changes. However, the alignment problem of academia to AI is destined to become exponentially more acute in a matter of years. The train is accelerating and if the operators will not stay in control, the passengers-carrying cars will likely go off the rails in the coming years. The time is ripe to act.
My university, Harvard, is primarily engaged with educating students and researchers on how to use AI for their classwork and research. But the university is slow in addressing the urgent strategic challenge of alignment with the new AI reality outside its control. Currently, the response of universities to AI is tactical, not strategic. Here’s hoping that University leaders will notice how fast the academic landscape is changing and come up with a strategy on how to align with it.
ABOUT THE AUTHOR
Avi Loeb is the head of the Galileo Project, founding director of Harvard University’s — Black Hole Initiative, former director of the Institute for Theory and Computation at the Harvard-Smithsonian Center for Astrophysics (2005–2026), 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.
Professional website:
https://lweb.cfa.harvard.edu/~loeb/
Social media:
https://avi-loeb.medium.com/
https://www.youtube.com/@ProfessorAviLoeb
https://open.spotify.com/show/1zhndXkvSY2b8FdjspFpCd
https://x.com/ProfAviLoeb
