Commentary
Can you name one great innovation funded by OpenAI? How about one funded by Google? There are plenty. But where are all the innovations from the National Science Foundation (NSF)?
President Biden’s executive order on AI expects the NSF to work to “promote the adoption of leading-edge privacy-preserving technologies by federal agencies.” The NSF controls around $10 billion of taxpayers’ money, according to their website, “to promote the progress of science,” and it “envisions a nation that capitalizes on new concepts in science and engineering and provides global leadership in advancing research and education.”
It has the power to fund up to $100 million or probably even more in research infrastructure costs. Why, then, are general-purpose widely impacting breakthroughs similar to OpenAI’s ChatGPT not coming out of the funds lately? Why does the NSF website not at least list some major widely used products it funded in the last decade?
There have been many critical scientific phenomena, such as the mental health issues caused by social media, needing immediate attention in the last few years. Why is the NSF not at the center of solving these issues? What ails the NSF?
A simple survey involving human subjects requires Institutional Review Board (IRB) approvals and whatnot in order to get published. A huge social experiment called social media was rolled out, and it is now causing mental health issues, addiction, and suicides. The NSF could have easily funded studies to predict the impact and propose solutions.
Apparently, the NSF is not getting its priorities right. For instance, a project titled “The Development of Computational Thinking among Middle School Students Creating Computer Games” was awarded more than a million USD ($1,092,908.00) in 2009, and the same Principal Investigator was awarded $701,767.00 in 2014 for a study on “Can Pair Programming Reduce the Gender Gap in Computing? A Study of Middle School Students Learning to Program.”
I do not think these projects made any difference to the middle school students even after nine years.
A study published last year found NSF grant funding to be racially biased. Having been at both ends of the grant funding process—as a reviewer and a reviewee—I see substantial scope for improvement in the way the grant application processing works. The turnaround time for a review of a grant application to the NSF, at least in my case, was more than six months.
Vishnu Pendyala, assistant professor in the Department of Applied Data Science at the College of Professional and Global Education at the San Jose State University campus on May 8, 2023. (Jim Gensheimer/Courtesy of Vishnu Pendyala)
In rapidly progressing areas like generative AI, that kind of timeframe can obsolete the proposed ideas and create an entirely newer body of scientific knowledge. Some of the comments made in the reviews that I received reflect upon the ignorance of the reviewers. Apparently, there are no binding criteria for selecting the reviewers, nor is there a mandate for recusal when the reviewer does not have the knowledge to judge the proposed ideas.
Although they perform the most important function of the NSF, namely, help with the adjudication of the grant applications, the reviewers are neither accountable nor sufficiently paid to demand accountability. It is widely rumored that you need to be in the circles of the adjudicating reviewers and the program directors for your project to be funded by the NSF.
Apparent conflict of interest is taken seriously, but not latent biases such as ethnicity, affiliation, or even area of research. Equity, diversity, and inclusion are given priority, but is it being taken too far so that merit is being sacrificed at the altar of social justice? Science is advancing rapidly. How pragmatic and even ethical is it to take months to reject a researcher’s ideas, that too with flimsy review comments?
This question is not just for NSF, but to all those publishers and funders who sit on applications and research papers for months. Can the lawmakers step in to make a difference?
There may not be many researchers who turn into lawmakers or lawmakers with researchers in their circles, but researchers are an important community that lawmakers represent. In the long run, the economy is primarily driven by innovation, particularly in science and relevant fields.
There must be a wide referendum on the practicalities of current research adjudication processes, and legislative action must ensue from it. To start with, the law can make it mandatory for funding agencies—and for that matter, even journals—to announce review turnaround times.
Reviewing plays a substantially important role. It is well known that Google’s PageRank was not accepted for the SIGIR conference in 1998 and Einstein’s theory of relativity did not win the Nobel Prize in 1921. But the scientific community hasn’t learned from any of such incidents.
Reviewing must be made a top priority, paid, and made accountable. Manual peer reviews must be supported and complemented by AI-based tooling. Compared to the bias in AI models, human bias is complex. A number of factors such as ego, patriotism, and nepotism compound the problem.
AI models are devoid of such factors, even if they are impacted by other kinds of bias. President Biden’s executive order makes it a priority to combat bias in AI models and algorithmic discrimination, and it had better start with the Fed’s use of AI.
AI-based reviews have a much faster turnaround, and the technology is mature enough to give it a try. Significant projects and huge funding must have multiple rounds of oversight.
In the days of increasing science skepticism, it is crucial that the NSF lives up to its mission “to promote the progress of science.”
Views expressed in this article are opinions of the author and do not necessarily reflect views of his employer or any other entities that he is affiliated with.
