Divyansh Kaushik is Associate Director for Emerging Technologies and National Security at the Federation of American Scientists.
Federal Newswire
What does the Federation for American Scientists do?
Divyansh Kaushik
A lot of the people who are reading the news probably find the Federation of American Scientists in the context of nuclear weapons. The scientists who worked on the Manhattan project founded this organization as the Federation of Atomic Scientists. But over the last 75 plus years it has focused a lot on reducing harms posed by nuclear weapons, working on nuclear disarmament, and nonproliferation. Helping with the treaties that we've had in the past including, New START.
The lines between national security and economic security have gotten more blurred. I think we've started to think more in the broader mission and that has to include science innovation and technology. How are we thinking about a stronger and safer America, which is what we want, and how we think of the emerging technology landscape that we're finding ourselves in today.
How do we think of the competition with authoritarian regimes like the Peoples Republic of China, which wants to use these technologies to conduct genocide, whereas we want these technologies to uplift the word? We are not in this competition to make someone lose. We're in it because we believe that the world wins with us and to find an organization for me was my life's mission.
Federal Newswire
When it comes to technological development and advancement, how do we make sure it aligns with human values?
Divyansh Kaushik
It's very important to ask whose values we are thinking about. Those cannot be the values of authoritarians who want to suppress their people. The values that we should be enshrining into our technological principles should be of individual liberty and freedom for all.
We want people to live their best lives and have a prosperous future. Not one that is pre-written by Xi Jinping--that if you're born in a particular family that follows a particular religion or live in a particular region this is your future.
We want technology to increase our GDP enough that we can start solving things like finding a cure for Alzheimer's. We want to increase productivity to make sure we are reaching the next frontier and are empowering everybody along the way. We are an imperfect democracy but we always aspire to be better and that's what these technological values need to enshrine.
There are people who have legitimate arguments about whether we are only looking at western values. If you’re looking down upon the global south, those are reasonable arguments to have.
There are some aspects but there are underlying fundamental principles of equal rights, liberty, justice, and freedom for all, those are indisputable. Those aren’t the only disputes there are between democracy and authoritarianism, so which side do we want the tech companies to fall on? I think that's a question they have to answer for themselves.
Federal Newswire
Is it possible to have an authoritarian-free supply chain for AI?
Divyansh Kaushik
The supply chain for the AI stack today comprises three different things: data, semiconductors, and talent or people. On the data side, one of the reasons you're seeing all these large language models being developed in-house in China is not as good as the models that we have here. They do not want to use all the data that we are able to use because that data may have something about Tiananmen Square or something relating to Xi [and] Winnie the Pooh.
That is one of the biggest factors that we've seen right now. We will have to see what comes out of China, but right now our models, even the Mandarin specific models developed in the United States are outperforming the ones developed in China for that particular reason.
I think that's very important and to have an authoritarian free supply chain I think is rather hard, because of the kinds of disinformation campaigns that we are seeing flood free media. These are spoiling the social media that we have today. If you look at TikTok, how much of that content is just engineered by the CCP? How much of that content is designed to influence people to be against the ideas that we stand for in this country. It’s hard to filter out that content.
But so far we've been saved from that in terms of large language models because they're primarily dealing with text data. The majority of the text data that we see is coming from things like Wikipedia, common crawl, or the web at large where there are a variety of opinions. There isn’t that big of a selection bias that you would see on say social media like TikTok. But once we start to get into those videos on audio it's likely that we will start seeing that pop up as well.
Federal Newswire
How can America compete with a country like China if it is unencumbered by ethics?
Divyansh Kaushik
To first address China, they could get ahead by stealing this technology. I think they have been and that has been a big concern for us for a long time. That was also the impetus behind the presidential proclamation last year which prevented certain graduate students from Chinese universities with civil-military fusion ties from coming to the United States. The current administration has continued implementing that proclamation.
At the same time I will go back to President Reagan when he issued the national security decision directive 189 which said, “we grow because we keep our basic fundamental science open.”
Now you have specific national security imperative technology or side-effect development that should be classified. [We should] try to keep whatever we can open, but whatever is national security imperative let's classify that.
That approach has driven us to a success that the USSR could never replicate because they were a fully closed society…Now how do we see that in respect of China stealing all our technology?
If someone in China wants to steal some large language model, are they going to trust the outputs of that large language model? Will there be an alignment with the CCP? No, it is easier to steal hard tech than to steal information tech.
As for the people's side of things, we have about 60% of our computer science PhDs going to international students, and that's a blessing. Our population is a quarter of China’s and even if we increase our population by 1-4%, China has to only increase by 1%. For every four people we train with this computer science PhD they only have to train one. That is really hard for us to match with our population alone, so we have to have our allies and partners join us in the rest of the world.
The biggest supplier of international students to the United States has become India rather than China. India, Saudi Arabia, South Korea are all our allies and partners, but obviously we have to ensure that we are keeping them in a way that is safe and secure for us.
Look at Xi Jinping's speech at the 20th Party Congress last year. He emphasized how talent was the primary source of innovation now.
We very much say that attraction and retention of top AI talent around the world is a priority for us, and our allies are taking note as well.
The UK created a new high potential individual list for graduates of top universities from the U.S., offering a decision on their application in two months or less. Beijing, at the same time, has doubled its funding for STEM education in the last decade. Some of their provinces are now requiring elementary and middle schools to study AI while they're already graduating more STEM doctoral students each year than us. It’s double the number of STEM graduates overall than the United States, so obviously it is a battle of the brains.
Federal Newswire
How does the private sector compare with the government in this balance between national security and development?
Divyansh Kaushik
The private sector is the biggest driver of technological innovation in the United States today. That was arguably not true back in the days of the Cold War. But, at the same time, look at apparatus within the US government, whether it's DARPA or the Department of Energy National Labs; much of the research that the private sector is capitalizing on is coming out of those.
The basic research that the federal government invests in is what is driving private sector growth. If the private sector is the biggest developer of technology, [government is] still the biggest procurer. I think that symbiotic relationship has benefited both. But there are warning signs.
It is not as easy to constrain technology that should have to be classified, which is probably why the administration in its AI executive order decided to use the Defense Production Act to have the industry tell the Administration when they're developing something that is dual-use in nature. That way at least we are aware that it is being developed, and if we do need to classify it we will be able to. If we do not know that someone's developing it, then I think that will be a problem.
There's another warning sign where China's R&D budget has grown so much. In 2022, they spent about $65.1 billion US dollars in their R&D activities, which was about 70% of their overall science and technology budget. A 34% increase from 2018.
At the same time the United States's R&D expenditure has not kept pace. We’re no longer the world's leading spender in R&D based on GDP.
If China starts to think of science and technology as their comparative advantage–because they do have the talent pool and they have the resources to spend on–they can do whatever they want. Their state-run economy is a challenge for us. Obviously we cannot keep spending like China, but we have to identify strategic areas where we have to invest in.
I think the Chips and Science act was passed in a very big bipartisan way and laid out some of those key technology focus areas that the United States should focus on including AI and quantum. These things that are of critical nature to our national defense. I think if we continue to prioritize and bring the best minds together we will still win.
Federal Newswire
How would a practice of ‘reciprocity’ with China work when applied to the field of advanced technology?
Divyansh Kaushik
I think the answer to that will very much depend on the use case. We still want to be the shining beacon of freedom. We are not going to stop the media from around the world from reporting on American events. We do not censor, even if China does not want to let our media report events accurately and freely. At the same time when it comes to technology collaboration there are areas where we do not and should not collaborate.
Should we be working with a military civil fusion institute on some technology that could potentially be dual-use? No. But, there are certain low-hanging fruits where we've collaborated with Chinese nationals, and that is something we also did during the Cold War with Russian nationals. For instance, there was a recent paper where a group of US and Chinese researchers developed a new AI model for early detection of pancreatic cancer. Now that is a very difficult problem in medicine to solve.
If their interests are going to be to use whatever we are developing for the People's Liberation Army then it should be a no-go.
Whether it's AI or quantum, there are areas that should be off limits. The science and technology agreement with China expired last year. The Administration renewed it for another six months to give themselves some leeway to negotiate. I think that's a good opportunity for the to try to strike a balance … where we are collaborating with China.
What conditions are we setting? What should we want China to do in terms of clamping down on intellectual property theft? Why is it that someone can just take something on a thumbdrive [from the U.S.] and get it patented in China.
Those are things that the Administration needs to work through. I don't see a one size fits all answer there, I see it as use case to use case based where we have shared interests and where we do not.
The China Desk podcast is hosted by Steve Yates, a former president of Radio Free Asia and White House national security advisor.
