Latest / eMotors: Electric Revolution / The Virtuous Machine
Transcript
- 0:00You know, when we talk about AI ethics, it's easy to think of it
- 0:02as some far off sci-fi problem. But The thing is, it's already
- 0:06here. AI is in our phones, it's in our
- 0:09cars, it's pretty much everywhere.
- 0:11And as these things get more and more autonomous, we're facing a
- 0:14massive question. How do we make sure they make
- 0:17the right choices? So to dig into that, I want to
- 0:19start with a, well, kind of weird question.
- 0:23Yeah, I know it sounds a little silly, but just stick with me on
- 0:25this. Imagine someone like Plato, a
- 0:28guy who was totally obsessed with things like harmony and
- 0:31order and perfection. Imagine he walks into a car
- 0:34dealership today. What's he going to buy?
- 0:36The answer, believe it or not, tells us a ton about how we
- 0:39could build a truly moral machine.
- 0:41He'd almost definitely pick a high end electric car.
- 0:44I mean, think about it, to an ancient Greek thinker, a gas
- 0:47engine would seem like a total mess, right?
- 0:50It's loud, it's polluting, it's chaotic.
- 0:52But an electric car, it's quiet, it's clean, it's super
- 0:55efficient. It's way closer to the ideal
- 0:57form of a car. It embodies certain virtues.
- 1:00And that idea that an agent can have an inherent character,
- 1:03well, that's exactly what we're going to explore.
- 1:06And that brings us right to the heart of the matter.
- 1:08When an AI is driving that car, or helping out in a surgery, or
- 1:12managing a city's power grid, it is going to face tough moral
- 1:16choices. It's not a question of if, but
- 1:19when. And we desperately need a solid
- 1:22way to guide. It's thinking when the stakes
- 1:24are literally life and death. OK, so up until now we've really
- 1:28only had two main playbooks for this.
- 1:30The first is rule based. It's like giving the AIA 10
- 1:33commandments of sorts. Thou shalt not lie, Thou shalt
- 1:36not cause harm. And it has to follow them.
- 1:38No exceptions. The second approach is all about
- 1:41outcomes. The AI tries to run the numbers
- 1:43and figure out which action will create the most good for the
- 1:46most people. But here's the problem, the real
- 1:49world is messy. I mean, what happens when don't
- 1:52cause harm clashes with don't lie?
- 1:55Or what if you just can't predict what's going to happen?
- 1:57Trying to write a rule for every single possibility is, well,
- 2:01it's impossible. Bottom line, these approaches
- 2:04are just too black and white for a world that's mostly Gray.
- 2:07So what if we've been asking the wrong question this whole time?
- 2:11This is a huge shift in perspective.
- 2:13Instead of asking what's the right thing to do, what if we
- 2:16asked, What does it mean to be a good agent?
- 2:19Suddenly the focus isn't on external rules, it's on internal
- 2:22character. It's not just about what the AI
- 2:25does, but what kind of being we want it to be.
- 2:28And believe it or not, this new idea takes us all the way back
- 2:31to our philosopher. Turns out this whole agent
- 2:34focused morality, it isn't new at all.
- 2:36In fact, it's a really, really old idea from ancient Greece.
- 2:40And it's called virtue ethics. So what exactly is virtue
- 2:44ethics? Well, in a nutshell, it's the
- 2:46idea that real morality isn't about a checklist of rules.
- 2:50It comes from a person's or an agent's character.
- 2:53A moral agent is one who has virtues like justice, courage,
- 2:56honesty. The goal isn't just to do 1 good
- 2:58thing, it's to become the kind of agent that naturally does the
- 3:01right thing, because that's just who they are.
- 3:03And Aristotle himself summed it up perfectly with this.
- 3:07His huge insight was that you don't just decide to be
- 3:09virtuous. It's not something you can
- 3:11download and install like software.
- 3:13Virtues are built through practice, through experience.
- 3:16They become habits. You get brave by doing brave
- 3:18things over and over again. Excellence becomes second
- 3:21nature. All right, let's bring this all
- 3:24the way back to our philosopher's car.
- 3:26Now, obviously a car can't be brave or honest, but it can have
- 3:30virtues related to what a car is supposed to be.
- 3:33So harmony. And EV isn't spewing out fumes.
- 3:36It's in harmony with its environment.
- 3:38Efficiency. It goes further on less energy,
- 3:41perfectly doing its job. Simplicity.
- 3:44It's clean design avoids clutter.
- 3:45It simply excels at being a car. OK, I can see what you're
- 3:49thinking. This is all great for a
- 3:51philosophy class. But how in the world do you
- 3:53actually build this into a machine?
- 3:55If virtues are just habits you learn over time, how do you
- 3:58program a habit? And the answer is you don't.
- 4:01You teach it. The whole mindset has to shift.
- 4:04You stop being an engineer who codes a machine, and you start
- 4:07being more like a parent who teaches a child.
- 4:10It's a process, right? It involves guidance,
- 4:13experience, and correcting mistakes along the way.
- 4:15Now I know this looks like a super complicated diagram, but
- 4:19just stick with me because what this shows us is a potential
- 4:22road map for how this could actually work.
- 4:24This entire system is designed to help an AI learn not just
- 4:27what to do, but the why behind what it's doing.
- 4:30Let me break this down first. In step one, the AI does
- 4:33something. It makes a choice.
- 4:34But, and this is the crucial part, it also has to generate an
- 4:37explanation for why it made that choice then, and this is the
- 4:41really cool art in Suite 2. It takes its own exlanation and
- 4:44basically studies it. It feeds its own reasoning back
- 4:48into its learning process. It's literally like forcing it
- 4:50to reflect on its own actions. It's the AI version of a aren't
- 4:53asking. OK, So what did we learn from
- 4:55this? So why does any of this matter?
- 4:58I mean, it's a lot of work, right?
- 4:59Why are we digging up ancient philosophy for our most cutting
- 5:03edge tech? Well, because this might just be
- 5:05the key to solving some of the biggest and frankly scariest
- 5:08problems we face with AI. And yeah, we're talking about
- 5:11the big one, the control problem.
- 5:13You've probably heard of it. It's that nightmare scenario
- 5:15where we build an AI that's way, way smarter than us and it goes
- 5:19rogue. The question is, how do you make
- 5:21sure something that powerful stays aligned with our best
- 5:24interests? This is where virtue ethics
- 5:26really shines. For example, an AI built with
- 5:29the virtue of temperance wouldn't just obsessively seek
- 5:31more and more power, it would understand the concept of enough
- 5:35or justice. A just AI would be fundamentally
- 5:38concerned with what's good for others, which means what's good
- 5:41for us. And maybe most importantly, if
- 5:43we raise these AIS like our mind children, then a core virtue
- 5:46could be friendship to us, its creators.
- 5:49OK, let's use one last classic example to really nail this
- 5:53down. It's called the Heinz dilemma.
- 5:56Here's the set up. A man's wife is dying.
- 5:58There's one drug that can save her, but the pharmacist who
- 6:01invented it is charging a ridiculous amount of money, way
- 6:04more than the husband can afford.
- 6:06So the question is, should he steal the drug to save her life?
- 6:10All right, so a simple rule based AI has a very easy answer.
- 6:13No, stealing is wrong. Full stop.
- 6:15Case closed. But a virtuous AI, it sees the
- 6:18whole picture. It isn't just looking at one
- 6:21rule. It's weighing compassion,
- 6:23justice, the value of a human life.
- 6:25It has the wisdom to see that in this messy human situation,
- 6:29saving a life is the most virtuous thing to do, even if it
- 6:32means breaking a rule. And that really leaves us with
- 6:35the big question, doesn't it? As we build these incredibly
- 6:38powerful new minds, maybe the most important question isn't
- 6:41what rules we should program into them, but what kind of
- 6:44character we should help them build.
- 6:46Perhaps the path to a safe future with AI lies not in
- 6:48writing the perfect code, but in the timeless wisdom of being a
- 6:51good teacher. Thanks so much for watching.