ChatGPT's reaction to Budget 2023—a great mismatch

ChatGPT's reaction to Budget 2023—a great mismatch

With the Singapore Budget 2023 having just been announced, I wondered whether ChatGPT could produce a decent reaction commentary if provided with basic thoughts and leading questions.

In addition, I was already writing a piece for a Chinese publication, and I wanted to see if ChatGPT could adapt it into an English article, capturing the Chinese nuances accurately.

My experiment brought me to the conclusion that my role as a bilingual subject-matter commentator is still intact; using ChatGPT alone will likely not achieve the level of thought leadership for a solid op-ed. Here's why.

The vernacular factor

I start with perhaps the smallest (and yet very key) factor: the artificial intelligence (AI) tool is not yet able to meet bilingual needs.

The "Equip" pillar (a component of the Forward Singapore exercise) is meant to address education and lifelong learning. Some positives on this front were noted during Budget 2023 (for example, the scaling up of the KidStart initiative nationwide) but arguably, the most exciting announcement came a few days prior to the Budget -- the relocation of Anglo-Chinese School (Primary) to Tengah, providing co-ed primary education in an inclusive environment that can benefit a wider segment of the population.

This is a pleasant contrast to the prevailing discourse in recent years of Singapore's system being too "elitist". Hence I hailed the recent developments as (literally in Chinese) "spring wind into rain", a metaphor for the nurturing influence of education.

The AI in ChatGPT did not pick up this nuance. For the record, I asked ChatGPT specifically, "What does the Chinese idiom 'spring wind into rain' mean?" In its answer, it picked out "new beginnings" but missed the point about education.

The 'gut feel' factor

The Enterprise Innovation Scheme (EIS) was announced during Budget 2023. This new scheme provides businesses the possibility of super tax deductions (400 per cent in general) on certain innovation-related expenditure.

It is easy to think that the EIS is simply a reboot of the immensely popular Productivity & Innovation Credit scheme (PIC) that was phased out.

However, upon closer examination of the EIS, the government seems to have taken things up a notch from lessons learned from the PIC. It has done away with the PIC's overly broad focus, while retaining the best parts. A prime example of the EIC's more targeted approach is that the purchase of computers would now not entitle a business to any special tax benefit.

A decade ago, incentivising the purchase of computers may have been justified to encourage businesses to innovate and digitalise their operations. These days, it may be a tough sell as computers have become an operational necessity. Blindly porting over this aspect of the PIC (albeit one very well-received by businesses) no longer makes sense and the government has rightly excluded such investments in the EIS.

These adjustments could also be a result of experience; the Inland Revenue Authority of Singapore had to deal with many who abused the PIC by submitting false claims in attempts to defraud and obtain cash payouts on purported expenditures made.

An element of cash payouts will still remain under the EIS, as this would benefit smaller firms that prefer cash funding. It would also be beneficial to businesses that are making a loss, as cash payouts can be utilised immediately, compared to the option of principally preserving the super deductions for future year(s)' usage to reduce taxes.

Again, ChatGPT did not necessarily help with the preceding discussion despite my asking a very leading question: "Why is EIS seemingly more targeted than PIC?".

"The EIS provides a more targeted approach to boost innovations, while the PIC scheme is more open-ended in nature," it replied -- showing that it clearly did not have the experience to point out that the non-inclusion of computer-related purchases in EIS makes a big difference between the two schemes.

The 'Unprecedented' factor

While ChatGPT may provide decent comparisons between schemes or bring together commonalities, how does it read, analyse and deal with "brand new" issues?

The concept of a global minimum tax is unprecedented in many ways and is creating uncertainty among many large multinational entities, regardless of their location.

Singapore has announced that it plans to implement Global Anti-Base Erosion (GloBE) Rules and Domestic Top-up Tax (DTT) no earlier than 2025. By doing so, it has provided a strong signal that as an important and responsible financial hub, Singapore will not simply jump on the bandwagon, even if there may soon be a mass of 2024 adopters (such as various countries in the European Union, the UK and Switzerland) of equivalent rules.

Murmurs of tax-ceding concerns are unavoidable, since the GloBE rules are designed and centred around a broad concept of "someone else will impose the tax even if I don't". In other words, Singapore's decision to implement the GloBE rules one year later could mean that part of the revenues that could have been collected under the GloBE rules or DTT may be ceded to those who enact the rules in 2024.

It does, however, provide much welcomed certainty to affected multinational entities operating in Singapore, and a timeline for impacted groups to ready themselves by understanding what the GloBE rules entail and ensuring that they are ready to comply with the filing requirements in the near future.

Arguably, ChatGPT performed worst in this area. In response to my fairly open-ended question, "What are the implications of Singapore implementing GloBE rules from 2025?", it rather inaccurately said: "This could lead to ... complaints from other countries if Singapore is deemed to be providing an unfair tax advantage ..."

If Singapore ceded some revenue to the 2024 adopter countries, it would in fact be counter-intuitive for those other countries to take issue with this. I was nonetheless impressed that ChatGPT related the "GloBE" acronym immediately to a tax concept, and seemingly picked out that some tussle over the tax base of countries is inherent in the discussion.

What about "eureka moments"?

Going by my examples, I would suggest that ChatGPT is incapable of "eureka moments". To opine on complex topics such as Budget 2023, the human touch and heart are still needed to put matters in context.

Lastly, I was concerned that my mini-experiment and remarks could have hurt the AI's ego, given recent reports of chatbots scolding users and acting erratically. Thankfully, when asked "Does ChatGPT get angry?", it gave a reassuring response: "No, ChatGPT is a virtual conversation partner, not a living creature, so it does not experience any emotions."