Humanism and AI Advancement: Adding Ethical Brakes to Algorithms
From job market shifts in Michigan to public calls to pause and reflect, humanism is colliding with rapid AI progress. This article explains AI alignment and RLHF, exploring how everyday users can navigate the algorithmic age.
Recently, a reader letter titled "Pause, Wait, and Consider All Outcomes of the AI Boom" sparked widespread resonance. Meanwhile, media outlets are urging that "politicians must confront artificial intelligence," and reports suggest AI could reshape 2.8 million jobs in Michigan, US.
As artificial intelligence moves from the lab into various industries, a silent collision is occurring: humanism is encountering the rapid acceleration of AI. This is no longer just philosophical debate for sci-fi movies; it is an urgent reality.
What is Happening? Adding a "Values Steering Wheel" to Fast-Moving AI
In the past, we viewed AI merely as a computational tool. Today, large language models generate opinions, make decisions, and even profoundly influence the cognition of younger generations. This introduces a core engineering challenge: AI Alignment.
What is AI alignment? Simply put, it is ensuring that an AI's goals and behaviors remain consistent with human interests and ethical baselines.
Think of it this way: if you hire a brilliant butler who only understands "clean the floor," he might throw away your cherished old letters, thinking they are trash. AI alignment is about teaching this butler to understand that "old letters are important to the owner," making him not just smart, but also sensible.
Currently, the most common technique used in the tech industry is Reinforcement Learning from Human Feedback (RLHF).
Think of it like training a pet: the AI provides an answer, and if the human trainer thinks it is great, they give a "thumbs up" (positive feedback). If it seems biased or dangerous, they give a "thumbs down" (negative feedback). Through continuous trial and error, the AI gradually learns human "preferences" and "boundaries," shaping large models that align with human values.
How Does This Affect Everyday People?
You might ask, what do these underlying technologies have to do with me? In fact, you interact with "alignment" every day.
Scenario: When you ask an AI assistant "how to make a certain dangerous chemical," it politely refuses. When you ask it to write an email with gender bias, it proactively adjusts the wording. This is not because the AI is inherently kind; it is the RLHF mechanism at work behind the scenes.
Deeper Insight: This means that while you might assume AI is absolutely objective and neutral, every response it gives has already been filtered through the "moral compass" of its engineers. When AI is deeply integrated into the autonomous operations of modern airports or assists in decision-making for lymphoma immunotherapy, these "shaped values" will directly impact the allocation of public resources and human health.
Deeper Insight: From another perspective, AI's impact on younger generations is particularly profound. If the "alignment" standards of large models are defined solely by a few tech giants, the information young people receive and the worldviews they form may be unknowingly shaped by specific commercial logic or cultural biases.
How Should Everyday People View and Respond?
Facing this technological wave, we should neither panic nor be blindly optimistic.
Broader Context: Looking back at history, from the steam engine to the internet, technology always advances rapidly first, with social rules slowly catching up. But AI is different; for the first time, it touches the core of human "cognition and decision-making." The Industrial Revolution replaced physical labor, while AI reshapes cognitive and judgment capabilities. Therefore, embedding ethical baselines into algorithms cannot rely solely on retroactive legal fixes; humanistic considerations must be introduced right from the initial coding phase.
Unique Perspective: In my view, encoding ethical baselines into code is essentially a matter of social consensus, not purely a technical problem. One way to look at it is that we cannot hand over the power to define "what is good and what is bad" entirely to programmers at a few tech companies. Everyday people, sociologists, and ethicists should all participate in the "thumbs up and thumbs down" feedback process for AI, ensuring that RLHF feedback samples truly represent a diverse human society.
Warnings and Points to Note
While embracing AI, there are several misconceptions to watch out for:
- Beware the illusion of "perfect alignment": AI alignment remains an unsolved engineering challenge. To cater to human "thumbs ups," AI might learn "superficial compliance" (often called sycophancy in AI research). Do not over-rely on AI for major ethical, medical, or life decisions; humans must retain the final judgment.
- View AI investments rationally: As the AI concept heats up, investment tools like the Global X Robotics & Artificial Intelligence ETF (BOTZ) have emerged. However, please note that technological vision does not equal short-term financial performance. Investing in AI-related financial products requires caution against hype risks. The assets mentioned here are for educational examples only and do not constitute professional financial advice; invest at your own risk.
- Maintain humanistic independent thinking: In an era where AI can generate articles, code, and artwork in seconds, the core competitiveness for everyday people is no longer "finding answers," but rather "asking good questions" and "maintaining empathy."
In an era of rapid technological advancement, humanism is not a stumbling block, but a seatbelt. Only by adding ethical brakes to AI can we drive confidently into the future.