#84 Big Innovation, Big Impact
Why Major Firms Must Step Their R&D Efforts Now; (De)Generative AI - Is ChatGPT Killing the Essay?
Today, Arindam Goswami writes an incisive essay on why India’s big firms must ramp up their in-house R&D efforts to keep pace with their growing size.
Rohan Pai follows with an excerpt from an upcoming Takshashila research document that discusses the impact ChatGPT has had on the “art” of essay writing.
Also,
We are hiring! If you are passionate about working on emerging areas of contention at the intersection of technology and international relations, check out the Staff Research Analyst position with Takshashila’s High-Tech Geopolitics programme here.
Cyberpolitik 1: Why Major Firms Must Step Their R&D Efforts Now
— Arindam Goswami
An effective innovation system has innovation happening majorly at firms. Profit-maximising, export-oriented firms benefit from research and development (R&D) done in-house and from research happening elsewhere, but without their own efforts, the collateral benefits don’t really materialise. The reverse is also true: firm competitiveness and growth also impact innovation. Around the world, an average of 1.5% of the GDP is spent on in-house R&D by industry (in India, this number is just 0.3%).
As per this paper written quite a while ago, there is an elasticity of R&D with respect to sales that is close to unity. In simpler words, there is a 1-to-1 proportionality of R&D to sales. But as argued here, does this mean that firm sizes don’t matter to aggregate level of innovation – does it matter if we have 10 small firms or just 1 big firm doing R&D? Interesting question - we’ll come back to this in a while.
The Indian Scene
Before that, let’s dig a little deeper into the Indian scene. As Naushad Forbes argued in this column of his, assuming that innovation and firm competitiveness are tied to each other, did India lose its way by trying to build technological heft through public R&D investment immediately after independence, instead of going via the private sector route as did Japan, South Korea, Taiwan, Singapore and China (later on)?
Their path to building an innovation system had a particular sequence. The first step is for firms to enter export markets as original equipment manufacturers (OEMs) in usually labour-intensive and low-technology industries, then move up the value chain to technology-intensive industries with accompanying in-house R&D for sustaining competitiveness, and lastly, move to higher-technology sectors. This, of course, requires an ecosystem of research talent, which these countries created by funding innovation through publicly funded research in the higher education system, with its knowledge spillovers and positive externalities.
Contrast this with India, where, genuinely speaking, innovation was thrust upon Indian industry only after the liberalisation process started in 1991. But the necessary step of increased in-house innovation hasn’t really happened. Even our big firms are doing woefully less innovation. As the data in this article by Naushad Forbes shows,
“The top 2,500 firms investing in R&D worldwide account for around 90 per cent of all industrial R&D, … top 10 sectors account for close to 80 per cent. India has no firms in the top 2,500 in five of these 10 sectors, and just one firm each in two more. Within some sectors, we are much less R&D-intensive: Our software firms, large by world standards of profitability, are small in R&D, investing around 1 per cent of turnover on average, against a world average of 12 per cent (and a Chinese average of 10 per cent). And we are simply missing any giant investors in R&D. The world’s 26th-largest investor in R&D, Bosch, invests more than all Indian firms combined.
… Our 10 most profitable non-financial firms… invested under $1 billion, or about 2 per cent of profit, in R&D. Contrast China:… 29 per cent of profit.”
Incremental Innovations, Cost Spreading, Innovator’s dilemma
So, what can explain this? Let’s return to the question posed at the start—how does firm size relate to innovation? Evidence shows that while R&D programs grow more significantly with firm size, patents per dollar or per employee are fewer. Even if we use different measures to evaluate innovation output, we would reach the same conclusion of a negative relationship between firm size and innovation.
Evidence, however, finds that larger firms do more incremental innovations aimed at improving existing products than venturing into new ones for markets where they don’t have any existing revenues. Also, those innovations are less impactful. Now, this is paradoxical. If the productivity of their R&D activities is lower, why do large firms keep investing the same effort as small firms into R&D?
One possible explanation is that this is perhaps illusory—large firms apparently invest more in process innovations that focus on delivering a service or manufacturing a product better than in building new product lines, as this paper finds. Process innovations are difficult to observe and survey and sometimes are not patented as firm sizes grow.
Another explanation is attributed to the “cost spreading” advantage of larger firms in conducting R&D. Larger firms have many product lines and products, over which it makes sense to spread the cost of their R&D activities. They’d like to focus on improving their existing products to service their known customer base rather than look for newer consumer segments. Also, larger firms are more likely to report their R&D spending as the cost of doing so is easier for them to bear than small firms, and the incremental benefits of doing so are spread out over multiple product lines and products, giving better returns on investment. This push towards hard-to-observe process innovation could lower R&D productivity in larger firms.
Now let’s look at the “replacement effect”, an old argument in economics dating back to this paper. As per this argument, incumbent firms (generally large firms) have a smaller incentive to work towards a new product because that would disrupt their old products and thus not give enough returns from older product lines. This, therefore, makes them invest less in product innovation than process innovation.
This also ties in nicely with another concept called the innovator’s dilemma. Large firms listen to their existing customers and provide incrementally better, highest-value products. This is also the lowest risk for them because they don’t want to do anything to unnecessarily disrupt their existing customer base, thereby impacting sales and turnover, which then impacts shareholders. However, new firms that don’t have an established consumer base are more likely to experiment with both their products and consumer bases. They can iterate faster over different versions of the technology underpinning their products, even going so far as to release low-quality technology to gauge the sentiments in a new niche customer segment. If the feedback is encouraging, they can iterate faster than larger firms to improve the technology. If the feedback is not encouraging, they can pivot faster, as they are nimbler and don’t have to continuously look over their shoulders at sales numbers and shareholders. In time, the smaller firms capture the new customer segment, leaving the incumbent, larger firms caught unawares. It is too late for them to play catch-up with the startups. This then perpetuates the vicious cycle where larger firms keep focussing more on incremental process innovations than product innovations.
One of the possible illustrations of this dynamic is how Google seems to have underinvested initially in the type of AI technology that powers OpenAI’s ChatGPT as that would have directly impacted their ad revenues from their search engine.
In conclusion, what does this mean for R&D efforts at large Indian firms?
To tie all of the above into one coherent argument for the Indian scene, it could be argued that while the post-1991 liberalisation set us on a path where our industries have grown because the ecosystem of research talent, research institutes and higher education sector wasn’t (and isn’t) as developed as is required to foster innovation, their inhouse R&D was severely lacking. And now that they have grown in size, their R&D investment, which is too low even now, is not geared towards product innovations, due to the multiple factors we saw above (incremental process innovations, cost spreading, innovator’s dilemma).
The funding aspect is something which, anyway, has to be solved and improved upon. That is a fundamental and necessary, but not sufficient, requirement. Without reorienting away from the structural problems associated with larger firms when it comes to innovation, it won’t be possible to come up with the big innovations needed to catapult the Indian innovation scene forward. Some strategies can help, like building autonomous, startup-like organisations within a larger firm to drive disruptive innovation. There could be tie-ups between incumbent firms, research institutes, and the higher education sector to drive attention towards product innovations.
As Naushad Forbes argues, drawing on the chain-linked model of innovation advanced by Steve Kline,
“Knowledge — both technological and scientific — plays a key role not as a trigger for innovation, but as a repository one draws on during the innovation process to help solve problems. And research is what you do when the existing stock of knowledge is not enough to solve the problem… For the great bulk of the world’s R&D effort that takes place in hundreds of thousands of firms worldwide, Science (capital S) as an organised research effort doesn’t matter; science (small s) — excellently educated students — does.”
To do that, Indian firms, like others, need a massive reorientation.
India is set to host the Quad Leaders' Summit in 2024. Subscribe to Takshashila's Quad Bulletin, a fortnightly newsletter that tracks the Quad's activities through the Indo-Pacific.
Your weekly dose of All Things China, with an upcoming particular focus on Chinese discourses on defence, foreign policy, tech, and India, awaits you in the Eye on China newsletter!
The Takshashila Geospatial Bulletin is a monthly dispatch of Geospatial insights for India’s strategic affairs. Subscribe now!
Cyberpolitik 2: (De)Generative AI - Is ChatGPT Killing the Essay?
— Rohan Pai
**This is a sneak peek into an upcoming research document. It will be available to read on Takshashila’s research page soon**
ChatGPT was released to the public in November 2022, and the initial fears of students using it as a tool for academic dishonesty are still yet to wind down. A paper published by Hindawi found that high school teachers were able to correctly identify a ChatGPT-generated essay only 70% of the time. While this statistic may not appear especially dire at first glance, it can have serious ramifications for innocent students falsely accused of using the chatbot to write an assignment.
Against this backdrop, educational institutions have been forced to think outside the box and devise newer methods of evaluation that circumvent the capabilities of ChatGPT.
The first method of circumvention is the personalised prompt, that is to say, a prompt which requires students to draw from their personal experiences while writing an essay. The chatbot, at the end of the day, is merely a large-language model or LLM that extracts information from a fixed set of training data from webpages such as Wikipedia and Reddit, as well as e-books. While the conversational tone of its responses certainly mimics those of a human being, ChatGPT is still a far cry from the dystopia of artificial intelligence operating on emotion.
The second method of circumvention involves framing essay prompts about current affairs, whether at a global, national or local scale. Students attempting to use ChatGPT for an assignment on a hot-button issue are sure to be disappointed, because the free version of the chatbot (based on GPT 3.5) currently has a knowledge cutoff around the time of January 2022, after which training was halted. This constraint in ChatGPT’s most widely available version’s functionality perhaps explains why it hasn’t yet threatened to take Google’s spot as the most popular search engine.
This method of circumvention, however, is a ticking time bomb with an expiration date. In September 2023, former OpenAI CEO Sam Altman announced proudly that ChatGPT’s premium version could now browse the internet and generate responses using up-to-date information. It must be noted that the new feature is currently available only to users with a monthly subscription to the premium version ChatGPT Plus (based on GPT 4). OpenAI has confirmed, though, that it’s only a matter of time before the feature is democratised and made freely accessible to the public.
Circumvention certainly puts up a good fight against the academic dishonesty triggered by ChatGPT, but it also fails to offer complete assurance and has led to many institutions blocking user access to the OpenAI website on school-sponsored devices and networks. A common method of detection used by many educators is simply asking ChatGPT itself whether a piece of writing was written by a human being or auto-generated by AI. As it turns out, however, relying on AI for protection against the misuse of AI may not always be the best path forward. Using this method caused a professor at Texas A&M University to erroneously fail more than half the students in his class.
False positives have also been generated by universally-renowned plagiarism checkers. Institutions such as Vanderbilt University and Northwestern University have publicly issued statements on their decisions to halt the usage of Turnitin’s AI detection tool. Protection of students from a false accusation has proven to be equally, if not more, concerning for universities than the preservation of academic integrity. Tiptoeing between these extremes and searching for a happy medium is the current dilemma that continues to stump educational institutions.
The plot thickens after scholars from Stanford discovered that ChatGPT detectors possess a considerable bias against writing produced by those for whom English is a second language. According to their paper, roughly 61.22% of essays written by non-native English speakers for TOEFL (Test of English as a Foreign Language) were flagged as AI-generated, while a negligible percentage of the same was flagged for English-speaking students born in the US. This imbalance can largely be attributed to ChatGPT’s general and relatively less sophisticated writing style that could, at times, mimic that of a non-native English speaker who may lack a strong vocabulary. Not only do ChatGPT detectors run the risk of falsely accusing students of academic dishonesty, but they disproportionately target those already at a disadvantage.
This has led a number of educational institutions to re-evaluate their strategies of prevention. The Russell Group, for example, released a comprehensive set of principles in July 2023 on the use of generative AI tools by students. Far from placing bans, the Russell Group has tasked university staff with the responsibility of ensuring students are AI-literate, and adapting assessment formats with the use of generative AI tools.
Such radical acceptance of cutting-edge technology may be a wise choice because, as past experience shows, attempts to ban a product or service outright have not always sown the seeds of deterrence. More than simply being ineffective, however, bans on ChatGPT may inadvertently harm the career prospects of students growing up in the era of AI. Proficiency in ChatGPT might, after all, constitute a hiring criterion for entry-level positions in the near future.
The widespread accusations made against ChatGPT for destroying the very foundation of education, whether warranted or not, ultimately commit the sin of equating intelligence with one’s ability to write. Essays have always been an incomplete measure of a student’s abilities, and the ChatGPT era serves as a wake-up call for educational institutions to put more time and effort into the process of evaluation. The invention of calculators did not render the field of mathematics obsolete, and the arrival of ChatGPT is just history repeating itself.
Admissions are open for Takshashila’s Postgraduate Programme in Public Policy. This comprehensive, multidisciplinary, and entirely online 48-week program is tailored for those who want to delve deep into the theoretical and practical aspects of public policy.
What We're Reading (or Listening to)
[Takshashila Policy Brief] A Case for Quad Cooperation in Biofuels, by Bharat Sharma
[Opinion] What SSF’s disbandment tells us about Chinese policymaking under Xi Jinping, by Manoj Kewalramani
[Newsletter] How to Build a $20 Billion Semiconductor Fab [Construction Physics]