AI and the Filipino

Bad news: the Philippines will be slow in adopting artificial intelligence (AI) and its improvements – artificial general intelligence (AGI) and artificial general superintelligence (ASI).

Good news: the Philippines will be slow in adopting AI and its improvements.

AGI is the intelligence of a single human. ASI is superhuman abilities across virtually all tasks and domains of human interest and activity.

Why is slow adaption of AI both bad news and good news for us?

Bad because Filipinos will be left behind by the world as the world goes super digital at super speed.

Good because not being so intelligent could actually be more of a benefit than a curse. As they say, innocence is bliss.

Why is innocence bliss? Because there is a limit to what AI can do. That limit is the available data – text, voice, video, etc – mankind has produced in the last 2,000 years.

In theory, AI’s computing power should grow 10,000 times it is today. That’s the definition of “progress” – what Google’s DeepMind think tank refers to as advances in AI capabilities, plus scientific and technological progress, and even societal progress more broadly.

Under Moore’s law, in the past 60 years, computing power per dollar was measured at a rate of 1.5 times per year. It was called exponential growth, doubling computing power every two years. There is a new metric called hyperbolic growth – computers doubling in power next year, then doubling again in the next six months, then again in the next three months, then again in the next 1.5 months, until you reach infinite technological growth called technological singularity.

Don’t worry, Filipinos, we won’t reach that singularity, physically. In the first place, a nation with 7,600 islands, 100 dialects or tribes and a multi-party political system that thrives on stealing flood control and other forms of tax money cannot reach singularity. Just look at the quality of our senators, individually and as a group - stupid, corrupt, plunderers, obstructionists, dynastic, traitors to country and people.  They have no singularity – except when it comes to money, people’s money.

This brings me to three Cs that hinder AI or computing power, Philippine style – cost, corruption, capability.

Cost. Computing power or AI is very expensive, for two reasons: 1) a global shortage of microchips and 2) the global high cost of energy.

Google says AI, particularly large language models (LLMs), require enormous computational resources. Training these models involves thousands of graphics processing units (GPUs) running continuously for months, leading to high electricity consumption.

In 2023, data centers consumed 4.4 percent of US electricity – a number that could triple by 2028. AI’s rapid expansion also drives higher e-waste, raising urgent sustainability concerns, says Penn State Institute of Energy and the Environment.

Data centers consume 70 percent of memory chips, particularly DRAM and NAND. Chip production is inadequate and delayed by six months to a year. The same chips are needed to run iPhones and laptops. This September when Apple launches its latest iPhone, it will raise prices by $200 for the most basic model.

Corruption. The Philippines ranks 120th out of 182 countries in Transparency International’s Corruption Perception Index. Nine of every 10 Filipinos believe corruption in government is widespread.

I once asked to be connected to a reliable Meralco meter. Meralco is a private company. But the local barangay asked me to produce 17 permits. It takes 47 permits and two years to get a permit to build a house. It takes 100 permits and 10 years to get a permit to install a windmill. Our own Senate is a house of corruption; at least half of the 24 senators – a quorum – are guilty of the most severe form of corruption. Our Congress itself is the Philippines’ biggest criminal syndicate.

And why is red tape worse today despite claims of rapid digitalization of government procedures? Answer: corruption. Digital hates corruption. What do government humans do? They corrupt the system.

Capability. Per international tests, our 15-year-olds cannot read, cannot write and cannot count beyond 20 – the most basic form of arithmetic. Even if they can read, they cannot understand what they read. These are the people AI, no matter how advanced the model, cannot penetrate. If you have no knowledge and no competence, you cannot outsource that to AI. Therefore, AI cannot replace you. You are in good company. There are about 15 million 15-year-old Filipinos.

In computing, there is such a word as FLOPS, a measure of computing power in fields requiring floating point calculations. In our educational system, our 15-year-olds are flops - complete, unsuccessful failures. A failure is something AI cannot handle; because AI is supposed to be intelligent.

Meanwhile, asks Google’s DeepMind: “If continued AI progress relies mostly on scaling compute, data and inputs like energy and computer hardware, then a main question is whether the economic cost of continued scaling (over orders of magnitude) can be sustained and for how long?”

To answer that, you need to hurdle six bottlenecks, says DeepMind: 1) data wall; 2) too fast a growth in economic and natural resource demand; 3) insufficient neural paradigm; 4) research is getting harder; 5) abstraction barrier and 6) deliberate slowdown.

Data wall means running out of sufficient (or rather, sufficiently growing) amounts of high-quality data for pre-training, post-training, fine-tuning and test-time adaptation.

A neural network (specifically, an artificial neural network) is a machine learning model inspired by the human brain. It uses interconnected nodes, or “neurons,” organized in layers to process information, recognize patterns and make predictions. Neural networks power modern AI technologies like image recognition, natural language processing and generative AI, says Google.

On deliberate slowdown of AI, DeepMind explains: “Rogue-actor use, accidents or severe risks, military or political (ab)use, or sociocultural harm and societal backlash, might lead to deliberate slowdown or regulatory capping of AI capability improvements.”

*      *      *

Email: [email protected]