English
Based on โHow AI Will Play Out, Explainedโ from Economics Explained Watch the original video
The Invisible Tsunami: How AI is Reshaping Our World, One Task at a Time
Episode summary
This video from Economics Explained revisits the automation debate, contrasting early predictions with AIโs current impact. The channelโs 2019 thought experiment explored three futures โ good (UBI, leisure), bad (two-tiered society, gig work), and ugly (mass unemployment, starvation) โ for a fully automated world. However, AI, particularly LLMs like ChatGPT, defied expectations by first impacting white-collar outsourced service jobs in developing nations, like the 89% of such roles in the Philippines at high risk, rather than factory floors.
The practical takeaway is that AI is creating a stark divide: it acts as โcomplementary capitalโ for skilled workers, boosting their productivity, but โsubstitutive capitalโ for routine jobs, replacing them entirely. This dynamic, coupled with AI capital ownership concentrated in a few wealthy nations (USA and China projected to capture 70% of AIโs $15.7 trillion GDP boost by 2030), is accelerating global inequality and brain drain, making it harder for developing economies to catch up.
Four years ago, the conversation around automation felt like a distant hum โ a theoretical โif and whenโ rather than a pressing โnow and how fast.โ Economists pondered factory robots and software rendering accountants obsolete, running thought experiments about a fully automated world. Then, in November 2022, ChatGPT burst onto the scene, transforming the academic debate into a lived reality, impacting everything from lunch breaks to legal briefs. The future wasnโt just coming; it had arrived, and it was far more nuanced and unsettling than anyone had predicted.
This isnโt merely a story of technological advancement; itโs a rapidly evolving economic saga. Our understanding of AIโs impact has shifted dramatically, moving from initial abstract concerns to a stark confrontation with its real-world implications, particularly for global economies and the very fabric of work. The journey of understanding has been a winding one, marked by surprising turns and a growing realization that our traditional economic lenses may no longer be sufficient.
The Old View: Automationโs Inevitability and Economic Fundamentals
In 2019, the argument for automationโs inevitability was built on historical precedent. Steel made bronze obsolete, the internal combustion engine replaced steam power, and self-checkout kiosks displaced human workers. The logic was simple: if a machine could do a job more efficiently, it would. The question was merely a matter of scale and timing.
To illustrate this, economists often turn to the fundamental principles of supply and demand. Imagine a graph where the x-axis represents the quantity of apples and the y-axis represents their price. The demand line slopes downwards: if apples are cheap, everyone wants them; if theyโre expensive, few do. Conversely, the supply line slopes upwards: if apples are cheap, no one bothers to grow them; if theyโre expensive, everyone rushes to supply them. The point where these lines cross represents the market equilibrium โ a reasonable price for a reasonable quantity, where both buyers and sellers feel theyโre getting a fair deal.
Now, substitute โapplesโ for โaccountants.โ The โpriceโ becomes their salary. If accountants earned a paltry sum, every business would want to hire them, even if only to fetch coffee. But few would choose the profession. Conversely, if they commanded exorbitant salaries, everyone would aspire to be an accountant, but businesses would be hesitant to hire. The market equilibrium, where supply meets demand, determines the actual number of accountants employed and their average salary.
This is where automation enters the picture. The supply and demand curves for labor are dynamic. When companies outsource accounting work to countries like the Philippines, they effectively increase the global supply of accountants. This allows them to hire the same number of accountants for less money, contributing to wage stagnation in wealthier economies. On the demand side, technological advancements like calculators, spreadsheets, and accounting software mean that fewer accountants can do the work of many. A company that once needed 200 accountants might now only need 20. The remaining skilled accountants are then forced to compete on wages, driving down the overall value of their labor. The logical, if chilling, conclusion of this trajectory is a highly automated machine capable of performing all accounting tasks without any human intervention at all.
Imagining the Future: Three Paths for an Automated World
In 2019, economists bravely ventured into predicting what a fully automated world might look like, outlining three potential scenarios: the good, the bad, and the ugly.
The Good: This utopian vision paints a picture of humans freed from toil, served by robot butlers, with ample time for recreational and creative pursuits. Businesses running these machines would be heavily taxed, or the machines themselves would be government-owned, funding a universal basic income (UBI) for all. People could still work in niche human-centric roles or invest their UBI for extra income, forming an upper-middle class. Living standards for most would surpass todayโs middle class, thanks to the abundance generated by automated labor.
However, even this ideal future presents unforeseen challenges. With no career strain, no childcare costs, and no need for employment, birth rates would likely soar. While humans currently contribute more to society than they consume, in a post-automation world, machines would handle all productive labor. Humans would still require food, housing, and transport โ finite resources on a finite planet. This scenario could lead to a future where humans, despite their leisure, become a net drain on societal resources, posing a fundamental question about our โvalue.โ
The Bad: A more dystopian outcome envisions a basic UBI, barely covering essentials. Society would cleave into two distinct classes: the ultra-rich who own the companies and the robots, and a vast โpeasant classโ forced into precarious gig-style jobs. Having children would be discouraged, perhaps even financially crippling, leaving procreation largely to the wealthy. The rich, despite their fortunes and robot servants, would live in isolated โfortresses,โ much like the wealthy enclaves in Johannesburg, surrounded by a world filled with violence and resentment. This scenario suggests a breakdown of social cohesion, where immense wealth exists amidst widespread discontent.
The Ugly: This is the bleakest vision, likely to emerge in countries where the ethos of โwork for a livingโ persists. Universal basic income would be non-existent or minimal, and social welfare would be starved. In a purely transactional economic system, unemployable humans would have no economic value. Businesses, no longer needing human labor or a mass consumer base, would primarily trade with each other, catering to the needs of their wealthy owners. The average consumer, often considered the backbone of the economy, would become economically irrelevant. This scenario predicts massive population declines, stemming from people choosing not to have children they cannot support, or, in a much darker turn, widespread starvation. It forces us to confront a future where humans are no longer the creators of value, and their very existence may be deemed expendable.
AIโs Unforeseen Arrival: The Real-World Impact (2025 Perspective)
The 2019 predictions, while thought-provoking, couldnโt anticipate the speed or direction of AIโs arrival. The perceived threat was primarily for manual, repetitive factory work. The reality was almost the opposite. When AI truly landed, it targeted call centers in Manila, data entry workers in Dhaka, and the vast outsourced service economies that developing countries had meticulously built over three decades. Suddenly, office workers, not factory laborers, found themselves on the front lines of disruption.
This shift has had a profound and immediate impact on countries like the Philippines and Bangladesh. For years, these nations have thrived on business process outsourcing (BPO), a multi-billion-dollar industry encompassing customer service, billing, transcription, and tech support. These jobs were once considered safe, requiring language skills, contextual understanding, and a โhuman touchโ that machines couldnโt replicate. However, large language models (LLMs) like ChatGPT have rapidly become adept at these very tasks, performing them in seconds, at a fraction of the cost, and without the need for health insurance, vacation days, or HR departments.
The numbers are staggering: the IMF estimates that 89% of outsourced service jobs in the Philippines are at high risk of AI automation, impacting over a million people. Bangladesh faces a similar predicament. This doesnโt just threaten individual livelihoods; it jeopardizes entire industries that contribute significantly to these countriesโ GDP, undoing decades of economic growth. The ability of AI to deliver the same quality of work faster and cheaper means thereโs little economic incentive to continue outsourcing. This could lead to โreshoring,โ where jobs return to wealthier nations where local automation can now rival offshore labor on price, completely flipping the script on the global outsourcing model.
Even in rich countries, AI is creating new divides. The US Bureau of Labor Statistics predicts declines in roles like cashiers, bank tellers, and customer service representatives. Some estimates suggest 7.1 million US jobs could vanish in the next five years, with up to 47% of current roles at risk. While this sounds alarming, itโs also worth noting the immense financial investment in AI development. Companies that have poured trillions into this technology are now seeking returns, and cutting millions of workers from payroll is a direct, albeit ruthless, path to achieving those returns. This creates an incentive to amplify the โscare campaign,โ as what sounds horrifying to the general public often translates into opportunity for investors. The trend lines are clear: AI is accelerating the gap between those who get ahead and those who fall behind.
The Widening Divide: AI as Concentrated Capital
The disparate impact of AI โ supercharging growth in rich countries while threatening the economic survival of others โ stems from its nature as a specific kind of capital. Economists distinguish between two types:
Complimentary Capital: Most past technologies functioned as complimentary capital, enhancing human productivity rather than replacing it. A combine harvester didnโt eliminate farm workers; it made each worker dramatically more efficient. Similarly, a financial analyst using AI to scan reports gains insights faster, freeing them for strategic thinking. A doctor leveraging AI for diagnostics can dedicate more time to direct patient care. In these scenarios, AI acts as a multiplier, making skilled human expertise more valuable and leading to increased productivity, wages, and living standards.
Substitutive Capital: AI, however, increasingly acts as substitutive capital for routine, process-driven work. An AI-powered chatbot doesnโt make a customer support agent faster; it replaces them entirely. A sophisticated code generator doesnโt assist a junior developer; it renders their entry-level skills obsolete. As capital becomes more capable, it requires less human labor to function, leading to direct displacement.
The problem is compounded by the highly concentrated ownership of this new capital. Most major AI breakthroughs originate from a handful of elite firms in the US and China. These companies benefit from a powerful โdata network effectโ: the more data they collect, the better their AI models perform; the better their models, the more users they attract; the more users, the more data they generate. This feedback loop rapidly concentrates market power and profits. PwC estimates that while AI could add $15.7 trillion to global GDP by 2030, a staggering 70% of that wealth is projected to go to just two countries: the USA and China, because they own AI.
This concentration extends beyond software to the very hardware that powers AI. Over 90% of CPUs and GPUs are designed and manufactured in just five countries: the US, Taiwan, China, South Korea, and Japan. This means a tiny handful of nations not only control the AI systems themselves but also the foundational components that make AI possible. The result is a system where AI primarily benefits those who already own the assets, while displacing those who donโt.
Furthermore, AI rewards specialized skills that are difficult to scale globally. The advanced education, reliable infrastructure, and access to cutting-edge technology required to build and train AI models are overwhelmingly concentrated in wealthy nations. This means the most valuable AI jobs are largely inaccessible to workers in emerging markets. When talented engineers from these countries do acquire such skills, they are often recruited by global tech giants or relocate to AI hubs, accelerating a โbrain drainโ that further hinders poorer nationsโ ability to catch up.
Echoes of the Past: Lessons from Industrial Disruption
History offers stark warnings about the social and economic consequences of rapid technological disruption. The industrial automation and large-scale outsourcing of the 1980s and 1990s decimated manufacturing sectors in the US and Western Europe. In America alone, over 7 million factory jobs disappeared between 1980 and 2010, most never returning. The Midwest, once the industrial heartland with cities like Detroit, Cleveland, and Youngstown, bore the brunt. Robotic welders, computer-run assembly lines, and cheaper overseas labor evaporated stable middle-class jobs, leading to factory closures, soaring unemployment, and the collapse of local economies. Beyond mere job loss, these towns experienced drops in life expectancy, rises in opioid addiction, and struggling schools. The few jobs that returned often paid less and lacked the stability that once supported entire communities.
The UK witnessed a similar transformation, with coal mining, shipbuilding, and steel plants across northern England and Scotland shutting down. Even today, areas like Sheffield and Sunderland lag behind the rest of the country in income and social mobility. The lesson is undeniable: even if technology promises long-term improvements, its short-term impact can be devastating, and once inequality takes root, it is incredibly difficult to reverse.
Charting a New Course: Strategies for Adaptation
Given the clear and present challenge, what can be done? The frontline effects observed in lower-income countries like the Philippines and Bangladesh offer a critical playbook. These economies demonstrate which jobs are most vulnerable, where the risks are highest, and the consequences of governmental actionโor inaction.
For instance, the Philippine government has launched a national AI strategy aiming to retrain over a million workers by 2028. Bangladesh is developing a policy framework to cultivate AI talent, modernize education, and support tech startups, striving to position itself in the AI-enabled services market while safeguarding jobs through upskilling and digital inclusion. While the sufficiency of these efforts remains to be seen, they provide a valuable blueprint for wealthier nations.
Two simultaneous investments are crucial:
- Heavy investment in AI infrastructure: This is fundamental to participate in and benefit from the AI revolution.
- Equally heavy investment in people: This includes conventional computer science education, but critically, also the development of uniquely human skills that AI struggles to automate: critical thinking, complex problem-solving, effective communication, and creative decision-making. Research indicates that AI adoption actually increases demand for these human-centric skills more often than it eliminates jobs entirely.
Building an accessible digital economy is paramount. With nearly 2.6 billion people worldwide still lacking internet access, the opportunity to compete or even participate in the emerging AI economy is non-existent for a significant portion of the global population. The World Bank estimates that every 10% increase in broadband access can boost GDP growth in developing countries by up to 1.4%, even before factoring in AIโs benefits. Alongside retraining, countries need policies that expand broadband access, reduce device costs, and equip more people with essential digital skills.
Robust social safety nets are also vital. They act as economic buffers, providing displaced workers with the time and resources needed to adapt, retrain, and re-enter the labor market from a position of strength. If AI allows businesses to grow while workers lose income, the economy risks hollowing out: productivity rises, but consumption falls, innovation continues, yet inequality grows, ultimately dragging down overall economic growth. To ensure AI broadly boosts productivity, not just corporate profits, we must fundamentally rethink how we design and share the value it creates, addressing questions of who builds, governs, and ultimately benefits from AIโs massive productivity gains.
The Iceberg Beneath: MITโs New Lens on AIโs Impact
Just seven months after these strategies were discussed, a groundbreaking study from MIT reframed the entire conversation. The question of who AI impacts turned out to be far more complex than anticipated because we had been asking about jobs when we should have been asking about tasks.
The popular narrative, fueled by headlines, focused on programmers, junior developers, and tech layoffs โ jobs supposedly under direct threat. While this anxiety is partly justified, the MIT study argues itโs largely misdirected. AI doesnโt typically replace entire jobs overnight; it replaces the tasks within them. A lawyer doesnโt disappear, but the hours spent reviewing routine contracts might shrink. A journalist still writes, but the time dedicated to background research, pulling quotes, and fact-checking becomes compressed. This subtle distinction is critical because our entire economic system โ how we measure work, track productivity, and plan for the future โ is built around jobs, not tasks. GDP, unemployment figures, and wage data count jobs and people, but they were never designed to peer inside a job and assess which specific components AI can already perform. Consequently, the disruption emerges where we arenโt looking, rendering existing workforce preparation plans ill-suited.
To address this blind spot, MIT developed the Iceberg Index: a new tool to map where AI capabilities and human skills overlap, weighted by the economic value of that work. The name itself is telling. When measuring the work AI can technically perform within the tech sector, it accounts for about 2.2% of total US labor market wage value, roughly $211 billion. This is the visible tip of the iceberg, where all the public discourse has been concentrated.
However, when the same methodology is applied to the entire economy, the number dramatically jumps to 11.7% of total US labor market wage value, or approximately $1.2 trillion โ five times larger. This is the vast, hidden underwater portion of the iceberg. It encompasses highly educated, well-paid professionals in industries and sectors that have generated virtually no anxious headlines about AI.
The Iceberg Indexโs methodology is rigorous. It began by creating a digital representation of 151 million American workers across 923 occupations and 3,000 counties. To map the skills required for each occupation, they utilized ONET, a US Department of Labor database that breaks down jobs into component skills (e.g., analyzing data, critical thinking, programming), each with an importance and difficulty rating derived from surveys of real workers. They then applied the same ONET skill taxonomy to over 13,000 production-ready AI tools, cataloging what each AI tool can do. This allowed for a direct, โapples-to-applesโ comparison between human worker skills and AI system capabilities.
The result for each occupation is a percentage indicating how much of its wage value AI can technically perform. This focus on โwage valueโ is crucial. Automating 60% of an accountantโs time doesnโt necessarily mean automating 60% of their economic value. The index reflects where the true economic exposure lies, capturing the worth of the tasks AI can take on. The index deliberately avoids accounting for physical robotics or predicting actual job losses or timelines; itโs an โearthquake risk map,โ showing which buildings are on a fault line, not when or if a tremor will hit.
Unveiling the Hidden Threat: Who is Really Exposed?
The Iceberg Indexโs findings are startling. While 2.2% of wage value in the tech sector, or $211 billion across 1.9 million workers, has dominated headlines, this is just the tip. The same capabilities that make a coding assistant useful (document processing, routine analysis, data handling) are also vital to hundreds of non-tech occupations: financial analysts, HR coordinators, insurance claims processors, and legal secretaries. When the index is run across all these skills, the exposure jumps to 11.7% of the total US labor market โ $1.2 trillion in wage value. This means the anxiety about AI has been aimed at roughly one-fifth of the actual problem; the remaining four-fifths have been sitting on a fault line for which no one โ no government, no company, no individual worker โ has been preparing.
And the people on this fault line are not who youโd expect. According to a separate study tracking actual AI usage, the most exposed group earns 47% more on average than the least exposed, is nearly four times as likely to hold a graduate degree, and is 16 percentage points more likely to be female. In essence, these are professionals whose working day revolves primarily around reading, writing, analyzing, and summarizing information โ individuals who, by all reasonable societal measures, did everything right and excelled.
While the technical capability of AI is already immense (e.g., theoretically handling 94% of computer and math workersโ tasks), observed professional use is currently much lower (around 33%). This โgapโ is whatโs keeping much of the exposure theoretical, for now. Friction points like regulation, integration challenges, and the continued need for human oversight are holding AI back. However, these are temporary barriers that are likely to diminish as the technology matures.
The leading edge of this disruption is already visible in hiring data. IBM has replaced portions of its HR department with AI tools, and Salesforce has paused hiring engineers and lawyers, noting AIโs increasing capabilities. Entry-level employment in AI-exposed occupations has dropped 14% since the pre-ChatGPT era, and entry-level job postings across the US have fallen 35% since January 2023. This is likely just the beginning, as job postings typically decline before actual employment figures.
Perhaps most surprisingly, the states with the highest exposure values are not the tech hubs of California, Washington, or New York, but rather South Dakota, North Carolina, and Utah. Their economies are heavily concentrated in administrative and financial work โ precisely the sectors where AIโs hidden capabilities pose the greatest threat.
The implications of the Iceberg Index are profound. It reveals that the economic transformation driven by AI is far more pervasive and insidious than previously understood. We are not just facing a localized storm in the tech sector, but a silent, systemic shift that is already underway, threatening the economic foundations of well-educated professionals and entire regions that have yet to grasp the full extent of their vulnerability. The challenge is not just to adapt to AI, but to fundamentally redefine work, value, and societal structures before the hidden depths of the iceberg fully reveal themselves.
ํ๊ตญ์ด
โHow AI Will Play Out, Explainedโ โ Economics Explained ๊ธฐ๋ฐ ๊ธฐ์ฌ ์๋ณธ ์์ ๋ณด๊ธฐ
AI๊ฐ ๋ฐ๊พธ๋ ๋ฏธ๋: ๊ฒฝ์ ๋ ์ด๋๋ก ํฅํ๋๊ฐ?
์ํผ์๋ ์์ฝ
AI์ ๋ฐ์ ์ ๊ณผ๊ฑฐ ์๋ํ ๋ ผ์์๋ ์ฐจ์์ด ๋ค๋ฅธ ๊ฒฝ์ ์ ํ๊ธ๋ ฅ์ ๊ฐ์ ธ์ค๋ฉฐ, ํนํ ๊ฐ๋ฐ๋์๊ตญ์ ์๋น์ค ์์์์ฑ ์ฐ์ ์ ์ํํ๊ณ ์ ์ง๊ตญ ๋ด์์๋ ๊ณ์ธต ๊ฐ ๊ฒฉ์ฐจ๋ฅผ ์ฌํ์ํฌ ๊ฒ์ด๋ผ๋ ๊ฒฝ๊ณ ๊ฐ ๋์จ๋ค. 2019๋ ๋น์์๋ ๊ณต์ฅ ๋ ธ๋์ ๊ฐ์ ๋ฐ๋ณต์ ์ธ ์ก์ฒด๋ ธ๋์ด ์๋ํ์ ์ฃผ์ ๋์์ด ๋ ๊ฒ์ผ๋ก ์์ํ์ง๋ง, ์ฑGPT(ChatGPT)์ ๊ฐ์ ๋๊ท๋ชจ ์ธ์ด ๋ชจ๋ธ(LLM)์ ๋ฑ์ฅ์ผ๋ก ์ฝ์ผํฐ, ๋ฐ์ดํฐ ์ ๋ ฅ, ์ ์ฌ(่ฝๅฏซ) ๋ฑ ์ธ์ด ๋ฐ ๋งฅ๋ฝ ์ดํด๊ฐ ํ์ํ ์ฌ๋ฌด์ง์ด ๋จผ์ ๋์ฒด๋ ์๊ธฐ์ ๋์๋ค. ์ผ๋ก๋ก ๊ตญ์ ํตํ๊ธฐ๊ธ(IMF)์ ํ๋ฆฌํ ์์์์ฑ ์๋น์ค ์ง์ ์ 89%๊ฐ AI์ ์ํด ์๋ํ๋ ์ํ์ด ์๋ค๊ณ ์ถ์ ํ๋ฉฐ, ์ด๋ ์ง๋ 30๋ ๊ฐ ๊ฐ๋ฐ๋์๊ตญ์ด ๊ฒฝ์ ์ฑ์ฅ์ ํต์ฌ ์ ๋ต์ผ๋ก ์ผ์๋ ์์์์ฑ ๋ชจ๋ธ์ ๊ทผ๋ณธ์ ์ผ๋ก ๋คํ๋ค ์ ์๋ค.
์ด๋ฌํ ๋ณํ ์์์ AI๋ ๋ ๊ฐ์ง ํํ๋ก ์๋ณธ ์ญํ ์ ํ๋ค. ๊ณ ์๋ จ ์ง์ ์์๋ ์ธ๊ฐ์ ์์ฐ์ฑ์ ๋์ด๋ ๋ณด์์ ์๋ณธ(complimentary capital)์ผ๋ก ์์ฉํ์ฌ ๊ธ์ต ๋ถ์๊ฐ๋ ์์ฌ์ ์ญ๋์ ๊ฐํํ์ง๋ง, ๋ฐ๋ณต์ ์ธ ์ ๋ฌด์์๋ ์ธ๊ฐ ๋ ธ๋์ ์์ ํ ๋์ฒดํ๋ ๋์ฒด์ ์๋ณธ(substitutive capital)์ผ๋ก ๊ธฐ๋ฅํ๋ค. AI ๊ธฐ์ ๊ฐ๋ฐ๊ณผ ์์ ๊ฐ ๋ฏธ๊ตญ๊ณผ ์ค๊ตญ์ ์์ ๊ธฐ์ ์ ์ง์ค๋์ด ์๊ธฐ ๋๋ฌธ์, AI๋ก ์ธํ ์์ฐ์ฑ ํฅ์๊ณผ ๋ถ์ ๋๋ถ๋ถ(PWC ์ถ์ ์น์ ๋ฐ๋ฅด๋ฉด 2030๋ ๊น์ง ๊ธ๋ก๋ฒ GDP์ 15.7์กฐ ๋ฌ๋ฌ ์ฆ๊ฐ ์ค 70%๊ฐ ์ด ๋ ๊ตญ๊ฐ์ ์ง์ค)๋ ์ด๋ค ๊ตญ๊ฐ์ ๊ท์๋ ๊ฐ๋ฅ์ฑ์ด ๋๋ค. ๋ฐ๋ผ์ ๊ฐ์ธ๊ณผ ๊ตญ๊ฐ๋ AI๋ฅผ ํ์ฉํ๋ ๋ฅ๋ ฅ์ ํค์ฐ๊ณ , AI๊ฐ ๋์ฒดํ ์ ์๋ ์ฐฝ์์ , ์ ๋ต์ ์ญ๋์ ํฌ์ํ๋ฉฐ, ๊ต์ก ์์คํ ์ ์ฌ์ ๋นํ์ฌ ๊ธ๋ณํ๋ ๋ ธ๋ ์์ฅ์ ๋๋นํด์ผ ํ ๊ฒ์ด๋ค.
์ธ๊ณต์ง๋ฅ(AI)์ ๋จ์ํ ๊ธฐ์ ํ์ ์ ๋์ด, ์ฐ๋ฆฌ ์ฌํ์ ๊ฒฝ์ ์ ๊ทผ๊ฐ์ ๋คํ๋ค๊ณ ์์ต๋๋ค. ๊ณผ๊ฑฐ์๋ ๊ณต์ฅ ๋ก๋ด์ด๋ ํ๊ณ ์ํํธ์จ์ด์ ์๋ํ๋ฅผ ๋จผ ๋ฏธ๋์ ์ผ๋ก ์ฌ๊ฒผ์ง๋ง, ์ฑGPT์ ๊ฐ์ ๋๊ท๋ชจ ์ธ์ด ๋ชจ๋ธ(LLM)์ ๋ฑ์ฅ ์ดํ, AI๋ ์ด๋ฏธ ์ฐ๋ฆฌ ์ถ ์์ ๊น์์ด ์นจํฌํ๋ฉฐ โ์ง๊ธ, ๊ทธ๋ฆฌ๊ณ ์ผ๋ง๋ ๋น ๋ฅด๊ฒโ ๋ณํ๋ฅผ ๊ฐ์ ธ์ฌ ๊ฒ์ธ๊ฐ์ ๋ํ ์ง๋ฌธ์ ๋์ง๊ณ ์์ต๋๋ค. ์ด ๊ธ์ AI๊ฐ ๋ ธ๋ ์์ฅ๊ณผ ๊ธ๋ก๋ฒ ๊ฒฝ์ ์ ๋ฏธ์น๋ ์ํฅ์ ์ฌ์ธต์ ์ผ๋ก ๋ถ์ํ๊ณ , ์ฐ๋ฆฌ๊ฐ ๋์๊ฐ์ผ ํ ๋ฐฉํฅ์ ๋ชจ์ํ๊ณ ์ ํฉ๋๋ค.
๊ฒฝ์ ์ฑ๋ โEconomics Explainedโ๋ ์ง๋ 7๋ ๊ฐ AI์ ์๋ํ์ ๋ํ ํต์ฐฐ์ ๊พธ์คํ ์ ๊ณตํด ์์ต๋๋ค. 2019๋ ์์ ์๋ํ๋ ์ธ์์ ๋ชจ์ต์ ์์ํ๋ ์ฌ๊ณ ์คํ๋ถํฐ, 2025๋ AI๊ฐ ์ค์ ๊ฒฝ์ ์ ๋ฏธ์น๋ ์ํฅ, ๊ทธ๋ฆฌ๊ณ ์ต๊ทผ MIT๊ฐ ๋ฐํํ โ๋น์ฐ ์ง์(Iceberg Index)โ ์ฐ๊ตฌ๊น์ง, ์๊ฐ์ ํ๋ฆ์ ๋ฐ๋ผ AI์ ๋ํ ์ฐ๋ฆฌ์ ์ดํด๊ฐ ์ด๋ป๊ฒ ์งํํด ์๋์ง ์ดํด๋ณด๊ณ , ์์ธก ์ค ์ด๋ค ๊ฒ์ด ํ์ค์ด ๋์๊ณ ์ด๋ค ๊ฒ์ด ๋น๋๊ฐ๋์ง, ๊ทธ๋ฆฌ๊ณ ํ์ฌ์ ๋ ธ๋์์ ๊ฒฝ์ ๋ ์ด๋์ ์ ์๋์ง๋ฅผ ์ข ํฉ์ ์ผ๋ก ๋ค๋ฃน๋๋ค.
1๋ถ: ์์ ์๋ํ ์๋์ ์ธ ๊ฐ์ง ๋ฏธ๋ ์๋๋ฆฌ์ค (2019๋ ์ ์์ธก)
์๋ํ๋ ์ญ์ฌ ์์์ ๋์์์ด ๋ฐ๋ณต๋์ด ์จ ํ๋ฆ์ ๋๋ค. ์ฒญ๋๊ธฐ๊ฐ ๊ฐ์ฒ ์ ๋ฐ๋ ค๋๊ณ , ์ฆ๊ธฐ๊ธฐ๊ด์ด ๋ด์ฐ๊ธฐ๊ด์ ์ํด ๋์ฒด๋์๋ฏ, ์ค๋๋ ๋งํธ์ ์ ํ ๊ณ์ฐ๋๋ ์๋ง์ ๊ณ์ฐ์๋ค์ ์ผ์๋ฆฌ๋ฅผ ์์ ๊ณ ์์ต๋๋ค. ๊ธฐ๊ณ์ ์๋ํ๋ ๊ฒฐ๊ตญ ์ฐ๋ฆฌ ๋ชจ๋์ ์ผ์๋ฆฌ๋ฅผ ๋์ฒดํ ๊ฒ์ด๋ผ๋ ์์ธก์ ๊ฑฐ์ค๋ฅผ ์ ์๋ ํ์ค๋ก ๋ค๊ฐ์ค๊ณ ์์ต๋๋ค.
๋ ธ๋ ์์ฅ์ ๊ณต๊ธ๊ณผ ์์ ์๋ฆฌ
๊ฒฝ์ ํ์๋ค์ ์ฌํ์ ๊ฐ๊ฒฉ์ด ๊ณต๊ธ๊ณผ ์์์ ์ํด ๊ฒฐ์ ๋๋ฏ, ๋ ธ๋ ์์ฅ์์๋ ์๊ธ์ด ๋ ธ๋๋ ฅ์ ๊ณต๊ธ๊ณผ ์์์ ์ํด ๊ฒฐ์ ๋๋ค๊ณ ์ค๋ช ํฉ๋๋ค. ์ฌ๊ณผ๋ฅผ ์๋ก ๋ค์ด๋ด ์๋ค. ์ฌ๊ณผ ๊ฐ๊ฒฉ์ด 2์ผํธ๋ผ๋ฉด ๋ชจ๋๊ฐ ์ฌ๊ณผ๋ฅผ ์ํ๊ฒ ์ง๋ง, ์์ฐ์๋ ์์งํ์ฐ์ด ๋ง์ง ์์ ์ฌ๊ณผ๋ฅผ ํค์ฐ์ง ์์ ๊ฒ์ ๋๋ค. ๋ฐ๋๋ก ์ฌ๊ณผ ๊ฐ๊ฒฉ์ด 10๋ฌ๋ฌ๋ผ๋ฉด ๋ชจ๋๊ฐ ์ฌ๊ณผ๋ฅผ ํค์ฐ๋ ค ํ๊ฒ ์ง๋ง, ๊ตฌ๋งค์๋ ์ธ๋ฉดํ ๊ฒ์ ๋๋ค. ๊ฒฐ๊ตญ ๊ณต๊ธ๊ณผ ์์๊ฐ ๋ง๋๋ ์ง์ ์์ ํฉ๋ฆฌ์ ์ธ ๊ฐ๊ฒฉ๊ณผ ๊ฑฐ๋๋์ด ํ์ฑ๋ฉ๋๋ค.
์ด ์๋ฆฌ๋ ์ง์ ์๋ ๋์ผํ๊ฒ ์ ์ฉ๋ฉ๋๋ค. ํ๊ณ์ฌ๋ฅผ ์๋ก ๋ค๋ฉด, ์ฐ๋ด์ด 10๋ฌ๋ฌ๋ผ๋ฉด ๋ชจ๋ ๊ธฐ์ ์ด ํ๊ณ์ฌ๋ฅผ ๊ณ ์ฉํ๋ ค ํ ๊ฒ์ด๊ณ , ์ฐ๋ด์ด 100๋ง ๋ฌ๋ฌ๋ผ๋ฉด ๋ชจ๋ ์ฌ๋์ด ํ๊ณ์ฌ๊ฐ ๋๋ ค ํ ๊ฒ์ ๋๋ค. ํ์ง๋ง ํ์ค์์๋ ๊ธฐ์ ์ด ํ์ํ ๋งํผ์ ํ๊ณ์ฌ๋ฅผ ๊ณ ์ฉํ๊ณ , ํ๊ณ์ฌ๋ค์ ์์ฅ์์ ํ์ฑ๋ ์๊ธ์ ๋ฐ๊ฒ ๋ฉ๋๋ค. ํ์ฌ ๋ฏธ๊ตญ์์๋ ์ฝ 126๋ง ๋ช ์ ํ๊ณ์ฌ๊ฐ ํ๊ท 6๋ง ๋ฌ๋ฌ์ ์ฐ๋ด์ ๋ฐ๊ณ ์ผํ๊ณ ์์ต๋๋ค.
๋ ธ๋๋ ฅ ๊ณต๊ธ์ ๋ณํ: ์์์์ฑ์ ์ํฅ
ํ๋ฆฌํ์ ์ฌ๋ฌด์ค์ ์ด์ด ํ๊ณ ์ ๋ฌด๋ฅผ ํด์ธ๋ก ์์์์ฑ(Outsourcing)ํ๋ ๊ฒฝ์ฐ๋ฅผ ์๊ฐํด ๋ด ์๋ค. ์ด๋ ํน์ ์๊ธ ์์ค์์ ์ผํ๋ ค๋ ํ๊ณ์ฌ์ โ๊ณต๊ธโ์ ํจ๊ณผ์ ์ผ๋ก ์ฆ๊ฐ์ํค๋ ๊ฒฐ๊ณผ๋ฅผ ๋ณ์ต๋๋ค. ๊ธฐ์ ๋ค์ ๋ ์ ์ ๋น์ฉ์ผ๋ก ๋์ผํ ์์ ํ๊ณ์ฌ๋ฅผ ๊ณ ์ฉํ ์ ์๊ฒ ๋๋ฉฐ, ์ด๋ ์๊ธ ์ ์ฒด์ ์ฃผ์ ์์ธ์ด ๋์ด ๋ง์ ์ ์ง๊ตญ ๊ฒฝ์ ์ ์ํฅ์ ๋ฏธ์น๊ณ ์์ต๋๋ค.
๋ ธ๋๋ ฅ ์์์ ๋ณํ: ๊ธฐ์ ๋ฐ์ ์ ์ญํ
๊ธฐ์ ๋ฐ์ ์ ๋ ธ๋๋ ฅ โ์์โ ์ธก๋ฉด์์๋ ๋ณํ๋ฅผ ๊ฐ์ ธ์ต๋๋ค. ๊ณผ๊ฑฐ 20๋ช ์ ํ๊ณ์ฌ๊ฐ ์ฃผํ์ผ๋ก ํ๋ ์ผ์, 5๋ช ์ ํ๊ณ์ฌ๊ฐ ์ ์๊ณ์ฐ๊ธฐ๋ก ์ฒ๋ฆฌํ๊ณ , ์ด์ ์์ ๊ณผ ํ๊ณ ์ํํธ์จ์ด๋ฅผ ์ฌ์ฉํ๋ 2๋ช ์ ํ๊ณ์ฌ๊ฐ 5๋ช ์ ๊ณ์ฐ์๋งํผ์ ์ผ์ ํด๋ผ ์ ์์ต๋๋ค. ์๋ณธ ์์ฐ์ธ ๊ธฐ์ ์ด ๋ฐ์ ํ ์๋ก, ๋์ผํ ์์ ์์ ์ ๋ ์ ์ ์ธ์์ผ๋ก ์ํํ ์ ์๊ฒ ๋ฉ๋๋ค. ์ด๋ 200๋ช ์ ํ๊ณ์ฌ๋ฅผ ๊ณ ์ฉํ๋ ๊ธฐ์ ์ด ์ด์ 20๋ช ๋ง ํ์ํ๊ฒ ๋จ์ ์๋ฏธํ๋ฉฐ, ์๋ จ๋ ํ๊ณ์ฌ๋ค๋ง์ ์ต์ ์๊ธ์ ๋ฐ๊ธฐ ์ํด ๊ฒฝ์ํด์ผ ํ๋ ์ํฉ์ ๋์ด๊ฒ ๋ฉ๋๋ค. ๊ถ๊ทน์ ์ผ๋ก๋ ์ด ๋ชจ๋ ํ๊ณ ์ ๋ฌด๋ฅผ ์ธ๊ฐ์ ๊ฐ์ ์์ด ์ฒ๋ฆฌํ ์ ์๋ ๊ณ ๋๋ก ์๋ํ๋ ๊ธฐ๊ณ๊ฐ ๋ฑ์ฅํ ๊ฒ์ ๋๋ค.
์๋ํ๋ ์ธ์์ ์ธ ๊ฐ์ง ๋ชจ์ต
๋ฏธ๋๋ฅผ ์์ธกํ๋ ๊ฒ์ ๋ถ๊ฐ๋ฅ์ ๊ฐ๊น์ง๋ง, ๊ฒฝ์ ํ์๋ค์ ๊ธฐ๊ณ๊ฐ ๋๋ถ๋ถ์ ์ผ์ ์ํํ๋ ์์ ์๋ํ๋ ์ธ์์ ๋ชจ์ต์ ์ธ ๊ฐ์ง ์๋๋ฆฌ์ค๋ก ์ ์ํฉ๋๋ค. โ์ข์ ๋ฏธ๋(The Good)โ, โ์์ธํ ๋ฏธ๋(The Bad)โ, ๊ทธ๋ฆฌ๊ณ โ์ต์ ์ ๋ฏธ๋(The Ugly)โ์ ๋๋ค.
1.1. ๋๊ด์ ๋ฏธ๋: ํ์์ ์ฌ๊ฐ (The Good)
์ด ์๋๋ฆฌ์ค๋ ๋ก๋ด ์ง์ฌ์ ์๋น์ค๋ฅผ ๋ฐ๊ณ , ์๋ํ ๋๋ถ์ ์ธ๊ฐ์ ์ฌ๊ฐ์ ์ฐฝ์์ ์ธ ํ๋์ ์๊ฐ์ ํ ์ ํ๋ โ๋ฏธ๋ํ์์ ๊ฟโ๊ณผ ๊ฐ์ต๋๋ค. ๊ธฐ๊ณ๋ฅผ ์ด์ํ๋ ๊ธฐ์ ์ ๋ฌด๊ฑฐ์ด ์ธ๊ธ์ด ๋ถ๊ณผ๋๊ฑฐ๋, ๊ธฐ๊ณ ์์ฒด๊ฐ ์ ๋ถ ์์ ๊ฐ ๋์ด ๋ณดํธ์ ๊ธฐ๋ณธ ์๋(Universal Basic Income, UBI)์ด ๋ชจ๋ ์ฌ๋์๊ฒ ์ง๊ธ๋ฉ๋๋ค. ์ฌ๋๋ค์ ์ถ๊ฐ ์์ ์ ์ํด ์ฌ์ ์ ํ๊ฑฐ๋, ๊ธฐ๊ณ๊ฐ ๋์ฒดํ ์ ์๋ ์์์ ์ฐ์ ์์ ์ผํ ์ ์์ต๋๋ค. ๋๋ถ๋ถ์ ์ฌ๋๋ค์ ์ค๋๋ ์ ์ค์ฐ์ธต๋ณด๋ค ํจ์ฌ ๋์ ์ํ ์์ค์ ๋๋ฆฌ๊ฒ ๋ฉ๋๋ค.
ํ์ง๋ง ์ด ๋๊ด์ ์ธ ์๋๋ฆฌ์ค์๋ ์ ์ฌ์ ์ธ ๋ฌธ์ ๊ฐ ์์ต๋๋ค. ๋ชจ๋ ์ฌ๋์ด ๊ณ ์ฉ์ ์๋ฐ์์ ๋ฒ์ด๋๋ฉด ์ถ์ฐ์จ์ด ๊ธ์ฆํ ๊ฐ๋ฅ์ฑ์ด ๋์ต๋๋ค. ์๋ ์์ก์ ๋ถ๋ด์ด ์ค์ด๋ค๊ธฐ ๋๋ฌธ์ ๋๋ค. ๋ฌธ์ ๋ ์ธ๊ฐ์ด ์์ ์๋ชจ์ ์ฃผ์ฒด๋ผ๋ ์ ์ ๋๋ค. ํ์ฌ๋ ์์ฐ ํ๋์ ํตํด ์ฌํ์ ๊ธฐ์ฌํ์ง๋ง, ์๋ํ๋ ์ธ์์์๋ ๊ธฐ๊ณ๊ฐ ๋ชจ๋ ๊ฒ์ ์ฒ๋ฆฌํ๋ฏ๋ก ์ธ๊ฐ์ ๊ฒฝ์ ์ ๊ฐ์น๋ฅผ ์ฐฝ์ถํ์ง ๋ชปํ ์ ์์ต๋๋ค. ์ด๋ ์ ํํ ์์์ ๋ํ ์๋ฐ์ ๊ฐ์ค์ํค๊ณ , ์ฅ๊ธฐ์ ์ผ๋ก๋ ์ธ๊ฐ์ด ์ฌํ์ โ๋ถ์ ์ ์ธ ๊ฐ์นโ๋ฅผ ๊ฐ์ง๊ฒ ๋ ์ ์๋ค๋ ์ฐ๋ ค๋ฅผ ๋ณ์ต๋๋ค.
1.2. ์์ธํ ๋ฏธ๋: ์๊ทนํ์ ๊ฐ๋ฑ (The Bad)
์ด ์๋๋ฆฌ์ค์์๋ ๋ณดํธ์ ๊ธฐ๋ณธ ์๋(UBI)์ด ์ฌ์ ํ ์กด์ฌํ์ง๋ง, ๊ฒจ์ฐ ์๊ณ๋ฅผ ์ ์งํ ์ ์๋ ์ต์ํ์ ์์ค์ ๋จธ๋ญ ๋๋ค. ์ธ์์ ๋ก๋ด์ ์์ ํ ๊ธฐ์ ์ ์์ ์ฃผ์ธ โ๋ถ์ ์ธตโ๊ณผ ๊ทธ ์ธ์ โ๋์คโ์ด๋ผ๋ ๋ ๊ณ์ธต์ผ๋ก ๊ทน๋ช ํ๊ฒ ๋๋ฉ๋๋ค. ๋์ค์ ์๋ํ๋ก ๋์ฒดํ ์ ์๋ ํ์ ์์ฅ์์ โ๊ธฑ(gig) ๊ฒฝ์ โ ํํ์ ์ผ์๋ฆฌ๋ฅผ ์ฐพ์์ผ ํฉ๋๋ค. ์๋ ์ถ์ฐ์ ๊ฐ๋ ฅํ ๋น๊ถ์ฅ๋๋ฉฐ, UBI๋ ์ฑ์ธ์ด ๋ ํ์์ผ ์ง๊ธ๋์ด ์๋ ์์ก์ด ์์ฒญ๋ ์ฌ์ ์ ๋ถ๋ด์ด ๋๊ฑฐ๋ ๋ถ์ ์ธต์ ์ ์ ๋ฌผ์ด ๋ ์ ์์ต๋๋ค.
๋ถ์ ์ธต ์ญ์ ์ํฉ์ด ๊ทธ๋ฆฌ ์ข์ง๋ง์ ์์ต๋๋ค. ๋ง๋ํ ๋ถ๋ฅผ ๋๋ฆฌ๊ฒ ์ง๋ง, ์ฌํ ์ ๋ฐ์ ํผ์ง ํญ๋ ฅ๊ณผ ์ฆ์ค ์์์ โ์์โ์ ๊ฐ์ ์ถ์ ์ด์์ผ ํ ๊ฒ์ ๋๋ค. ๋ถ์ ์ธต๊ณผ ๋น๊ณค์ธต์ ๊ฒฉ์ฐจ๊ฐ ๊ทน์ฌํ ๋จ์ํ๋ฆฌ์นด ๊ณตํ๊ตญ์ ์ํ๋ค์ค๋ฒ๊ทธ์ ๊ฐ์ ๋์์ ํ์ค์ด ์ ์ธ๊ณ๋ก ํ์ฐ๋ ์ ์์ต๋๋ค.
1.3. ์ต์ ์ ๋ฏธ๋: ๋ฌด๊ฐ์นํ ์ธ๊ฐ (The Ugly)
์ด ์๋๋ฆฌ์ค๋ โ์ผํด์ผ๋ง ์ด ์ ์๋คโ๋ ์ฌ๊ณ ๋ฐฉ์์ด ์ง๋ฐฐ์ ์ธ ์ผ๋ถ ๊ตญ๊ฐ์์ ํ์ค์ด ๋ ์ ์์ต๋๋ค. ๋ณดํธ์ ๊ธฐ๋ณธ ์๋(UBI)์ ๊ฑฐ์ ์กด์ฌํ์ง ์์ผ๋ฉฐ, ์ฌํ ๋ณต์ง ์์คํ ๋ง์ ๊ฒฝ์ ์ ์๋ ฅ์ผ๋ก ์ธํด ๊ณ ์ฌํ ๊ฒ์ ๋๋ค. ๊ณ ์ฉ๋ ์ ์๋ ์ธ๊ฐ์ โ๊ฒฝ์ ์ ๊ฐ์นโ๊ฐ ์๋ค๋ ๋ํนํ ์ง์ค์ด ๋๋ฌ๋ฉ๋๋ค. ๋๋ถ๋ถ์ ์ฌ๋๋ค์ด ์๊ฐ๊ณผ ๋์ ๊ตํํ๋ฉฐ ์ด์๊ฐ๋ ์ค๋๋ ์ ๊ฒฝ์ ์์คํ ์์, ๊ทธ๋ค์ ์๊ฐ์ด ๊ฐ์น ์์ด์ง๋ค๋ฉด ๋์ ๋ฐ์ ์ด์ ๊ฐ ์ฌ๋ผ์ง๊ธฐ ๋๋ฌธ์ ๋๋ค.
์ผ๋ถ์์๋ ๊ธฐ์ ์ด ๊ณ ๊ฐ์ ์์ผ๋ฉด ๋งํ ๊ฒ์ด๋ผ๊ณ ์ฃผ์ฅํ์ง๋ง, ์ด๋ ์์งํ ์๊ฐ์ผ ์ ์์ต๋๋ค. ๊ธฐ์ ์ ๋ ์ด์ ์ธ๊ฐ์๊ฒ ์ฃผํ, ์๋, ์๋ฅ๋ฅผ ์ ๊ณตํ ํ์๊ฐ ์์ต๋๋ค. ๋์ ๊ธฐ์ ๋ค์ ์๋ก ๊ฑฐ๋ํ๋ฉฐ ๋ถ์ ํ ์์ ์ฃผ๋ค์ ์์๋ฅผ ์ถฉ์กฑ์ํฌ ๊ฒ์ ๋๋ค. ์ผ๋ฐ ์๋น์๋ ๊ฒฝ์ ์์คํ ์์ ์๊ฐ๋ณด๋ค ์ค์ํ์ง ์์ ์ ์์ต๋๋ค. ์ด ์๋๋ฆฌ์ค์์๋ ์ธ๊ตฌ ๊ฐ์๊ฐ ํ์ฐ์ ์ ๋๋ค. ์๋ ๋ฅผ ์์กํ ์ ์์ด ์ถ์ฐ์จ์ด ์ค์ด๋ค๊ฑฐ๋, ๋ ๋์๊ฐ ๋ง์ ์ฌ๋๋ค์ด ๊ตถ์ด ์ฃฝ๋ ๋น๊ทน์ ์ธ ์ํฉ์ด ๋ฐ์ํ ์ ์์ต๋๋ค. ๊ธฐ์ ๋ฐ์ ์ด ์ธ๊ฐ์ ๊ฐ์น ์ฐฝ์กฐ์๊ฐ ์๋ โ์์ฌโ๋ก ๋ง๋ค ๋ ๋ฒ์ด์ง ์ ์๋ ๋์ฐํ ๋ฏธ๋์ ๋๋ค.
2๋ถ: AI, ์์์น ๋ชปํ ๊ณณ์ ๋จผ์ ๊ฐํํ๋ค (2025๋ ์ ํต์ฐฐ)
2019๋ ์ ์์ธก์ AI๊ฐ ๊ณต์ฅ ๋ ธ๋์๋ ๋ฐ๋ณต์ ์ธ ์ก์ฒด๋ ธ๋์ ๋จผ์ ๋์ฒดํ ๊ฒ์ด๋ผ๊ณ ๋ณด์์ง๋ง, ํ์ค์ ์ ๋ฐ๋์์ต๋๋ค. AI๊ฐ ์ฒ์ ๋ฑ์ฅํ์ ๋, ๊ทธ๊ฒ์ ํ๋ฆฌํ์ ์ฝ์ผํฐ, ๋ฐฉ๊ธ๋ผ๋ฐ์์ ๋ฐ์ดํฐ ์ ๋ ฅ ๋ ธ๋์, ๊ทธ๋ฆฌ๊ณ ๊ฐ๋ฐ๋์๊ตญ์ด 30๋ ๊ฐ ๊ณต๋ค์ฌ ๊ตฌ์ถํด ์จ ์์์์ฑ ์๋น์ค ๊ฒฝ์ ๋ฅผ ๊ฐํํ์ต๋๋ค. ๊ณต์ฅ ๋ ธ๋์๋ค์ ๊ด์ฐฎ์์ง๋ง, ์ฌ๋ฌด์ค์์ ์ผํ๋ ์ฌ๋๋ค์ด ์คํ๋ ค ๊ฑฑ์ ํด์ผ ํ ์ฒ์ง๊ฐ ๋ ๊ฒ์ ๋๋ค.
์์๊ณผ ๋ค๋ฅธ AI์ ์ํฅ๋ ฅ
โ๊ธฐ๊ณ๊ฐ ์ฐ๋ฆฌ๋ฅผ ๋์ฒดํ๋ค๋ฉด ์ด๋ป๊ฒ ๋ ๊น?โ๋ผ๋ ์ง๋ฌธ์ โ๊ธฐ๊ณ๊ฐ ์ด๋ฏธ ์ฐ๋ฆฌ๋ฅผ ๋์ฒดํ๊ณ ์๋๋ฐ, ์ด์ ์ฐ๋ฆฌ๋ ๋ฌด์์ ํด์ผ ํ ๊น?โ๋ผ๋ ์ง๋ฌธ์ผ๋ก ๋ฐ๋์์ต๋๋ค. ์ฌ๋๋ค์ AI๊ฐ ์ค์นด์ด๋ท(Skynet)๊ณผ ๊ฐ์ ๋์คํ ํผ์์ ๋ฏธ๋์ ์์ ํ์๋ก์ด ์ ํ ํผ์ ์ฌ์ด ์ด๋๊ฐ์ ์์ ๊ฒ์ด๋ผ๋ ๋ถ์๊ฐ์ ๊ฐ์ง๊ณ ์์ต๋๋ค. ๊ธฐ์กด์ ๊ธฐ์ ํ์ ์ ๊ฒฝ์ ๋ฅผ ๋ ๋ถ์ ํ๊ฒ ๋ง๋ค๊ณ , ๋์ฒด๋ ์ผ์๋ฆฌ๋ณด๋ค ๋ ์ข์ ์ผ์๋ฆฌ๋ฅผ ์ฐฝ์ถํ๋ค๊ณ ์ฃผ์ฅํ๋ ๋๊ด๋ก ๋ ์์ง๋ง, ๊ธฐ๊ณ๊ฐ ์ก์ฒด๋ ธ๋์ ๋์ฒดํ์ ๋๋ ์ธ๊ฐ์ ์ง๋ ฅ์ ํ์ฉํ ์ ์์์ง๋ง, AI๊ฐ ์ง๋ ฅ๋ง์ ๋์ฒดํ๋ค๋ฉด ์ธ๊ฐ์๊ฒ ๋ฌด์์ด ๋จ์ ๊ฒ์ธ๊ฐ๋ผ๋ ๋น๊ด๋ก ๋ ์กด์ฌํฉ๋๋ค.
๊ฐ๋ฐ๋์๊ตญ ์๋น์ค ์ฐ์ ์ ์๊ธฐ
๋ฏธ๋๋ฅผ ์์ธกํ ํ์๋ ์์ด, ์ด๋ฏธ ์ผ๋ถ ๊ฒฝ์ ์์๋ ์ด๋ฌํ ๋ณํ๊ฐ ๊ด๋ฒ์ํ๊ฒ ๋ํ๋๊ณ ์์ต๋๋ค. ํ๋ฆฌํ๊ณผ ๋ฐฉ๊ธ๋ผ๋ฐ์์ ๊ฐ์ ๊ตญ๊ฐ์์๋ AI์ ์ํ์ด ํจ์ฌ ๋ ์ ๋ฐํฉ๋๋ค. ์ด๋ค ๊ตญ๊ฐ๋ค์ ์ง๋ 30๋ ๊ฐ ์ฝ์ผํฐ, ๋ฐ์ดํฐ ์ ๋ ฅ, ์ ์ฌ(่ฝๅฏซ) ์์ , ๊ธฐ๋ณธ์ ์ธ ์ํํธ์จ์ด ์ง์๊ณผ ๊ฐ์ ์์์์ฑ ์๋น์ค ์ฐ์ ์ ์ค์ฌ์ผ๋ก ๊ฒฝ์ ์ฑ์ฅ์ ์ด๋ฃจ์ด์์ต๋๋ค. ์ด๋ฌํ ์ผ์๋ฆฌ๋ ์ธ์ด ๋ฅ๋ ฅ, ๋งฅ๋ฝ ์ดํด, ์ธ๊ฐ์ ์ธ ์๊ธธ์ ์๊ตฌํ๊ธฐ ๋๋ฌธ์ ์๋ํ๋ก๋ถํฐ ์์ ํ๋ค๊ณ ์ฌ๊ฒจ์ก์ต๋๋ค.
๊ทธ๋ฌ๋ ๋๊ท๋ชจ ์ธ์ด ๋ชจ๋ธ(LLM)๊ณผ ๊ฐ์ AI ๋๊ตฌ๋ ์ด์ ์ด๋ฌํ ์์ ์ ๋ช ์ด ๋ง์, ํจ์ฌ ์ ๋ ดํ ๋น์ฉ์ผ๋ก ์ฒ๋ฆฌํ ์ ์๊ฒ ๋์์ต๋๋ค. ๊ตญ์ ํตํ๊ธฐ๊ธ(IMF)์ ํ๋ฆฌํ์ ์์์์ฑ ์๋น์ค ์ผ์๋ฆฌ ์ค 89%๊ฐ AI์ ์ํด ์๋ํ๋ ์ํ์ด ๋๋ค๊ณ ์ถ์ ํฉ๋๋ค. ์ด๋ 100๋ง ๋ช ์ด์์ ์ผ์๋ฆฌ๊ฐ ๋ช ๋ ์์ ์ฌ๋ผ์ง ์ ์์์ ์๋ฏธํฉ๋๋ค. AI๋ ์ด๋ฏธ ์ธ๊ณ์ ๋ถ์ ํ ๊ตญ๊ฐ๋ค์ ๋์ฑ ๋ถ์ ํ๊ฒ ๋ง๋ค๊ณ ์์ผ๋ฉฐ, ๋ค๋ฅธ ๊ตญ๊ฐ๋ค์ด ๋ฐ๋ผ์ก๊ธฐ ์ด๋ ต๊ฒ ๋ง๋ค๊ณ ์์ต๋๋ค.
์ ์ง๊ตญ ๋ด์์์ ์ง์ ๋ณํ
์ ์ง๊ตญ์์๋ AI์ ์ํฅ์ ๋ถ๋ช ํฉ๋๋ค. ๋ฏธ๊ตญ ๋ ธ๋ํต๊ณ๊ตญ(US Bureau of Labor Statistics)์ ๊ณ์ฐ์, ์ํ ์ฐฝ๊ตฌ ์ง์, ์ฐ์ฒด๊ตญ ์ง์, ๊ณ ๊ฐ ์๋น์ค ๋ด๋น์์ ๊ฐ์ ์ง๋ฌด๊ฐ ๊ฐ์ํ ๊ฒ์ด๋ผ๊ณ ์์ธกํฉ๋๋ค. ํ ์ถ์ ์น์ ๋ฐ๋ฅด๋ฉด, ํฅํ 5๋ ๋ด์ 710๋ง ๊ฐ์ ์ผ์๋ฆฌ๊ฐ ์ฌ๋ผ์ง ์ ์์ผ๋ฉฐ, ํ์ฌ ์ง๋ฌด์ ์ต๋ 47%๊ฐ AI๋ก ๋์ฒด๋ ์ํ์ด ์๋ค๊ณ ํฉ๋๋ค. ๊ธฐ์ ๊ณผ ํฌ์์๋ค์ด AI ๊ธฐ์ ๊ฐ๋ฐ์ ์์กฐ ๋ฌ๋ฌ๋ฅผ ์์๋ถ์ ๋งํผ, ์ด๋ค์ ๊ฒฐ๊ตญ ํฌ์ ์์ต์ ์ํ๋ฉฐ, ์๋ฐฑ๋ง ๋ช ์ ์ธ๋ ฅ์ ๊ฐ์ถํ๋ ๊ฒ์ด ๊ฐ์ฅ ์ฆ๊ฐ์ ์ธ ์์ต ์ฐฝ์ถ ๋ฐฉ๋ฒ์ผ ์ ์์ต๋๋ค.
AI๊ฐ ๋ถ์ ๊ฒฉ์ฐจ๋ฅผ ์ฌํ์ํค๋ ์ด์
AI๋ ๋ถ์ ํ ๊ตญ๊ฐ์์ ์ฑ์ฅ์ ๊ฐ์ํํ๋ ๋์์, ๋ค๋ฅธ ๊ตญ๊ฐ๋ค์ ๊ฒฝ์ ์ ์์กด์ ์ํํ๊ณ ์์ต๋๋ค. ๊ฒฝ์ ์ ์ฑ ์ฐ๊ตฌ์ผํฐ(Center for Economic Policy Research)์ ๋ฐ๋ฅด๋ฉด, AI๋ก ์ธํ ์์ฐ์ฑ ํฅ์ ๋๋ถ์ ๋ฏธ๊ตญ์ ํฅํ 10๋ ๊ฐ GDP๊ฐ 5.4% ์ฆ๊ฐํ ์ ์์ผ๋ฉฐ, ์๊ตญ, ๋ ์ผ, ํ๊ตญ๋ ๋น์ทํ ์ฑ์ฅ์ธ๋ฅผ ๋ณด์ผ ๊ฒ์ผ๋ก ์์ธก๋ฉ๋๋ค. ๋ฐ๋ฉด ์ ์๋ ๊ตญ๊ฐ๋ค์ 2.7~3.5%์ ๋ ์๋งํ ์ฑ์ฅ๋ฅ ์ ๊ธฐ๋กํ ๊ฒ์ผ๋ก ์์๋์ด, ๋ถ์ ํ ๊ตญ๊ฐ์ ๋น๊ณคํ ๊ตญ๊ฐ ๊ฐ์ ๊ฒฉ์ฐจ๊ฐ ๋์ฑ ๋ฒ์ด์ง ๊ฒ์ ๋๋ค.
ํ๋ฆฌํ์ ๋น์ฆ๋์ค ํ๋ก์ธ์ค ์์์์ฑ(Business Process Outsourcing, BPO) ์ฐ์ ์ 370์ต ๋ฌ๋ฌ ๊ท๋ชจ๋ก 130๋ง ๋ช ์ด์์ ๊ณ ์ฉํ๋ฉฐ GDP์ 7% ์ด์์ ์ฐจ์งํฉ๋๋ค. ํ์ง๋ง ์ด๋ค ์ผ์๋ฆฌ์ ๋๋ถ๋ถ์ ์ฑGPT์ ๊ฐ์ LLM์ด ๋น ๋ฅด๊ฒ ์๋ํํ ์ ์๋ ๋ฐ๋ณต์ ์ธ ํ ์คํธ ๊ธฐ๋ฐ ์์ ์ ๋๋ค. AI๊ฐ ๋ ๋น ๋ฅด๊ณ ์ ๋ ดํ๊ฒ, ๊ทธ๋ฆฌ๊ณ ๊ฑด๊ฐ ๋ณดํ์ด๋ ํด๊ฐ ์์ด ๋์ผํ ํ์ง์ ์์ ์ ์ํํ ์ ์๋ค๋ฉด, ํด์ธ ์์์์ฑ์ ๊ณ์ํ ๊ฒฝ์ ์ ์ด์ ๋ ์ฌ๋ผ์ง๋๋ค. ์ด๋ ๊ธฐ์ ๋ค์ด ์ผ์๋ฆฌ๋ฅผ ๋ค์ ๋ถ์ ํ ๊ตญ๊ฐ๋ก ๊ฐ์ ธ์ค๋ โ๋ฆฌ์ผ์ด๋ง(Reshoring)โ ํ์์ผ๋ก ์ด์ด์ง ์ ์์ต๋๋ค.
๋ณด์์ ์๋ณธ vs. ๋์ฒด์ ์๋ณธ
๊ฒฝ์ ํ์๋ค์ ์๋ก์ด ๊ธฐ์ ์ โ๋ณด์์ ์๋ณธ(Complementary Capital)โ๊ณผ โ๋์ฒด์ ์๋ณธ(Substitutive Capital)โ์ผ๋ก ๊ตฌ๋ถํฉ๋๋ค.
- ๋ณด์์ ์๋ณธ: ์ธ๊ฐ ๋ ธ๋์์ ์์ฐ์ฑ์ ๋์ด๋ ๊ธฐ์ ์ ๋๋ค. ์๋ฅผ ๋ค์ด, ์ฝค๋ฐ์ธ ์ํ๊ธฐ๋ ๋์ฅ ๋ ธ๋์๋ฅผ ๋์ฒดํ๋ ๊ฒ์ด ์๋๋ผ, ํ ๋ช ์ ๋ ธ๋์๊ฐ ํจ์ฌ ๋ ํจ์จ์ ์ผ๋ก ์์ ํ ์ ์๊ฒ ๋ง๋ญ๋๋ค. AI๋ ๊ณ ์๋ จ ์ง๋ฌด์์ ์ด๋ฌํ ๋ณด์์ ์๋ณธ ์ญํ ์ ํ ์ ์์ต๋๋ค. AI๋ฅผ ํ์ฉํ์ฌ ๋ณด๊ณ ์๋ฅผ ๋ถ์ํ๊ณ ์ด์ ์งํ๋ฅผ ๋ฐ๊ฒฌํ๋ ๊ธ์ต ๋ถ์๊ฐ๋ ๋ ๋น ๋ฅด๊ฒ ํต์ฐฐ๋ ฅ์ ์ป๊ณ ์ ๋ต์ ์ฌ๊ณ ์ ์ง์คํ ์ ์์ผ๋ฉฐ, AI ์ง๋จ ๋๊ตฌ๋ฅผ ์ฌ์ฉํ๋ ์์ฌ๋ ํ์ ์ง๋ฃ์ ๋ ๋ง์ ์๊ฐ์ ํ ์ ํ ์ ์์ต๋๋ค.
- ๋์ฒด์ ์๋ณธ: ์ธ๊ฐ ๋ ธ๋์ ์์ ํ ๋์ฒดํ๋ ๊ธฐ์ ์ ๋๋ค. AI ๊ธฐ๋ฐ ์ฑ๋ด์ ๊ณ ๊ฐ ์ง์ ์๋ด์์ ์ ๋ฌด ์๋๋ฅผ ๋์ด๋ ๊ฒ์ด ์๋๋ผ, ๊ทธ๋ค์ ๋์ฒดํฉ๋๋ค. ์ ๊ตํ ์ฝ๋ ์์ฑ๊ธฐ๋ ์ฃผ๋์ด ๊ฐ๋ฐ์๋ฅผ ๋๋ ๊ฒ์ด ์๋๋ผ, ๊ทธ๋ค์ ๋์ฒดํฉ๋๋ค. AI ๊ฒฝ์ ์์๋ ์๋ณธ ์์ ๊ฐ ํ๋ ์ญ์ฌ์ ๊ทธ ์ด๋ ๋๋ณด๋ค ์ง์ค๋์ด ์์ต๋๋ค.
AI ์๋ณธ ์์ ์ ์ง์คํ
AI ๋ถ์ผ์ ์ฃผ์ ๋ํ๊ตฌ๋ ๋ฏธ๊ตญ๊ณผ ์ค๊ตญ์ ์์ ์๋ฆฌํธ ๊ธฐ์ ์์ ๋์ค๊ณ ์์ต๋๋ค. 2017๋ ์ดํ ๋ฏธ๊ตญ์ 135๊ฐ์ ๋๊ท๋ชจ AI ์์คํ ์ ๊ฐ๋ฐํ์ผ๋ฉฐ, ์ค๊ตญ์ 110๊ฐ๋ก ๋ค๋ฅผ ์์ต๋๋ค. ํ์ง๋ง ๊ทธ ๋ค์๋ถํฐ๋ ์๊ตญ 25๊ฐ, ํ๋์ค 24๊ฐ๋ก ๊ฒฉ์ฐจ๊ฐ ํฌ๊ฒ ๋ฒ์ด์ง๋๋ค. ์ด๋ค ์ ๋ ๊ธฐ์ ๋ค์ โ๋ฐ์ดํฐ ๋คํธ์ํฌ ํจ๊ณผ(Data Network Effect)โ ๋๋ถ์ ๊ธฐํ๊ธ์์ ์ธ ์ฑ์ฅ์ ๊ฒฝํํ๊ณ ์์ต๋๋ค. ๋ ๋ง์ ๋ฐ์ดํฐ๋ฅผ ์์งํ ์๋ก AI ๋ชจ๋ธ์ ์ฑ๋ฅ์ด ํฅ์๋๊ณ , ๋ชจ๋ธ์ด ์ข์์ง์๋ก ๋ ๋ง์ ์ฌ์ฉ์๋ฅผ ์ ์นํ๋ฉฐ, ๋ ๋ง์ ์ฌ์ฉ์๋ ๋ ๋ง์ ๋ฐ์ดํฐ๋ฅผ ์์ฑํฉ๋๋ค. ์ด๋ ์์ฅ ์ง๋ฐฐ๋ ฅ๊ณผ ์ด์ต์ด ์์์ ๊ธฐ์ ์ ์ง์ค๋๋ ๊ฐ๋ ฅํ ํผ๋๋ฐฑ ๋ฃจํ๋ฅผ ๋ง๋ญ๋๋ค. PwC๋ AI๊ฐ 2030๋ ๊น์ง ์ ์ธ๊ณ GDP์ 15์กฐ 7์ฒ์ต ๋ฌ๋ฌ๋ฅผ ์ถ๊ฐํ ์ ์์ง๋ง, ์ด ๋ถ์ 70%๋ AI๋ฅผ ์์ ํ ๋ฏธ๊ตญ๊ณผ ์ค๊ตญ ๋จ ๋ ๊ตญ๊ฐ์ ๋์๊ฐ ๊ฒ์ผ๋ก ์์ธกํ์ต๋๋ค.
AI ํ๋์จ์ด ์์ฐ์ ์ง๋ฐฐ๋ ฅ
AI๋ฅผ ๊ตฌ๋ํ๋ ๋ฌผ๋ฆฌ์ ๊ธฐ๊ณ์ธ CPU์ GPU๋ ์๋์ ์ผ๋ก 5๊ฐ๊ตญ(๋ฏธ๊ตญ, ๋๋ง, ์ค๊ตญ, ํ๊ตญ, ์ผ๋ณธ)์์ ์ค๊ณ ๋ฐ ์ ์กฐ๋ฉ๋๋ค. ํ๋์จ์ด์ 90% ์ด์์ด ์ด๋ค ๊ตญ๊ฐ์์ ๋์ต๋๋ค. ์ด๋ ์์์ ๊ตญ๊ฐ๊ฐ AI ์์คํ ์ ์ด์ํ ๋ฟ๋ง ์๋๋ผ, AI๋ฅผ ๊ฐ๋ฅํ๊ฒ ํ๋ ๊ทผ๋ณธ์ ์ธ ๊ตฌ์ฑ ์์๋ฅผ ์์ฐํ๋ค๋ ๊ฒ์ ์๋ฏธํฉ๋๋ค. AI๋ ์์ฐ์ ์์ ํ ์ฌ๋๋ค์๊ฒ ์ฃผ๋ก ์ด์ต์ ์ฃผ๊ณ , ๊ทธ๋ ์ง ์์ ์ฌ๋๋ค์ ๋์ฒดํ๋ โ์๋ณธโ์ธ ๊ฒ์ ๋๋ค. AI๋ฅผ ์์ฐ์ฑ ์ฆ๋ ๋๊ตฌ๋ก ํ์ฉํ ์ ์๋ ๋ฅ๋ ฅ์ด ํด์๋ก, ์์ฅ์์ ๊ฒฝ์ ์ ๊ฐ์น๊ฐ ๋์์ง์ง๋ง, ๋ฐ๋ณต์ ์ธ ์ ๋ฌด๋ฅผ ์ํํ๋ ๋ ธ๋์, ํนํ ์ฌ๊ต์ก ํ๋ก๊ทธ๋จ์ ์ ๊ทผํ ์ ์๋ ์ด๋ค์๊ฒ๋ ๋ฏธ๋๊ฐ ํจ์ฌ ๋ ๋ถํ์คํด ๋ณด์ ๋๋ค.
3๋ถ: ๊ณผ๊ฑฐ์ ๊ตํ๊ณผ ๋ฏธ๋๋ฅผ ์ํ ํด๋ฒ
AI๋ก ์ธํ ๋ณํ๋ ๋ ์ด์ ๊ฐ์ค์ด ์๋ ํ์ค์ ๋๋ค. 2010๋ ๊น์ง ๋ฏธ๊ตญ์์ 700๋ง ๊ฐ ์ด์์ ๊ณต์ฅ ์ผ์๋ฆฌ๊ฐ ์ฌ๋ผ์ก๊ณ , ์ด๋ค ๋๋ถ๋ถ์ ๋์์ค์ง ์์์ต๋๋ค. ์ด๋ 1980๋ ๋์ 1990๋ ๋ ์ฐ์ ์๋ํ์ ๋๊ท๋ชจ ์์์์ฑ์ด ์ ์กฐ์ ์ ๊ฐํํ๋ ์๊ธฐ์ ์ ์ฌํฉ๋๋ค. ๋ํธ๋ก์ดํธ, ํด๋ฆฌ๋ธ๋๋์ ๊ฐ์ ๋์๋ค์ ํ๋ ์ฒ ๊ฐ, ์๋์ฐจ, ์ฌ์ ์ฐ์ ์ ๊ณ ์๊ธ ์ผ์๋ฆฌ๋ก ๋ฒ์ฑํ์ง๋ง, ๋ก๋ด ์ฉ์ ๊ณต, ์ปดํจํฐ ์ ์ด ์กฐ๋ฆฝ ๋ผ์ธ, ๊ทธ๋ฆฌ๊ณ ํด์ธ์ ์ ๋ ดํ ๋ ธ๋๋ ฅ์ ๋ฑ์ฅ์ผ๋ก ์ธํด ์์ ์ ์ธ ์ค์ฐ์ธต ์ผ์๋ฆฌ๊ฐ ์ฌ๋ผ์ง๊ณ ๋์ ๊ฒฝ์ ๊ฐ ๋ถ๊ดดํ์ต๋๋ค. ์ด๋ฌํ ์ง์ญ์์๋ ๊ธฐ๋ ์๋ช ์ด ๊ฐ์ํ๊ณ , ๋ง์ฝ ์ค๋ ์ด ์ฆ๊ฐํ๋ฉฐ, ํ๊ต๊ฐ ์ด๋ ค์์ ๊ฒช๋ ๋ฑ ๋จ์ํ ์ผ์๋ฆฌ ์์ค์ ๋์ด์ ์ฌํ์ ๋ฌธ์ ๋ค์ด ๋ฐ์ํ์ต๋๋ค. ๋ถํ๋ฑ์ด ์ผ๋จ ๊ฒฝ์ ์ ๋ฟ๋ฆฌ๋ด๋ฆฌ๋ฉด ๋๋๋ฆฌ๊ธฐ๊ฐ ๊ทน๋๋ก ์ด๋ ค์์ง๋ค๋ ๊ตํ์ ์ป์ ์ ์์ต๋๋ค.
๋์ ๋ฐฉ์: ์ธํ๋ผ์ ์ธ์ฌ ํฌ์
AI๊ฐ ๊ธฐ์กด์ ๋ถํ๋ฑ์ ์ฌํ์ํฌ์ง, ์๋๋ฉด ํด๊ฒฐํ๋ ๋ฐ ๋์์ด ๋ ์ง๋ ์์ผ๋ก ๊ตญ๊ฐ์ ๊ฐ์ธ์ด ์ด๋ค ์กฐ์น๋ฅผ ์ทจํ๋๋์ ๋ฌ๋ ค ์์ต๋๋ค. ๋คํํ ์ฐ๋ฆฌ๋ ์ด๋ฏธ ์ ์๋ ๊ตญ๊ฐ์์ AI์ ์ต์ ์ ํจ๊ณผ๋ฅผ ๋ชฉ๊ฒฉํ๊ณ ์์ต๋๋ค. ํ๋ฆฌํ ์ ๋ถ๋ 2028๋ ๊น์ง 100๋ง ๋ช ์ด์์ ๋ ธ๋์๋ฅผ ์ฌ๊ต์กํ๊ฒ ๋ค๋ ๋ชฉํ๋ก ๊ตญ๊ฐ AI ์ ๋ต์ ์์ํ์ต๋๋ค. ๋ฐฉ๊ธ๋ผ๋ฐ์๋ AI ์ธ์ฌ ์์ฑ, ๊ต์ก ์์คํ ํ๋ํ, ๊ธฐ์ ์คํํธ์ ์ง์์ ์ด์ ์ ๋ง์ถ ์ ์ฑ ํ๋ ์์ํฌ๋ฅผ ๋ฐํํ์ต๋๋ค. ์ด๋ฌํ ๋ ธ๋ ฅ๋ค์ด ์ถฉ๋ถํ ์ง๋ ๋ฏธ์ง์์ง๋ง, ์ ์ง๊ตญ๋ค์ด ๋ฐ๋ผ์ผ ํ ๋ช ํํ ๊ฒฝ๊ณ ์ด์ ์ง์นจ์ ์ ๊ณตํฉ๋๋ค.
๊ฒฝ์ ์ ๋ ๊ฐ์ง ์ธก๋ฉด, ์ฆ AI ์ธํ๋ผ์ ๋ํ ๋ง๋ํ ํฌ์์ ํจ๊ป โ์ฌ๋โ์ ๋ํ ํฌ์๋ฅผ ๋์์ ์งํํด์ผ ํฉ๋๋ค. ์ฌ๊ธฐ์๋ ์ปดํจํฐ ๊ณผํ ๊ต์ก๋ฟ๋ง ์๋๋ผ, AI๊ฐ ์๋ํํ๊ธฐ ์ด๋ ค์ด ์ข ๋ฅ์ ๊ธฐ์ , ์ฆ ๋นํ์ ์ฌ๊ณ , ๋ณต์กํ ๋ฌธ์ ํด๊ฒฐ, ํจ๊ณผ์ ์ธ ์์ฌ์ํต, ์ฐฝ์์ ์ธ ์์ฌ ๊ฒฐ์ ๊ณผ ๊ฐ์ โ์ธ๊ฐ ๊ณ ์ ์ ๊ธฐ์ โ์ ๋ํ ํฌ์๊ฐ ํฌํจ๋ฉ๋๋ค. ์ต๊ทผ ๋ฏธ๊ตญ์์ 1,200๋ง ๊ฐ์ ์ฑ์ฉ ๊ณต๊ณ ๋ฅผ ๋ถ์ํ ๊ฒฐ๊ณผ, AI ๋์ ์ ์ด๋ฌํ ์ธ๊ฐ ๊ณ ์ ์ ๊ธฐ์