Transcription
DAVID KIRON: When implementing AI, most organizations obsess over the technology, but our research reveals a surprising truth. Philosophy is what truly determines AI success. Today, Michael Schrage and I will unpack how philosophical frameworks can transform your business initiatives, sharpen your AI strategy, and supercharge execution in ways algorithms alone never could. Think of this as your guide to extracting actual business value from AI, not just impressive demos, because the competitive edge isn't in having more sophisticated models, but in thinking more clearly about what they should actually accomplish. This isn't theoretical. It's about ensuring your AI investments deliver genuine competitive advantage. Let's dive in.
(bright music)
So we wrote this piece, "Philosophy Eats AI." Why don't we start off by talking a little bit about what the main thesis is?
MICHAEL SCHRAGE: Well, the main thesis was based on what's next. And so the obvious question is, "If software is eating the world and AI is eating software, what's going to eat AI?" And the logical, dare I say teleological, ontological, and epistemological answer to that question was clear: Philosophy is going to be eating AI. And the reason why this was not just a good bet, but a great insight, was GPT, generative pre-trained transformers, training, learning, education. These are all philosophical constructs.
DAVID: Yeah, they're all these embedded philosophies.
MICHAEL: Exactly. That's the beautiful irony that our piece doubled down on, AI should not be seen overwhelmingly as just an ethical or a technical or a digital innovation and platform. It's actually a philosophical capability and resource.
DAVID: Philosophy has had a large, almost outsized role in the development of computer systems.
MICHAEL: But what I think, though, is really, really different though, is when you look at generative pre-trained transformers, we're not looking at legacy notions of logic, or reason, or rationality. What we're looking for are what aspects and elements of patterns are similar or relevant to one another? So it really does offer, pun intended, a different philosophical insight into how meaning gets made.
DAVID: Yeah, I mean there's this relationship between computation, reason, and patterns.
MICHAEL: Yes.
DAVID: And that's... The value of that, of the dynamic between those has really shifted in this new world of agentic AI.
MICHAEL: And the notion of agents simply being mechanisms that perform tasks as opposed to entities that have, yes, agency, where they can make choices, where they can make evaluations, where they can make trades. I think that's profound. We're moving agents, agentic AI, from the notion of how should we ensemble agents as enablers or performers of tasks into what kind of emergent intelligence? How would you train an ensemble of agents or a swarm of agents to solve a problem, to rethink or refine a context? The ability to play with patterns, to meaningfully play with patterns, buys you way more than anybody would've expected. And the ability to simulate reason may be good enough, or I'll stick my neck out and say better than the real thing, for a lot of problems, situations, and use cases.
DAVID: What I find a really compelling angle to what we discuss in the article is the notion of bounded rationality is different from bounded patterns.
MICHAEL: Actually, that's the kind of thing we should put into ChatGPT. Here is the Herb Simon Nobel Laureate Prize-winning notion of bounded rationality, but what is the counterpart to bounded patterns or pattern constraints?
DAVID: Purpose. Purpose.
MICHAEL: Right.
DAVID: Could be an important part of what bounds, creates some boundaries around.
MICHAEL: So we could play with this. What is the purpose of patterns? What are the patterns of purpose? And that kind of reflexivity, I think, is going to be a source of inspiration for how large language models and small language models get trained.
DAVID: Well, let's connect that with the Google Gemini example that we discussed.
MICHAEL: Great case.
DAVID: Where Google Gemini in one of its early instantiations wound up misrepresenting historical things like World War II and America's Founding Fathers. There was a lot of diversity that didn't exist back then. And it was almost as if there were competing objectives that were going on in the model.
MICHAEL: That was your excellent phrase, "teleological confusion." And I think that's exactly right. What happens when you have purposes that are at odds or in the first gen Google Gemini case, genuinely conflicting and that you basically, and I picked this word deliberately, you privilege the diversity purpose over the historical accuracy purpose.
DAVID: I don't think they thought it through. It's like you have these diversity, equity, and inclusion mandates. And you have historical accuracy mandates. They had their ontology for what the world is like in terms of, and from an accuracy point of view, and they also had an ontological point of view about, what diversity should look like, and these were at odds with one another.
MICHAEL: The irony is they did this with an ethical imperative. My own view is not that they didn't think it through, it's that they didn't care until these things generated bad publicity as well as inaccurate history.
DAVID: All right, so you've got indifference and I've got like a lack of commitment.
MICHAEL: Which ironically are philosophical perspectives on this. I really believe that these corporate culture, technical value, what's our real philosophy, what do we really stand for issues have to become more important in a generative AI, predictive AI context and circumstance. I think these values issue and what is it that we're really seeking to optimize becomes more important and that's why philosophy eats AI. That's why this theme, this narrative, this argument, this assertion has to be taken more seriously.
DAVID: We promised in the article that there's some actionable thing that you can do with this talk about philosophy eating AI. What is, to your mind, one of the biggest things that executives can do with this discussion?
MICHAEL: To me, I think the most important actionable and inexpensive insight is organizations need to map what they believe... right now, with responsible and ethical AI, they're mapping the ethical and responsible elements and aspects.
DAVID: Responsibility mapping.
MICHAEL: Exactly, it's responsibility mapping. And you and I both agree that there's perhaps, been perhaps over-indexing and overinvestment in the ethical components of AI, the philosophy of AI. Our article argues that what is the teleology? And we had a good negative example. Well, what are the good positive examples in that regard?
DAVID: We also talk about Starbucks and Amazon.
MICHAEL: Exactly. The the whole notion of what does loyalty really mean? Is it the the superficial metric of repeat business or is it that actually your customers become advocates, they become champions, they become defenders, and you can track those elements and aspects of evangelism and defense and sharing communication that they celebrate. And that's the, you know, to me, one of the really intriguing things that an organization that has bothered to go through the agony and investment of digital transformation and then putting a intelligent stack or capability on top of that digital transformation, they really should be thinking and mapping, gee, what's the teleology? What's the purpose? What's the ontology? How do we label and categorize these things? Epistemology, what is the nature of knowledge that informs categorization and purpose? Do we have them in conflict? Is there a virtuous cycle? What do we want our software to learn? What do we want our AI to learn? What do we want our agents to learn? This is where you have a marriage of the technical capability with the philosophical need and the business purpose.
DAVID: One of the big practical insights is that if you are going to lead with metrics, you need to have a deep understanding of what you can actually measure in the enterprise.
MICHAEL: Yes.
DAVID: And that you have new techniques for measurement.
MICHAEL: Yes.
DAVID: With all due respect to Peter Drucker, like you manage what you measure. This was not like a complimentary thing, so you can only measure a small amount of your business activity. So now there's so much more business activity that you can measure that can drive performance, that you can, like, tailor metrics.
MICHAEL: I completely agree. My only pushback is it broadens the notion of what business means. You know, it's not just customer satisfaction or NPS, it's what is customer experience? What is the purpose of customer experience? What are the epistemological underpinnings or ontological aspects and categories of customer experience? Same thing with partnerships with suppliers. The whole notion of what the vocabulary of value creation can and should be is fundamentally disrupted and fundamentally changed. The essential narrative, the essential virtuous cycle there is you learn to prompt and you prompt to learn. And if you do it right, the way you learn to prompt informs how you prompt to learn. It's not just your learning, it's the machine's learning. If you are not learning as much as your AI models are, something is wrong with your human capital balance and human capital portfolio, and it's the same sort of thing. If your humans are learning faster than your models, gee, I think your ML ops aren't as efficient or as effective as it should be. What kind of virtuous cycle do we want to enable and empower between machine learning and human learning?
DAVID: Some of our recent work is focused on, like, reconsidering how decisions are made. But we're talking about right now reconsidering how thinking occurs in the enterprise, which is a significant transformation in how leaders conceive of their own role in the enterprise. If you can think out loud with an LLM or with an agentic AI, I mean, you can be augmented in all sorts of ways that you weren't able to before.
MICHAEL: As you come to have a greater understanding, in every meaning of the phrase, of what it is you seek to accomplish, what it means, and what the knowledge supporting that means and is, the whole notion of what you want to automate, make mindless, should become clearer and what you wanna augment, i.e., put the human in the loop, add value to the leadership's and the leaders' decision-making capabilities also becomes more acute. I think the whole notion of how we enable people to think more rationally, intelligently, and intentionally about the trade-off between automating and augmenting agents that think and agents that just perform tasks. I think that's a big aspect as well. So I think you're spot on.
DAVID: Are you in any way optimistic that those people who are inclined to think less rigorously now have a capability for thinking more rigorously as they think with AI?
MICHAEL: That's sort of like the cognitive counterpart to Ozempic versus Weight Watchers. Would I rather just take the drug or am I prepared to follow a diet? There really is a difference between leaders who take rigorous, comprehensive, philosophical thinking seriously, and those who are looking to optimize share price, market share, et cetera. Generative AI is the battleground and the battle space for competing in conflicting philosophies for value creation and experience.
DAVID: The bottom line, philosophy isn't just academic, it's a practical approach to AI that delivers meaningful business value. What philosophical questions are you asking before implementing AI in your organizations? Drop your thoughts in the comments below. And for a more detailed analysis, check out our article "Philosophy Eats AI" on the MIT Sloan Management Review website. And for more insights from our authors, check out this curated playlist. Thanks for watching.
(bright music)