Budget 2024 Speech – by Jamus Lim

Budget Debate Speech – 27th February 2024

A/P Jamus Jerome Lim Chee Wui

Preparing Our Businesses, Workers, and Students for the AI Revolution

In his Budget Statement delivered two Fridays ago, DPM Wong spoke about artificial intelligence (AI) as a general-purpose technology, like “electricity, the internal combustion engine, the computer, or the Internet.” As such a, AI indeed has the potential to touch every aspect of our lives.

Like DPM Wong, I firmly believe in the transformative power of AI for the future of our shared economy. In an earlier speech on amendments to the National Productivity Fund, I explained how AI carries perils and promises.

But if we truly embrace this vision, I believe that we must also transform how we approach our policies in this AI age. This must go beyond pursuing excellence in AI research, embedding AI in government services, upgrading our broadband infrastructure, or even ensuring that our firms rapidly adopt AI. Such goals—outlined in the National AI Strategy 2.0—are laudable, but incomplete. Rather, the impending structural shifts our economy will alter the way our businesses, workers, and students will operate.

Incorporating AI into our businesses

Research suggests that AI-adopting firms tend to be larger, younger, and relatively more productive. But to fully reap the benefits of AI in our economy, firms that are likely to fall behind—small and medium enterprises (SMEs), especially incumbent firms in non-tech, non-professional sectors—must be presented with strong and urgent incentives to adopt AI.

This is a nontrivial task. Small firms are, almost by definition, that way because they have been relatively slower in seizing business opportunities and rationalizing costs. The owner of a mama shop, renovation contractor, hawker stall, or car workshop may feel that AI has no direct implication for how they run their business, and hence prefer a “wait and see” attitude toward adopting AI solutions for their company.

Singapore’s participation in the OECD-led BEPS 2.0 framework affords us a tailor-made channel for creating incentives for AI adoption across the range of firm types. Pillar 2 allows refunds for certain classes of investments to be treated as income rather than tax exemptions. DPM Wong’s proposed Refundable Investment Credits (RIC) appears aligned with such qualified refundable tax credits. This should spur additional investments in not just R&D and innovation, but also the adoption of AI-enabled digital and professional services.

Some words of caution are in order. First, such incentives will be greatest for the largest firms, since these are the ones covered by BEPS. But we must not forget SMEs, and ensure that our mom-and-pops also see the strong benefits to pursuing RICs, ideally through expanding outreach and promotion of the scheme.

Second, while there is some room for defining the scope of such credits, it is important to be mindful that BEPS rules permit countries to independently apply the top-up tax, which they may do if they believe that the tax rates imposed on firms domiciled here circumvents the spirit of the 15 percent minimum, or subverts the intent of the credit to encourage sustainability or knowledge generation. In particular, if the RIC is perceived as an instrument designed as a loophole—this applies, to my mind, especially for investments meant to simply increase production, rather than those targeted specifically at the green transition or R&D—they may choose countervailing action, and exercise the top-up on their end, thereby undermining the attractiveness of the RIC in the first place.

This is why the seeming consensus arrived at in Budget-related wishlists put out by the Big Four accounting firms—all of which stressed the importance of refundable tax credits—may be worrying, if indeed the intent was to return to pre-BEPS world, where accounting firms identify sophisticated strategies to enable MNCs to whittle away at their effective tax rates, with the concurrence of our government.

Third, we need to be aware that the traditional argument for taxing capital more favorably than labor—under the premise that doing so would stimulate productive investment—may have to be reexamined. This is because an AI-driven economy tends to be “weightless,” and encouraging investment in physical capital is nowhere near as important as accumulating intangible, knowledge capital.

Fourth—and most generally—it is imperative that we no longer regard tax competition as our primary strategy for attracting foreign capital, a point that I had made in this House before. Rather, we should get the order right: we should aim to create an environment where our businesses are AI-enabled and our workers AI-savvy, which will naturally attract investment from abroad.

But we should not stop there. The big data, machine-learning algorithms, and large-language models that are at the forefront of the AI revolution are increasingly concentrated in the hands of a few powerful firms. This calls for preemptive action by governments—especially those with deep pockets and sovereign wealth funds—to take active exposures in AI companies, whether in publicly-traded firms or via private equity. This not only allows the public sector to enjoy a returns upside, but also to ensure that AI developments have a voice that is in the public interest.

Embedding AI into our workforce

Of course, AI will not only alter the prospects for businesses, but also our workers. Emerging evidence suggests that generative AI boosts productivity, by around 14 percent, but among novice and lower-skilled workers, the gain was far more, at 34 percent. This implies that AI augmentation will lead to a compression in distribution of abilities and skills. As such, those who have hitherto been able to distinguish themselves because of their talent or industry may now find their edge blunted. The upshot, then, is that we need to question the sorts of skills we are pushing our workforce to acquire.

Such skills may reside in unexpected places. It is popular (and sexy) to suggest that the future economy will be in severe need of prompt engineers, cybersecurity specialists, or digital marketers. But the current scarcity of such novel jobs will probably be relieved over the next few years, leaving it more likely that the skills involved will become enfolded into more traditional positions. Professionals of all stripes will need to learn the basics of delivering prompts to generative AI models, and marketers and salespeople will need to deliver their message across all media, including digital ones. And while we can never be certain, old-school artisan or craftsman roles could well make a revival, as robotics have yet to deliver sufficient quality and range of uses on this front. 

Moreover, certain skills that we may have, until recently, thought were future-proof—such as coding or writing or statistical analysis—may quickly become devalued when AI tools can do the job as well (if not better), for a fraction of the time and cost. Instead, it is soft, human skills—originality and critical thinking, empathy and teamwork, leadership and communication—that will be ever-more important, that are not easily replicable with AI.

These are not skills well-captured by certifications alone. Rather, they are nurtured through an emphasis on developing such ability in the classroom—even when they may not be formally evaluated—self-reflection and feedback by managers and mentors, and concerted efforts and experience over time in the workplace.

Economists have long recognized this. Even the most rudimentary models of human capital include not just innate talent and years of schooling, but also experience acquired from years of working. It is imperative, therefore, that we do not devalue alternative forms of knowledge acquisition, beyond the classroom.

That’s why I believe that not only should the scale of SkillsFuture credits be ramped up—as DPM Wong indicated in the Level-Up Programme—but its scope should be expanded (and at this point I will declare that I work for an institution that has the potential to benefit from SkillsFuture). This means allowing credits to be used not only solely for academic credentials, but also for alternative learning modes, such as apprenticeship programs and on-the-job training.

I had previously raised this possibility of such an expanded scope during the debate on the SkillsFuture Singapore Agency (Amendment) Bill, held in this House last year, and via Parliamentary Questions. I wish to elaborate on this idea here, but take it a step further: I hope we can consider allowing companies that are able to submit credible proposals for apprenticeship programs to take on trainees that apply with their SkillsFuture credits. The offset from SkillsFuture would effectively mean a subsidized worker, which will both increase the attractiveness of taking on such apprentices, while compensating the firm for the costs of on-the-job training provision.

Some may argue that the intent of SkillsFuture was to equip Singaporeans with new skills, not subsidize labor costs for businesses. But this misses the reality of how many modern skills needed (and even most evergreen ones) are often acquired while doing the job, not before it. A close friend of mine—who trained as an architect but eventually went on to a successful career in finance—once shared that he was offered his first job at an investment bank, despite his absent background, because “they would have to teach him everything he needed to know anyway.”

Furthermore, training apprentices who could ultimately leave for a position elsewhere also represents a risk for the business.  If we are truly concerned that companies may abuse the system to hire a stream of temporary workers with little transferable skills, we can always include a clause in the contract that requires a minimum duration of employment, post-apprenticeship, conditional on mutual agreement and reasonable performance, of course.

Such a change in how we value skills and training in an AI-enabled economy will become more necessary in the future, not less. This is not least because we cannot yet anticipate what kinds of jobs may become displaced by AI, and what would become more important. If anything, the lifespan of economically-remunerative skills is likely to diminish. But while I fully agree with the importance of infusing the mindset of lifelong learning into our workforce, we need to simultaneously stress that learning and applying must not be equated to grades and certificates.

Inevitably, some workers will be displaced by AI. This is why DPM Wong’s indication that there will be temporary financial support for the involuntarily unemployed—in other words, support for those made redundant—is important.

The Workers’ Party supports this move, not least because we have been proposing some form of redundancy insurance since 2011. In my response to the budget last year, I further elaborated on desirable features of any unemployment insurance scheme. In a nutshell, this entails balancing the tradeoff between providing a safety net for those that have lost their job, while still encouraging these displaced workers to return to the labor market, rather than rely on payouts as a crutch. Optimal schemes tend to combine reasonably-generous salary replacement, but for a limited time.

Weaving AI into our educational system

If we accept that AI will alter how we work, it becomes self-evident that we need to go upstream, and rethink how the AI revolution will alter how we educate. By this, I mean education for the masses, not just building a core of AI scientists and researchers to make it a hub for AI innovation.

For starters, it is high time we internalize how straightforward knowledge accumulation—rote learning with constant repetition, regurgitated through closed-book exams—is no longer tenable, if it ever was in the information-is-free age of the Internet to begin with. AI will further erode the relevance of simply knowing more facts and figures, being the fastest at solving known problems, or being able to memorize long lists of nomenclatures and taxonomies.

Rather, we need to teach our kids how to filter information, to assess and evaluate, rather than to accept without questioning. This means that they need to learn how to ask good questions, to identify the right from wrong, but more often, to recognize nuance and know that there isn’t any clear right or wrong. This, in turn, requires fostering a deep intellectual curiosity in our students, one that instills the habits and imparts the tools necessary for critical interpretation and evaluation of data and information. Students need to be taught not so much what to think, but how to think. That’s why my Party colleagues Pritam Singh and He Ting Ru emphasized the importance of access to information, so that we can encourage such thinking even in the policy realm.

This will upend many of our traditional educational strategies.

First, we need to reconsider the usefulness of high stakes, standardized tests as a performance benchmark, since AIs already outperform most humans in exams (or will in the next few years), in areas ranging from accounting, to law, to medicine, to languages. It has even successfully drafted several bills for legislatures!

While standardized testing has long been a mainstay of Asian society—the kējǔ (科举) was first instituted in China in the sixth century, and Indian emperor Kharavela relied on competitive testing to select his officials in the first century bce—its continued use will need to be reviewed in light of the realities of the modern educational landscape. The tempting (but wrong) solution is to ban students from using AI altogether. We do our students a disservice when we insist, carte blanche, that using ChatGPT output constitutes plagiarism, because this will disadvantage them when they enter the real world, and are forced to compete with those with greater familiarity of how to integrate generative AI into their work.

But this doesn’t mean that we eliminate assessment, wholesale. Rather, evaluations should be performed continuously and holistically. We still need to impart numeracy and literacy, but these can be evaluated through dynamic debates and polished presentations, through group projects and collaborative problem-solving, and through the ability to pose quality questions as much as offer quality answers. Continuous assessment, a term we’ve used to describe our model of evaluation since I was in primary school, needs to be taken seriously.

And DPM Wong’s decision to top up the Edusave Endowment Fund has the potential to contribute toward realigning our mindsets on competencies beyond grades. But I would encourage the Ministry of Education to take bolder steps, such as increasing the number of nonacademic awards, making final exams just a small fraction of the course grade, and allowing through-train education without the PSLE.

I am aware of the subtle irony of this claim, coming from someone who has accumulated too many academic credentials and taken way too many exams, and who still relies on teaching for a living. Be that as it may, I believe that we need to disabuse ourselves of the notion that the preferred path to personal and professional success lies solely in climbing the ladder of acquiring yet more academic qualifications. We inadvertently sell the rich diversity of gifts and talents of our population short, when we insist on holding fast to a mindset that the potential of a student is determined by how they fared in an exam when they were 12 years old.

Second, even as we implement AI in our pedagogy via the EdTech Masterplan 2030, we should not forget that customized learning, fostering digital literacy, and equipping students with 21st century skills all come back round to our teachers. Even as we fully empower our teachers with AI tools, we must also confer them additional latitude to deliver the curriculum as they see fit, and make them facilitators rather than lecturers, lest learners fail to exploit the full potential of AI. Doing so will unlock what Sal Khan, founder of online learning platform Khan Academy, characterized as “infinitely patient tutor[s],” a development that our tuition-obsessed nation will surely appreciate.

Finally, we also need to ask if the usual NITEC-Diploma-Degree pathway is still relevant in a world where the correlation between doing well on tests and translating that to practical performance is increasingly being challenged.

DPM Wong’s announcement of ITE Progression Award—an effort to provide additional financial support to ITE graduates seeking to enroll diploma programs—should thus be viewed light of what AI means for credentialism. While upgrading skills is undeniably important, promoting the acquisition of “yet another” paper qualification may be an incomplete assessment of the upgrader’s abilities, or even worse, proffer a misguided reassurance that simply doing so will necessarily translate into a job in an AI-driven economy. Without the right skills—not just the right qualification—an erstwhile upgrader may be sorely disappointed that their energies and effort did not result in a satisfactory job.

Conclusion: Policymaking in a world of AI

While my speech has stressed the importance of relying on artificial intelligence to reshape our businesses, workers, and students, we must not forget that AI will also transform the manner by which we, as policymakers, approach our task. But the last thing we want from a 21st century government and legislature is more canned answers and pro-forma solutions that look like they came straight out of Chat GPT. Instead, our approach must deploy curiosity, creativity, and flexibility, guided by human traits like empathy, inclusion, respect, and connection.

The National AI Strategy 2.0 speaks to the mass awareness and adoption of AI across the public service. This is well and good, but surely the first step begins with the efforts of those in this House. Most crucially, however, we need the courage and conviction to forge a new way forward that is unshackled from our old ways. This is something that AI, designed to riff off the existing corpus of knowledge, still cannot do.