Ep. 64: AI's Original Sins

One of the examples that I think is really helpful to illustrate this is a story by Ursula Le Guin.

The ones who walk away from Omilas, hopefully I'm saying that correctly, haven't read the source directly, but have looked at a number of works on sort of philosophy and ethics that have analyzed this.

In this fictional work, you have a highly prosperous utopian city that is wonderful for all of the citizens that are there, except for the only way that it can work is that there's a single child who's kept in this state of perpetual misery.

And if that changes, then everything in the society will fall apart.

So thinking about how you feel about that situation, is that something that you think can justify the suffering of one person for the benefit of many? But that's a foundational argument that we'll look at when we look at AI and AI ethics and what trade-offs are okay and what might not be okay.

And again, these are ethical value system, so there's not necessarily one right or wrong way to do it, but something that I would encourage you to examine and see how your own beliefs align to that.

Now, you could contrast that view with a utilitarian approach where you're simply looking at maximizing systems or choices that have the greatest good for the greatest number of people.

So in that same case, in that city of Omelas, you might do the calculations and say, if we look at all of the benefits of having this highly prosperous utopian city, this massive value for a high number of people.

And yes, we do have this one child that is in misery and suffering, and that is significant.

But if we turn that into a math equation and subtract that negative from the positive, it still results in this massive net positive.

So that's the utilitarian approach.

And I think that's the approach that we see in a lot of the leaders in AI labs and Silicon Valley.

So it's not necessarily surprising that we might see choices based on this worldview that you might have some negatives or negative consequences as a result of these systems.

But if that can be put together in an equation that adds up to what we see as an overall net benefit than from an ethical and philosophical framework that's justified.

So we use the term original sin sometimes looking at AI systems.

And I would say one of the differences with sort of quote unquote original sin for humanity is that's a one-time thing.

event, taking the bite of forbidden fruit and knowledge and all of that, that happened once and then sort of consequences are subsequent from that.

Whereas some of these sins and choices with AI systems are not just a one-time event, they're perpetuated and ongoing, which means there's a choice or potentially a choice about continuing them or taking other approaches.

And in this case, as often we're saying AI, we know AI can be a lot of different things.

We're referring to generative AI based on large language models.

So let's get into it.

Let's get into the original sins of AI and large language models.