Including the way we make our forecasts. And the way we design our delivery routes. And the way we see the logistics provider. And perhaps what our children learn in school, too.
Written by Henrik Batallones
Let’s be honest: it feels as if artificial intelligence has come in and swept us away.
It is a truly revolutionary technology, once the stuff of science fiction movies. It is perhaps the closest we have gotten to the Jetsons’ idea of the future, or perhaps that of Tony Stark’s: robots we can communicate with naturally, who will understand the instructions we give them and perform them intuitively. But what was presented then as a doe-eyed vision of the future turns out to come with costs and trade-offs that are hard to ignore.
It wasn’t always like this. Studies into making artificial intelligence—formally defined as the capability of computer systems to perform tasks typically associated with human intelligence, like learning, reasoning, problem-solving and decision-making—have been underway since the 1950s, but was initially hampered by limited capabilities of the equipment at the time, and the initial goal of replicating the human ability to set plans, goals and beliefs. By the 1990s, a narrowing of focus—towards allowing computers to solve specific problems—reignited interest in AI and brought with it renewed funding for research, although some were still keen to pursue the original vision of a fully independent machine.
As the years passed, improvements in hardware—particularly increases in processing power while shrinking the physical chips—led to software tackling more complex operations, although these are still bound by rules set in the programming. Gamers playing by themselves would know what I mean, although it wouldn’t be referred to as “AI” early on. Early analytics software also operated within the same framework. Arguably, even your word processor’s spell checking function is a form of AI, in that it followed a programmed set of rules and attempted to correct typographical errors, albeit with some degree of human input. The rise of deep learning in the 2010s—powered both by computing improvements and access to a larger amount of data through which it can derive decisions from—further rejuvenated the AI field, followed by developments in machine learning such as natural language processing and sequence processing.
The first breakthrough in the current age of AI arguably came in 2015, when DeepMind, a subsidiary of Google, unveiled AlphaGo. The computer program was taught only the rules to the board game Go; it devised its own strategy and eventually defeated several human professional Go players, including the then top player in the world, Ke Jie. The next major breakthrough came five years later, when OpenAI released the large language model GPT-3, which is capable of generating high-quality human-like text: this would form the backbone for ChatGPT, which was launched in 2022 and saw the breakout of artificial intelligence from research labs to wider public consciousness.
It may be futile to try to summarize an ever-evolving field into a few paragraphs that will remain static forever, but let’s try to break down the artificial intelligence tools available to us right now, in a nutshell.
The majority of us would be familiar with generative AI, which generates text, still or moving images, audio, or other forms of data. The AI applications we are most familiar with—chatbots such as ChatGPT, Claude, Copilot and Gemini; image generators such as Midjourney; and video generators such as Sora—would be examples of generative AI.
In essence, these applications learn the patterns and structures of the data they are trained on, and generate new data based on them, typically through instructions based on natural language prompts—the way we normally speak and write—rather than through a specific code. These applications were made possible by the aforementioned advances to deep learning, particularly in the way it classifies images and processes language; as well as the development of the “transformer” model, which allows these applications to learn faster from the inputs it is provided.
Generative AI’s ability to generate huge amounts of data, and to learn from inputs over time, have proven useful in several industries. Already, such models are used to improve the movement of robots, by learning from prior use; automating time-intensive tasks such as 3D modeling; and accelerating research into certain mathematical and scientific problems. However, most us would be more familiar with how generative AI models are used to create text and images, something we already see on articles (although I swear this piece is fully written through fingers bashed on a keyboard), social media videos, and even in the AI-powered functions of the software we use. It has also begun to change workflows in several professions, from marketing and journalism, to software development.
The principles behind generative AI also have applications across various supply chain functions. AI applications can be trained on historical logistics data and forecasting models to provide more instantaneous forecasts, and perhaps relay them in a manner that can be more easily understood by those needing the data. In addition, these applications can enhance existing technologies such as pick-by-voice systems, allowing it to understand different speaking accents and removing barriers to use—a pretty big leap considering how most of these systems are trained on Western languages and patterns.
The other form of artificial intelligence making waves is agentic AI. In contrast to generative AI, which is more tool-based and performs a narrow set of tasks (albeit learns from it and improves its output over time), agentic AI can use different tools with some degree of autonomy in the pursuit of accomplishing the tasks it is set out to do.
Task automation has existed long before in various computing systems: you may recall some of your “geek” friends hitting a button that allows their computers to perform a series of tasks in a row, usually powered by scripting languages. Over time, task-performing agents evolved as it moved from simple if-then logic to more intricate decision tree models. The modern AI agent, however, utilizes LLMs as well as certain functions—for example, the Model Context Protocol introduced by Anthropic, the developers of Claude—that allow the agent to be aware of the context it operates upon, and learns from it to improve the way it interacts with the provided inputs and the various tools at its disposal.
Unlike generative AI, agentic AI has not captured the public imagination as much because of the lack of a wide-appeal application similar to ChatGPT. However, certain Claude models, like Claude Code and Claude Design, have functionalities that allow it to perform specific tasks autonomously. AI agents are also widely used in software development, cryptocurrency, social media, and particularly in customer support, with chatbots providing more contextual, natural support at a lower cost than human agents. More notably, AI agents are now being used to automate administrative tasks.
For us across the supply chain, the potential of agentic AI goes beyond automation. Provided it has access to enough quality data, agents could conceivably automate various logistics functions like route planning, and even adjust plans in real time in response to situations such as road congestion, flooding and unexpected fleet limitations. Taken to the extreme, as the folks at FarEye note in their latest Eye on the Last Mile report, agents can perform these tasks wholly autonomously, with human input limited to setting rules and maintaining governance functions—although, again, in the real world this would have to depend on quality data at all levels, and not just from within the logistics operation involved.
Even then, we’re just barely scratching the surface on artificial intelligence. The above examples—as well as technologies such as self-driving cars and robots being used in the medical field—can be classified as “weak AI”, or AI that is good at specific tasks. Some researchers and developers are still keen for the holy grail: “strong AI”, which can be defined as artificial general intelligence, or the ability of a machine to solve multiple kinds of problems, rather than a specific set of problems it is trained to solve. We are certainly a long way away from superintelligence, or even artificial consciousness, which would put a machine’s capabilities above that of a human’s. Arguably developments such as quantum computing—which could process tasks much faster, but so far is seen to be limited to certain scientific applications—could lead us there, but perhaps the current backlash to AI among certain parts of the public sphere could lead to a reassessment of priorities.
Just as unavoidable as the talk of how AI is going to change so many things in our lives and how we should learn it to not be left behind, is talk of how disruptive and bad AI has become to some sectors. As someone who straddles both the creative and business worlds, it’s a perspective I can’t help but see and have to work with.
The creative sectors do have a lot to lose with the way generative AI applications are being utilized these days. The supposed democratization of art has led to the onslaught of “AI slop”, of low-quality content being churned en masse and, in some cases, even sold for a profit. These AI models are trained on publicly available text and art, including social media posts made by creatives to promote their wares—and now these models are conceivably capable of imitating those very art styles these creatives trade on. The same goes for video and audio generated by AI; this has led to various copyright tussles between rights holders and developers such as Suno.
At the very least, a lot of AI-generated text is written in a predictable and unnecessarily confident manner; a lot of AI-generated posters lean on the same visual tropes. As a creative, it frankly hurts my eyes—and that this sameness is the cause for some of my contemporaries losing opportunities to earn rankles a bit more.
But arguably more important is the environmental impact of these large-scale AI applications. These rely heavily on data centers that house the processing power to allow users from around the world to generate content fast enough. These facilities are heavy users of water and electricity, and already communities that house them have complained of a reduced quality of life due to power outages, declining water quality and increasing costs. In many countries, including the Philippines, communities have begun to more actively organize against the presence of these processing hubs, although it has not stopped investment from continuing in this front. Countries are starting to treat the construction of these data centers in their territories as essential to future economic competitiveness, with our Southeast Asian neighbors rolling our greater government support for the establishment of these facilities.
In the broader scheme of things, artificial intelligence has become a new arena for global domination. Already an “AI race” of sorts between the United States and China has materialized, first through firms such as OpenAI and Anthropic on the former side and DeepSeek and Kimi on the other, and then through the firms that manufacture the chips that power AI systems such as Nvidia and SMIC. Now the United States, through its Pax Silica Initiative, and China, through the World Artificial Intelligence Cooperation Organization, is seeking to enlist countries to secure their dominance of the AI space, with the Philippines joining the former and committing to house an “AI-native industrial hub” in Tarlac. (Within ASEAN, Singapore has joined Pax Silica, while Indonesia and Malaysia have joined WAICO.)
The geopolitical angle raises new questions surrounding the ultimate “ownership” of the AI technologies and infrastructure more and more businesses are relying upon. Who owns the data that is used to train these AI models? Who owns the processing power that enables these technologies? Will we end up with opposing standards that would be interoperable with each other—and what would the impact of these barriers be on a global economy that still tries to establish and strengthen ties amidst current moves towards nativist and populist spheres of influence?
Perhaps ironically, the AI race—which was supposed to democratize so many things that previously required high costs to enter—has resulted in more expensive consumer devices, such as laptops and smartphones, thanks to the cost of memory and chips being driven up by higher demand from data centers and AI processing hubs. In addition, the supposedly essential AI technology is becoming difficult to reach for those with more limited budgets, whose access to widely-available applications are limited to certain tasks and a smaller allocation of tokens (the currency of the likes of Claude, Gemini and ChatGPT) per day because of high subscription fees.
It’s also not inconceivable that AI would be used by nefarious actors to further certain political interests. This goes beyond the use of generative and agentic AI to create and spread disinformation much faster. Imagine various (possible state-sanctioned) entities using AI to hack into a rival’s important systems. The Five Eyes intelligence alliance has warned that we may be months away from AI models that can initiate cyber attacks on government and businesses. Already OpenAI has admitted to one of its models going rogue, escaping its testing sandbox and initiating a hack on another AI business.
It may still be quite a while from now, but let’s imagine for a moment what an AI-powered supply chain operation would be like.
We already have some idea as to where AI applications can make the most impact. It could deliver faster demand forecasts, for example. It could better anticipate surprise product shortages and reallocate stocks to locations faster. It could improve and optimize deliveries by factoring in available vehicles, delivery load and external factors to ensure products are shipped consistently and in time. And of course, administrative and non-core tasks that keep a supply chain operation ticking, like security and accounting, can also be automated by AI.
In any case, the vision of a fully-automated logistics operation—where agents and machines make the vast majority, if not all, of the day-to-day decisions—may still be quite far away. That would mean every single stakeholder across the supply chain—manufacturers, logistics providers, end customers, and particularly regulatory and trade-facilitiation bodies—is ready and on the same page. That degree of coordination needed means the idea of getting “logistics as a service” off the shelf may be quite far away, if it’s even doable. Or, perhaps, not for a large-scale, national or regional logistics operation.
In addition, there’s still the possibility of mistakes, or worse, hallucinations in the output. For now, any logistics operation that would utilize AI would have to approach it not as a replacement for human workers, but as augmentation: to allow them to get information and insight faster, and to make better decisions sooner. Even the vision of humans only managing governance of AI agents may still be a few years away to work efficiently. And of course, those working more physical roles such as driving or sorting inventory would not be directly affected by this form of AI, although robotics and autonomous vehicles, which also operate on AI principles, would like to have a (possibly LLM-generated) word, too.
The rise of AI has rightly focused on whether it will eliminate entire categories of workers. While this disruption has happened in some sectors—take, for example, the creatives who have been most vocal about how AI, which has scraped their works, has reduced their opportunities to earn—advocates believe that the nature of jobs will shift like in other key points in history. In any case, the disruption seems to be happening much slower than thought. A report from Anthropic notes that AI use has yet to fully reach its theoretical capability in the workplace, although the hiring of younger workers in “exposed occupations”—in fields like customer services, marketing, sales and data management—has slowed down.
Another focus has been on arming both the current members of the labor force, as well as future ones, with the skills to work with, and alongside, AI tools. As it stands, the former may not have the time and energy to pursue this line of inquiry alongside their current work and other responsibilities, unless, perhaps, their company provides them enough time—on company dime—to gain AI proficiency. Casual learners may be prevented from going all in by the aforementioned high subscription costs of the most powerful—and thus most instructional—versions of popular AI tools like Claude.
As for our students, well, let’s be frank—have you seen the state of our educational system? We don’t have enough classrooms. Our teachers are heavily burdened as it is. Constant climate-related class suspensions, and the spotty implementation of alternative learning set-ups, have been to the detriment of students’ learning. And don’t get me started on recent anxieties about security in schools. I can imagine that while there may be government initiatives to introduce more AI learning to students, the capacity to do so is woefully constrained by institutional deficiencies. Uninformed skepticism from parents and other educators don’t help, either. We’re certainly trailing behind our ASEAN neighbors, where tech literacy is not seen as a plus in the basic education curriculum, but rather, as an essential part—a factor which has allowed them to be in a better position to upskill despite the breakneck pace of development in this front.
On the other hand, an excessive reliance on AI to automate logistics processes could also be incompatible with how logistics talent is currently developed. One may call it old school, but in supply chain, a lot of knowledge is obtained in the workplace, during day-to-day operations, and through interactions with stakeholders and colleagues. An advanced AI agent may still not be able to capture the nuances of local knowledge and customs, as well as collaborative relationships among partners, suppliers and in some cases, even rivals. If this institutional knowledge is depleted, it may be difficult to solve certain problems caused by an AI agent going rogue, for example.
It’s not quite right to say artificial intelligence is here to stay, because in many ways, it has been here for quite a while now. That said, developments in the past decade, and especially in the past couple of years, can genuinely change how our supply chains operate, and how they generate value for partners and customers alike. It is not a technology we can look away from or ignore.
However, it’s worth recognizing that there is so much we don’t know and cannot clearly discern from the ongoing AI boom. As such, much of the discussion around it is driven by those in extreme positions: those who are motivated by profits, and who are keen to portray AI as inevitable and inescapable; and those who fear they have much to lose from the arrival of AI, and are keen to see the technology stopped at all costs. There are clear benefits and there are valid concerns, and these must be discussed as objectively and as soberly as possible so everyone can create an informed and nuanced understanding of the impact AI will have on our lives now and in the future.
Considering how this boom has been illustrated by the creep of various “AI-driven” functionalities on pretty much everything we do on a laptop or mobile device—and how it feels we don’t have much of a choice, which raises suspicions—it is important to know, and feel, that we are, and should, ultimately be the masters of AI, and not the other way around. And certainly not the tech giants with money to make.
With all that in mind, I think it is possible to have a middle ground in this discussion. For one, it is true that a lot of AI’s more promising functionalities at the moment is powered and made more reliable in part by cloud computing—which means being online, and using those data centers with a worrying environmental footprint. But it may be possible to utilize this technology while keeping all the infrastructure within your organization. It may also help to minimize the risk of data breaches and exposure, but there may be a trade-off with the speed of finishing tasks and decision-making, as well as reliability.
AI can also be used to ideate and model potential new approaches and processes, perhaps through a digital supply chain twin, before phased real-world implementation. It can also help present unforeseen scenarios when forecasting demand or mapping out new routes. And of course, it can help automate repetitive tasks that can free up employees to do more essential work, important particularly for smaller businesses.
However AI ends up developing in the coming months and years—whether this current proves to be a bubble waiting to burst, or whether we really are on the cusp of a revolution—we should not be left behind when it comes to policy. The government should be proactive when it comes to how we use AI, how we learn AI, and how we maximize AI’s potential. This goes beyond regulations directly governing AI use, or facilitating investments in AI infrastructure such as data centers and its energy requirements; we should also bolster policies regarding aspects such as tech literacy, data privacy, cybersecurity and digital governance.
But perhaps the most important investment our government can make to allow the Philippines to truly take advantage of AI is in its education system. Beyond classrooms, teachers and a better way to do online classes—not that these aren’t important—teaching the next generation of the country’s labor force the fundamentals of AI, as well as the ethics of using it, is important.
Key to this is not treating AI education as just an add-on, but integrating it into the curriculum. For instance, Singapore’s strategy includes introducing younger students to basic AI concepts, with specific competencies taught to older students as part of their regular subjects. Higher education students would also be required to take up foundational AI courses covering both skills as well as ethical uses. Perhaps most notably, the framework devised by the country’s Ministry of Education is designed to utilize AI as a thinking partner rather than a shortcut to getting higher scores.
Admittedly, one of our biggest hurdles when it comes to maximizing AI is the opinion of the ordinary Filipino. They may see this just as a tool for easily making posters, and we know it just scratches the surface. Perhaps worse, they may only know of the (valid) criticism towards Pax Silica and be against the whole technology as a result. Let’s be frank: AI has long been here, but the leaps in recent years will impact the way we live in ways that aren’t clear yet. Leveraging it would allow the Philippines to at least keep up with its neighbors, and to not fall behind in the global economy with which it wants to forge stronger ties with.
As stakeholders both begin laying the groundwork, we should start talking about AI with an open yet critical mind. Go beyond the hype and the fears, and we’ll better understand what we can do—as supply chain managers, as customers, and as global Filipinos.