Startups (Page 7)

The Economist has a good story on the rise of Chinese science. In terms of “high-impact” (i.e. highly cited) publications, China has soared past the US and Europe:

Now, these numbers have to be taken with a grain of salt. Measuring the impact of papers by looking at how many other papers cite them can be a biased measure of true impact — you can have a bunch of researchers who all cite each other copiously and thus inflate the metric.

Qiu, Steinwender, and Azoulay have a recent paper in which they argue that this phenomenon is especially common in China:

We highlight a novel source of bias in citations that is particularly relevant for cross-national comparisons: home bias, i.e., the tendency of researchers to excessively cite researchers from their own country…

We find that China exhibits by far the largest home bias among all countries. This is not a recent phenomenon. While China’s home bias has been steadily increasing over the past twenty years, Chinese citations were already strongly home biased in 2000, the start of our observation period. In addition, China’s home bias is not driven by any particular research field. Rather, China exhibits the strongest home bias in 18 out of 20 broad scientific fields…

Finally, we find that home bias has exaggerated the rise of China in science. While China ranks second behind the US in terms of raw citations, it falls back to the fourth position behind the US, the UK and Germany once we use our de-biased metric. Homedebiasing citation counts might be seen as especially informative if one believes that home citations are especially prone to reflect political or strategic considerations, rather than the acknowledgement of scientists cumulatively building on the ideas contained in the articles they choose to cite.

The researchers basically just identify “home bias” by controlling for a country’s size. Their measure of home bias still displays some apparent size dependence, with small European countries at the bottom end of the scale and large countries at the top end. So I do wonder if they controlled for size correctly. But China still definitely sticks out above all others, including India:

So while I think the authors’ conclusion that the UK is still ahead of China in high-impact science seems pretty suspect, there really does seem to be something going on here in terms of Chinese researchers citing each other an awful lot.

You can interpret this in a couple of different ways. One possibility is that Chinese science is just much more high-quality than people outside China realize and non-Chinese speakers fail to cite these high-quality Chinese papers due to the language barrier.

An alternative interpretation — which Qiu et al. suggest — is that China’s government told the country’s researchers to go out and write papers that get a ton of citations, and the researchers basically responded by establishing implicit or explicit citation rings. And, of course, it could be some combination of these two explanations.

(Update: In the comments, Zhicheng Lin, who has done research on authorship inflation and who has worked with Chinese scientists, suggests another explanation. Chinese researchers, he argues, are under greater pressure than researchers elsewhere to cite senior researchers within their own departments.)

Also, it’s not clear that China is outspending the developed world when it comes to science. As The Economist article shows, China’s R&D spending has grown rapidly since the 1990s:

But that’s partly because its economy grew rapidly; the percent of GDP China spends on R&D has also been increasing, but so far it’s just converging to the global norm of 2.5-3.5%:

(Note that the real research spending powerhouse here is South Korea, and the real laggard is France.)

And on top of that, the national R&D spending numbers that The Economist touts are actually mostly R&D spending by corporations, not by the Chinese government:

Much of this research is done in the corporate labs of state-owned enterprises, which took over much of the research function of Chinese government labs back in 1999. But the share of R&D output attributable to Chinese universities is fairly small, and has actually shrunk recently:

So China’s domination of global science, either in terms of citations or spending, isn’t really quite as dramatic as the Economist article makes it out to be.

But in the applied physical sciences — especially in materials science, chemistry, and engineering — China has definitely zoomed ahead of the West, even if you accept the “debiasing factors” from Qiu et al. (2024):

Health care is great and China is doing its people a disservice by skimping on health spending. But applied physical sciences are the key input to the export-oriented high-tech manufacturing industries — computer chips, EVs, and so on — that the US wants to foster.

And applied physical sciences are also crucial to winning wars — to building high volumes of highly accurate and destructive missiles and other weapons, and so on. The Economist certainly thinks military strength is a big factor behind China’s science push:

The Chinese Communist Party (CCP) has made agricultural research—which it sees as key to ensuring the country’s food security—a priority for scientists…“Engineering is the ultimate Chinese discipline in the modern period,” says Professor Marginson, “I think that’s partly about military technology and partly because that’s what you need to develop a nation.”

So it would seem like a good idea to beef up the US’ prowess in the physical sciences, not just because it improves the world and raises US GDP, but also in order to help keep the US and its allies strong.

Can US actually get more federal funding for science?

Discussions about science funding usually focus on the federal government. This is partly because government funding is simply the easiest policy lever to pull when you want something to change. It’s also because government funding is the weak link in the US research ecosystem. Over time, US private-sector R&D spending has risen steadily, while government funding for science has fallen relentlessly as a share of GDP:

Even within universities, the government’s role has shrunk over the last decade:

So it seems like we should do the obvious thing, and boost federal funding for science.

But there are a few reasons to be skeptical of increased federal funding as a silver-bullet solution. First of all, in an age of austerity like the one the US is probably headed into now, federal R&D funding is likely to suffer. The reason is that R&D funding doesn’t have much of a natural political constituency to go to bat for it on Capitol Hill — its benefits are diffuse and long-term.

You could see this play out in 2021-22. The CHIPS and Science Act started out as the Endless Frontier Act, a bold vision for increasing government research funding:

First introduced last year by Senators Chuck Schumer and Todd Young, the bill would have established a new Technology Directorate at the National Science Foundation (NSF) with a DARPA-like program structure equipped with flexible hiring and grant-making authorities.

With a $100 billion budget over five years, the Directorate would have been empowered to use grants, contracts, prizes, and cooperative agreements with industry, academia and research institutes to push the frontiers of US innovation in ten broad areas, ranging from cutting-edge technologies like Artificial Intelligence and quantum computing, to more mature but no less important sectors like robotics, manufacturing, biotechnology, advanced energy technology and material sciences.

That idea was based in part on the work of Jonathan Gruber and Simon Johnson, whose excellent book “Jump-Starting America“ was a clarion call to boost federal R&D spending to 1980s levels (this was before Johnson pivoted to calling for the government to slow down progress in artificial intelligence). Researchers from the Brookings Institution, as well as growth economists like Paul Romer, had also called for a big boost in government R&D.

It never happened. Congress significantly downsized the science spending in the bill, renaming it as the CHIPS and Science Act to reflect the shift in focus. There was still some science spending in there, but then Congress failed to appropriate most of the money for it, effectively gutting the remainder of the old Endless Frontier plan:

Two years in, Congress has fully funded subsidies for chipmakers. The big boost in science, however, is way off target…Congress has gnawed away at the law’s ambitions on fundamental research and development aimed at staying ahead of China and other rivals in competitive fields like artificial intelligence…

The latest example is the spending package lawmakers advanced over the past week: Biden’s signature enacts deep cuts to the National Science Foundation and stalls key offices in the Commerce and Energy departments that are supposed to deploy CHIPS money, turning a promised cash infusion of $200 billion over a decade into a humiliating haircut…

“These aren’t the numbers I’d like to see. I’m disappointed that we can’t provide funding to match what we authorized in CHIPS and Science,” House Science Chair Frank Lucas (R-Okla.) told Politico in an email. “Unfortunately, in our current fiscal environment we have to make difficult decisions and that’s reflected in the budgets for these agencies.”

It’s important to remember that when a bill passes Congress that “spends” an amount of money, that amount is only “authorized” — it’s actually just a notional target. The money isn’t really certain to be spent until it’s “appropriated” later. So basically Congress passed a bill promising to spend a bunch of money on science, and then just didn’t do it:

So you can see what an uphill battle this is.

Is direct federal funding all the US needs?

Then there’s the question of direct federal funding versus incentives for companies to fund research at universities. As you can see from the graph above, the percentage of federal funding in university science has been falling and is now just over 50%. But what we don’t really know is whether this is a good thing or a bad thing, on balance.

In fact, that’s an area of active debate in the economics world. For example, Fieldhouse and Mertens (2023) conclude that the economic returns to government-funded science are really large.

They do this by making a model in which government research creates “government R&D capital” which is then an input into the economy as a whole. Matching their model to the data, they find that government non-defense R&D is basically supercharges productivity growth:

[W]e find that a positive shock to appropriations for nondefense R&D robustly leads to a delayed and gradual increase in aggregate TFP that becomes highly statistically significant at long forecast horizons (8 to 15 years). For a shock that induces a one percent increase in government R&D capital, our baseline estimates show eventual increases in the level of TFP of about 0.2 percent. Positive shocks to nondefense R&D also induce increases in various indicators of innovative activity, such as patent grants, the number of doctoral recipients in STEM fields, the number of researchers engaged in R&D, or the number of technology publications.

This is a very big effect. The authors find a much smaller effect for defense R&D, but argue that this might be because the research results are kept secret for military purposes.

That’s a cool result, but there are lots of pieces of this analysis that might be wrong — the basic model relies on some theoretical assumptions, the time horizon is really long to be able to identify anything, etc. And there are some other papers that seem to contradict some of the conclusions. For example, Babina et al. (2020) find that federally funded university research is less likely to be commercialized:

[A] higher share of funding from federal sources reduces patenting activity…[A] 10% increase in the mean share of federal funding reduces the probability of any patenting by 0.4 percentage points, about half the mean.

The authors also find that more federal funding tends to keep researchers in academia, although it does also tend to increase their likelihood of starting startups.

Meanwhile, Arora et al. (2023) find similar results to Babina et al., and argue that federal research funding tends to crowd out private-sector research:

[W]e find that abstract public knowledge per se— publications in scientific journals—has little effect on the various components of corporate R&D. This means that corporate innovation is largely unresponsive to “pure” knowledge spillovers.

Second, public invention reduces corporate R&D. An increase in relevant university patents of one standard deviation reduces corporate patents by about 51%, corporate publications by approximately 33%, and the employment of AMWS scientists by about 8%. Further, we find that an increase in public invention reduces the firm’s profits, suggesting that, on balance, public inventions compete with corporate inventions more than they serve as inputs into corporate innovation…[F]irms on the technology frontier appear to respond less to public invention as compared to followers…

Taken together, our findings indicate that the public science that matters for corporate innovation—the science developed into patented inventions and embodied in the human capital of people—is both excludable and rivalrous. Thus, the expansion of public science may not lead to the sustained productivity growth that standard models of economic growth would predict.

Now your response to these findings may be something along the lines of: “Who cares about the private sector? Who cares about commercialization? For that matter, who cares about economic growth? The purpose of science is to discover the secrets of the Universe and increase human knowledge, not to make profit for some shareholder, you neoliberal shill!”

But regardless of your viewpoint on the value of discovering the secrets of the Universe, it’s probably the case that if research spending never makes its way into the creation of new commercializable products, it’s less likely to raise material living standards or to strengthen the national defense. So if we want to use science as a tool to enrich and strengthen the nation, we should be concerned about results showing that federal research spending is not the best way to do that.

So there are some conflicting results about whether federal R&D funding is the best way to fund science. I think a safe bet would be to go with a mix of direct federal funding and incentives for universities to work with corporations.

How else can the US support science?

The final question in my mind is whether there’s something else the US can be doing other than just spending more money on research. I think it’s instructive to realize that the meteoric rise of Chinese materials science, chemistry, and engineering has happened despite universities representing a slightly smaller share of China’s research spending.

That doesn’t mean more money for Chinese labs isn’t part of the story here — it is. China’s universities have reaped a share of the benefits of China’s rapid economic growth, even though corporate labs reaped an even larger share. But the fact that China is able to dominate the applied physical sciences without making academia more important in their system raises the question of whether the US might be able to accomplish something similar.

One possibility is that China focuses more on STEM education than the US. In fact, about 41% of Chinese college students major in STEM, compared to only 20% of Americans. But because more Americans go to college, the two countries have almost exactly equal STEM graduates as a percent of population:

If you multiply the US number by 4.26 (the population ratio between the countries), you come up with a number almost exactly the same as China’s.

Which raises the question: Maybe China is doing well in research just because it’s really, really big? Just as we couldn’t expect Germany to equal America’s scientific output in the long run, maybe it’s unrealistic to expect the US to keep pace scientifically with a country that has four times as many people. (Insert obvious pitch for mass high-skilled immigration here.)

In other words, maybe China is just becoming an average developed country along the dimensions of R&D spending and STEM education, but because it’s so huge, it looks like it’s doing better. I find this to be a common mistake Westerners make when looking at China — not realizing how many of its seeming outperformance is really just a function of size.

Even if size explains China’s overall R&D performance, though, there’s the question of whether the things China is researching are more important than the things the US and Europe are researching.

American and European research is much more biased toward life sciences, while Chinese research is much more biased toward applied physical sciences. So an important question is whether the US and its allies should shift spending out of biology and into materials science, engineering and chemistry.

In the US, this would probably involve diverting money from the NIH (part of the Department of Health and Human Services) to more physical science-focused agencies like the NSF and the Department of Energy. Currently, NIH utterly dominates nondefense research spending:

Changing that balance could be the key to competing with China in the applied physical sciences.

Anyway, I think the reports of China’s scientific dominance shouldn’t be causing policymakers in the West to panic. But it’s becoming pretty undeniable that China has now taken a commanding lead in applied physical science research and Western leaders need to ask themselves whether they can really afford to cede leadership in those fields.

This article was first published on Noah Smith’s Noahpinion Substack and is republished with kind permission. Read the original here and become a Noahopinion subscriber here.

Continue Reading

With over 640 million seats counted on June 5, 2024, spectators in India’s largest vote, which marked the end of the process, as well as what lessons can be learned for the rest of the world.

The efforts made extensive use of AI, including algorithmic imitations of candidates, celebrities and dying politicians. By some quotes, thousands of American citizens viewed deepfakes.

But, despite worries of common propaganda, for the most part the promotions, candidates and activists used AI positively in the vote. They used AI for normal political activities, including politicking, but generally to better interact with citizens.

Deepfakes without the fraud

Democratic parties in India spent an estimated US$ 50 million on authorized AI-generated material for precise connection with their constituencies this election period. And it was generally effective.

American political strategists have used Artificial to boost their communication because they have long understood the influence of character and emotion on their constituents. Young and future Artificial companies like The Indian Deepfaker, which began out serving the pleasure sector, immediately responded to this growing need for AI-generated promotion material.

In January, Muthuvel Karunanidhi, who was chief minister of the southwestern state of Tamil Nadu for two years, appeared via video at his party’s children aircraft event. He wore his signature yellow robe, white sweater, dark glasses and had his familiar attitude – nose slightly bent backward. Karunanidhi passed away in 2018, though. His party authorized the deepfake. A dead politician was introduced into the Indian election campaign by fake technology.

In February, the All- India Anna Dravidian Progressive Federation party’s official X account posted an audio clip of Jayaram Jayalalithaa, the iconic superstar of Tamil politics colloquially called” Amma” or” Mother”. Jayalalithaa passed away in 2016

Meanwhile, local representatives called voters to discuss local issues, but the voice on the other end of the phone was an AI impersonator. Shakti Singh Rathore, a member of the Bhhartiya Janta Party ( BJP), has been a frequenter of AI startups, sending personalized videos to specific voters about the government benefits they received and asking for their support over WhatsApp.

Multilingual boost

AI was present in the Indian elections in other ways than just deepfakes. Indian Prime Minister Narendra Modi addressed a crowd of a large number of people eager to celebrate the connections between the state of Tamil Nadu in the south of India and Varanasi in Uttar Pradesh’s northern state. As his Hindi speech was actually translated into Tamil, Modi proudly announced the launch of his “new AI technology” by requiring his audience to wear earphones.

The BJP used AI tools to make Modi’s personality accessible to voters in areas where Hindi is difficult to understand in a nation with 22 official languages and almost 780 unofficially recorded languages. Since 2022, Modi and his BJP have been using the AI- powered tool Bhashini, embedded in the NaMo mobile app, to translate Modi’s speeches with voiceovers in Telugu, Tamil, Malayalam, Kannada, Odia, Bengali, Marathi and Punjabi.

Some AI companies distributed their own viral versions of Modi’s well-known monthly radio show” Mann Ki Baat,” loosely translated” From the Heart,” which they voice-cloned to regional languages as part of their demos.

Adversarial uses

Indian political parties used AI to bolster their ongoing meme battles, and they doubled down on online trolling. The Indian National Congress uploaded a short clip to its 6 million Instagram followers early in the election season, using the song’s title from a brand-new Hindi music album called” Chor” ( thief ). Modi’s voice was copied by the video, which placed his digital likeness on the lead singer and reworked his lyrics to criticize his close ties to Indian business tycoons.

On its 7 million-follower Instagram account, the BJP launched a counter-retaliation video featuring a supercut of Modi campaigning on the streets, mixed with supporters ‘ clips, and played to unique music. It was a well-known singer Mahendra Kapoor‘s old patriotic Hindi song that was sung by him but was revived using artificial voice cloning. Kapoor passed away in 2008 and is now 84.

A common meme that alters footage of rapper Lil Yachty on stage, Modi himself tweeted an AI-created video of him dancing. Such inventiveness in the peak poll season is truly a delight, he said.

In some cases, generative AI tools were used to convey the violent rhetoric used in Modi’s campaign, which put Muslims at risk and stoked violence. However, the harm can be attributed to the offensive rhetoric itself, not necessarily the AI tools used to spread it.

The Indian experience

India is a first-time adopter, and its experiments with artificial intelligence serve as an example of what the rest of the world can anticipate from upcoming elections. Making it harder to tell the truth from fiction is made more accessible by technology’s ability to produce nonconsensual deepfakes of anyone, but its consensual applications are likely to increase that sameness.

The use of AI in participatory democracy that began with entertainment, political meme wars, emotional appeals to people, resurrected politicians, and persuasion through personalized phone calls to voters has opened a door for the role of AI in participatory democracy.

The BJP’s failure to win the anticipated parliamentary majority and India’s return to a highly competitive political system, among other things, highlights the potential for AI to play a positive role in deliberative democracy and representative governance.

Lessons for the world’s democracies

Any political party or candidate wants to have more specific conversations with their constituents, which is a priority in a democracy. In an unprecedented attempt to make their messages more accessible, especially to rural, low-income voters, by using AI for more individualized communication across linguistic and ethnically diverse constituencies.

Voters can communicate their demands and experiences directly with their representatives, and this could be done through AI and the development of participatory democracy, which would enable personalized communication as well as dialogue.

India can serve as an example of extending AI-aided party-to-people communications beyond politics with its recent proficiency. These platforms are already being used by the government to offer citizens in their native tongues government services.

This technology could provide a new era of representative governance, especially given the demands and experiences of rural residents.

Vandinika Shukla is a fellow at the Harvard Kennedy School’s Practicing Democracy Project, and Bruce Schneier is a visiting adjunct professor of public policy at the Harvard Kennedy School.

This article was republished from The Conversation under a Creative Commons license. Read the original article.

Continue Reading