Why AI Makes the Humanities More Important Than Ever Transcript, AI Summary & Key Points

00:00 AI can generate answers, but who decides the question? In the current AI era, we're rapidly improving machines that can summarize documents, answer questions, do research, write code, even compose music. These systems are becoming incredibly capable at producing outputs that look like understanding. But there's a critical gap emerging. We're building systems that can generate language.

00:25 Without necessarily understanding the purpose or value of what they're saying in any human sense of the word. To adapt a phrase from Oscar Wilde, AI knows the price of everything and the value of nothing. That leads to a difficult question. What does it take to responsibly use a technology that can speak so fluently without being grounded in good judgment?

00:49 Well, fundamental to that question is how do we know what is good and ethical in the first place. Where do we learn about the consequences of our actions and their impacts on people? Math class is great, but calculus can't answer that question. Same for physics and chemistry. Engineering is essential to making things work, but it can't tell us if we're making a good thing or a bad thing.

01:13 Computer science doesn't have the answer either. Code doesn't care. These things we call STEM, science, technology, engineering, and math subjects, They tell us. Really well how things work, but what they don't tell us is what all of this means. And they can't answer the why that we're doing all of these things, the purpose behind it all. Where can we go to inform our struggle for these really big questions, these overarching questions?

01:46 Well, the answer is the humanities. Subjects like literature, history, art, philosophy, things of that sort that may seem less important in a world dominated by STEM. But the reality is, we've never needed the humanities more than we do right now. And the good news is that these subjects have never been more accessible than they are right now, and that's due in large part, ironically, to advances in technology, such as the internet and AI.

02:18 Let's take a look at how the humanities can help inform us and make a better AI. I'm going to start with this. AI does not equal understanding. Understanding is not the same thing as just having a lot of data and a lot information. AI systems, especially large language models, operate on statistical pattern prediction. They learn correlations in language, not lived experience, intention or truth in the philosophical sense that we understand it.

02:48 Here's what I mean. So you take human knowledge and basically it's based on things like experience. It's based on intent. What are we trying to do? It could be- based on context. Something means one thing in one context and something means something else in another. It has to do with values, what's important to us and what isn't. There's also a sense of accountability that goes through all of this in the way that humans think about this.

03:16 But now AI as a model looks at it very differently. So AI models are based on patterns. So they're looking for probabilities. What's the next most likely thing that I should put in this? Token sequences, so I've seen these. Again, prediction of what should be next. Statistical associations. I've see a lot of these, so this one seems like it fits in here.

03:41 The key distinction is this. AI is fundamentally syntactic. So that's the way it looks at things. It's looking at the syntax, whereas humans look at things more semantically. What is the meaning of that? That's the sense that you and I think of this. AI is really good at manipulating symbols and things of that sort, but doesn't inherently anchor those symbols in lived reality like we do.

04:08 The humanities teach us really important things like what meaning really is, how language relates to reality, how interpretation changes across different contexts, and how ambiguity is resolved and sometimes actually exploited. Without this kind of grounding, AI outputs can seem correct, while being subtly misaligned, incomplete, misleading, or outright wrong.

04:33 Epistemology is another area that's important. This is a branch of philosophy that deals with things like what is knowledge? How do we know if something is true or not? What counts as evidence or justification? And what's the difference between belief and knowledge? Can we ever really be certain about anything? There's a lot of important questions here.

04:57 Those are really important, especially when we consider we're trying to judge the output of an AI that we know can hallucinate with confident errors. And when you consider that much of the knowledge AI has is based on what it scraped up off the Internet, well, that could present yet other problems for us. Look, if you remember nothing else from this video, please remember this.

05:19 Not everything on the Internet is true. Yeah, I know, that's a big shock to you, right? Well, you get my point. AI can give us tons of answers, but we still need to decide whether those answers are right. And more importantly, what should we do with that information? Again, not really a STEM question. This is one for the humanities. A common misconception is that AI is neutral because it's mathematical.

05:44 In other words, the numbers are just the numbers, but the truth is more subtle. The model may be mathematically neutral, but its use is not. So let's consider some different perspectives as we look at this. So we have the AI system, we have the designer, we had the user of the system, and then we have society. So the designer might have had one thing in mind and thought that this was right and this is wrong.

06:10 The AI system may come out with a different answer than that. And then the user may get yet a different answer still and have a different perspective on what they wanted, what their intent was, what was the right context. And then a user is not necessarily representative of all of society. So while an individual user might think this is okay, the rest of society may not.

06:33 So what we have to understand is that every training data set encodes error. And every interface encodes design assumptions. And every application encodes incentives. So these things will all guide the way the AI is going to actually work. And this is where ethics, political theory, and sociology become really critical. Humanities training helps us understand the answer to questions like, who benefits from this system?

07:06 These incentives will matter, and that will affect the output. What assumptions are embedded in the data? What kinds of harms might result from the way the system is used? What are plausible, if incorrect, outputs that could occur? Without this kind of lens, we risk confusing technical performance with ethical legitimacy. An AI system can be accurate and still do the wrong thing.

07:35 All outputs require interpretation. AI systems don't eliminate interpretation, they just shift it. Think about it this way. We have input into the AI system. It gives us some output. Okay, now what? Well, some human has to interpret this thing. And then they have to make a decision. And then we take action. So it's not as simple as input and action.

08:01 There are still a lot of steps that go on with this. And the key step in all of this is the one that the humans still own, and that's interpretation. Humanity's disciplines, especially things like literary studies, history, hermeneutics, which is the study of what things mean, train people to interpret under uncertainty. And we get a lot that with AI.

08:23 So these things teach us things like that multiple valid readings can coexist. You and I can read the same text and have entirely different meanings that come from that. That original intent is not always accessible. Sometimes we think we can discern that by reading a document, but if you weren't in the head of whatever created it, you might not always know what it was that they intended.

08:48 So text can also mislead without being wrong. So there's a lot of different things that we have to look at here. And AI maps this kind of behavior because AI outputs are probabilistic. They're context sensitive. Sometimes they're inconsistent. They're often plausible, but not always guaranteed to be true. So that gives us that level of uncertainty where something like hermeneutics becomes particularly important, where we're trying to determine what that meaning is and define that from what the outputs we've received.

09:22 So the user becomes the interpreter, not just the recipient, an active participant, rather than just a passive receiver. In other words, AI increases the importance of human interpretive skill rather than reducing it. A person trained only in technical correctness may miss subtle distortions, but a person trained in interpretation can detect ambiguity, bias, and omission.

09:47 One of the most underestimated shifts in the AI era is this. Communication is now a form of system control. In fact, I did a video previously on this where I tried to answer the question of what is the most important programming language in the era of AI. And spoiler alert is, it's the one you've been speaking since childhood. So go check out that video when you're finished with this one and you'll see why.

10:10 So let's consider we have a user here who is going to enter a prompt It goes into an AI system which is going to give us output back to this user. Very simple case. Well, that prompt and the quality of that prompt will make all the difference in the world in terms of the quality and the output we're going to get. So how we frame the prompt, what kind of constraints we put into it, what kind context, what kinds of things have we said previously to this AI that it may still be remembering, and what was our intent?

10:43 Is it clearly expressed or not? So this is directly analogous to rhetoric, which is the art of persuasion and structured communication. Humanity's disciplines like rhetoric, philosophy of language, and discourse analysis become practical tools for interacting with AI systems. Remember all those papers you had to write in composition class? Yeah, I didn't like them either.

11:05 But those skills will make you better at prompt writing as well, because that's where you learn that small changes in framing can change outputs. Ambiguity in instruction can produce variance. Implicit assumptions propagate into bad results. So the ability to clearly define intent, context, and boundaries is no longer just a writing skill, it's operational competence in AI systems.

11:34 A poorly framed prompt is not just unclear, it's computationally underspecified. AI systems reflect the structure of their training data, of course. So that means. That they can unintentionally reinforce majority viewpoints, they could under represent minority perspectives, they could ignore cultural nuance, they could even confuse frequency with truth.

12:00 In other words, the more they see, the more, they think this must be true, but that's not always the case. So humanities training helps introduce some corrective mechanisms that are really important in this. For instance, we learn about historical perspectives and awareness. We learn about cultural literacy, about critical analysis, about the awareness of narrative framing, what's going on around the story.

12:26 So these things all allow us to detect the difference between when likely output is not the same thing as appropriate or even useful output. In practice, this is what prevents us from overtrusting AI's fluent but potentially shallow misleading answers. Okay, so here's the point. AI expands the space of possible answers for sure, but it doesn't define which answers actually matter.

12:53 It doesn't tell us which values need to be prioritized or what interpretations are reasonable. It also doesn't us about which consequences are tolerable. We have to decide that stuff. Those remain the human responsibilities. And the humanities are the structured study of those questions, as we've covered before. So in an AI-rich world, it's not enough just to have technical fluency.

13:18 In fact, the most valuable skill, I would argue, is critical thinking, and that involves judgment. Things like ethical judgment, interpretive judgment, cultural judgment, communicative judgment. So the bottom line is this. We are becoming less human because of AI. In fact we're becoming more dependent on what it means to think like a human. Back when I was in school, and they told us...

13:43 That we study literature to learn things like man's inhumanity to man, and the struggle against nature, and pressures of society, and the conflicts within ourselves. Well, what we learn from all of those timeless lessons are the things that we need to use to inform our use of AI. In an interesting plot twist, the more we advance our technology, the more we need to be grounded in the humanities, in these things that I've talked about.

14:11 And with AI taking care of all the mundane tasks that used to burden our days, well, now we've got more time to enjoy a good book.