Funding for To The Contrary provided by: This week on To The Contrary: The protection of our physical personhood and the protection of our digital personhood in the law has not been rectified.

We need to be thinking about preemptive solutions and trying to get to a place where these harms aren't happening all the time.

Hello, I'm Bonnie Erbé.

Welcome to To The Contrary, a discussion of news and social trend from a variety of perspectives.

AI chatbots are being used in ways that fuel violence against women and girls.

New research show that chatbots can simulate rape, abuse and harmful role playing with few safeguards.

They could even simulate dangerous behavior and guide real world abuse like stalking.

These harms are linked to design choices and weak regulation.

Joining me this week are Dr.

Mary Anne Franks, a legal scholar and professor at the George Washington Law School, and Dr.

Brandeis Marshall, a data scientist and AI expert.

Welcome to you both.

So let's start with both of you, please.

And you go first, Dr.

Franks, tell us what is an AI chatbot and what does it do?

So an AI chatbot is a progra or an interface that simulates human conversation and it uses a sophisticated—AI Chatbots use this kind o sophisticated machine learning to simulate these conversations in a way that seems very natural and very much like a real conversation that you're having with a human being.

How does it do that?

A lot of it has to do with training on large data sets that are made up of all of our speech and all of our information, and a sort of training from that.

In general, what happens is that there are algorithm that are used in order to train data that is very muc based on human like responses.

So it's taking the information that we give it inside that algorithm, then performing training and testing on that data in order to simulat how a actual human will respond.

And then it will predict the text one word after the other, and then phrase after phrase, and then sentence after sentence, and then having thi simulated conversation thread, that's how chatbots work.

And it then learns from massive amounts o information that is given to it in training and testing, but also it uses massive data that we as humans provid it, intermediary and in between.

So it doesn't really understand what the human is actually doing and saying.

All it's doing is regurgitating statistical relationships and exploiting them to make it seem as though there is a human intelligence behind the algorithm.

But it's really just glorified math.

Right?

But you hear so much.

I'm not a techie, and I read so much about all the bad outcomes from AI and the potential bad outcomes.

And even this, the AI chatbots.

I mean, why aren't they regulated?

It would seem to me to be an easy thing to say, you can't make a pretend rape or you can't make a pretend sexual assault.

Why hasn't—?

Why can't that be done?

I think there's a lot of answers to that question.

And one of them is that it's not that it can't be done.

I would start there, and I wouldn't even say that it's necessarily the case that you— that these things are not illegal in some of the circumstances that we've seen reported in the media.

You think about how there are laws against abuse and harassment, even if it's not physical.

And so the assumption that I think many people have that if a bot does it, it's fine.

Not necessarily true.

A bot is capable of producing defamatory content and fraudulent content and deceptive content and exploitative conten and all of those things I think we should assume can be brought before some kind of— in some sense can be objected to on some legal grounds.

But I think the larger question that you're asking is, why is so much of this being allowed to happen and to flourish, and then sen people kind of scrambling around to see if they can, in particularly egregious situations, get some kind of justice?

And that answer I think, is because we have had more than 30 year of a really laissez faire attitu to the tech industry, actually worse than lassez faire.

It's been more of a, here's a blank check, do whatever you want, we believe in innovation, things are going to be great and really has made it difficult for any meaningful regulation to happen in this area, by which I mean any consequences or incentives for these major companies.

And right now we're talking about billionaire companies, right?

To have any reason to thin that they'd be held accountable for the terrible things that their products are inflicting upon the population.

That's really why we're facing that situation now, is because of that sort of terrible bargain.

I think that Congress, at least in the United States is made with the tech industry.

And your thoughts, Brandeis, on this.

Okay.

She answered it the legal way.

Let me answer it the algorithmic way.

Inside of tech, there is this understanding, in fact, acceptanc that innovation equals progress.

And that, to me is one of the biggest misnomers.

Not everything that tech produces is actually good.

Everything that tech produces can have a bad actor use it and manipulate it.

So there has been instances where AI chatbots have been helpful.

Like in customer service, there might be some writing assistance that could be helpful for folks or even health care triag in certain limited capacities.

So algorithmically, originally a lot of this work was done within research labs, very much in controlled environments.

But now tech has taken it outside of controlled environments.

And now anybody has access t these algorithms and can build and scaffold these technologie without any type of regulation.

So that means that unfortunately, tec has been given this blank check.

But what I think is so important to realize is that when it comes to the algorithms, the algorithms themselves sometimes the people developing them have no ide what the outcome is going to be.

That's part of the problem.

The algorithms are created.

There's no way to trace the algorithms.

So as a tech person, as someone that's within the AI and data governance realm, there is no way in order to predict what these algorithms are going to do.

You just raised something that seems incredibly consequential to me, which is there's no way to trace an algorithm?

So you can't tell if a data scientist has put a bad motiv into an algorithm or something that's going to end up with fake rapes and fake sexual assaults all over the place?

Very true.

And that's what tech doesn't want to talk about.

The algorithms have within it a certain amount of probability of certain outcomes being produced.

Now that outcome probability could increase or decrease based upon the choices of the user.

So there is some direct interaction between the user and the algorithm which again tech doesn't talk about.

So now the algorithm can go down a particular rabbit hole and not be abl to get out of that rabbit hole.

And this is where bad actors can really exploit.

This is where data breaches happen.

This is where cyber attacks can ferment and grow.

These algorithms originally did not have a bad intent.

Originally, the developers assumed that there was going to be a certain set of processes tha were going to happen, a certain workflow, a certain pipelin that was going to be executed.

But then another person decides that they are going to switch a lever, they are going to add more weight to a certain process, o they're going to limit a certain outcome from happening.

And now you have essentially a different set of outcomes that are possible.

In addition to the fact that when we're talking about agentic AI and generative AI, a lot of the outputs of one algorithm doesn't get vetted or verified before it becomes input into another algorithmic system.

So that means that there are n guardrails happening around it.

So there's no wa for a developer in order to know where their algorithm is going to take as far as the output, they have no idea because their original intent was one set of rules, but once it interacts with individuals, and once other people are able in order to manipulate that algorithm anything is completely possible.

So how do we get technology going on the route to ban what's clear— what would be illegal behavior if a human did it, like rape, like sexual assault?

And just because it's being done by a machine or by a computer doesn't make it legal.

One of the things I want to highlight about what Dr.

Marshall has expressed so eloquently about how this works, right?

How that there's a breakdown between original intent, let's say, and eventual consequences.

This is something that the law is capable of thinking about, because of course, thi can happen with human behavior too, that people will make certain decisions that maybe seem completely innocuous or inconsequential, and then someone else will add to that, or will manipulate that, or use that to exploit someone else.

And the law, normally, we say that that might be trick for purposes of accountability, but it doesn't make it impossible.

We have to think about, well, who contributed, who was negligent, who was a reckless, who was intentional.

And we can think about ways to impose consequences on those actors so that in the future they are incentivized not to act that way and to be more responsible.

The problem I think that we have this law called Section 230 of the Communications Decency Act that has been tellin the tech industry for 30 years that essentially they don't have to operat by the rules that anybody else operates by.

Now, there are some exceptions to that.

It's certainly going to be a question about whether Section 230 applies in any of these cases, because they're not the same thing as social media platforms or more typical tech questions that we've seen before.

But the attitude that it has cultivated among the industry is we, as the industry, are allowed to do things that can break u the kind of connection between a person acting in a certain way and certain consequences.

And you can break it u and break it up and fragment it so that you always hav a kind of plausible deniability to say, well, I didn't do anything.

I'm just one piece doing this one thing.

And the problem is, of course, that that is going to creat exactly the society we have now, which is you've now given this industry a license to do tha and to have incredible profits in doing so, and never have to worry about cleaning up their messes.

Now, why can't there just be a law passed, or you know, on the state level 50 times that says anything that's illegal to do in real life, like murder like sexual assault, like rape is also illegal if you do it in a chat bot or if you do it through a computer, why can't you just pass a law like that and get rid of all these awfu things that computers are able to do right now?

There's an assumption that that isn't true, right?

That is, somehow what happens online is not accountable in the same way.

And I think as a matter of law, that isn't the case.

As a matter of practical accountability, for all the reasons that Dr.

Marshall is saying, it's difficult to know who, right?

You can't just say, well, these things are illega and expect that to mean anything unless we can say, this is the person who's responsible for the illegal action.

So I think that that kind of fragmentation that I was talking about makes it hard as a practical matter.

But if you are soliciting someone's murder using a bot, that's still illegal, right?

That is still something that can be illegal for the person who's using the bot for those purposes.

The question is, are you going to be able to build a case, given the fact that all these complexities now can be built in, but it's not as though we actually have a rule that says you can do whatever you want using a bo that you couldn't do in person.

It just makes it more practically difficult to figure out who to hold accountable, and also really changes some hard questions in the law about if you didn't mean for something to happen.

Can you be held accountable for it?

And in the offline context, we often say the fact that you didn't have the purpose of causing harm isn't enough to sa you're not responsible for it.

We have differen levels of liability according to whether it was your purpose to do something, whether you kne something was going to happen, whether you were reckless with regard to the risk that it was going to happen or whether a reasonable person would have know that this was going to happen.

And all of those things should be possible here, too.

And I think that we don' want to assume we can't do that.

I think we need to be calling upon all of, you know, our processes and systems to do exactly that.

But again, that's going to be after the fact, and it's going to be after the harm has happened.

We need to be thinking about preemptive solutions and trying to get to a place where these harms aren't happening all the time.

When these algorithms are created, when these AI chat bots are then released into the world, it's not just one person.

These are teams of people, typically under an organizational umbrella.

Multiple individuals are touching the code at different junctures.

Some people are in the design, some people are in the implementation, some are in the performance, some are in the maintenance of these technologies.

So who has the responsibility to ensure that this AI chatbot is not going to be used for malfeasance?

There is no way of of even understanding how to make that happen, for developer, for the product team, for the software engineering team, for the data team.

There is no way within the organization, there's no architecture, there' no structure for that to happen.

Because the goal, remember, i innovation is equal to progress.

So you've got to make it, got to make it.

You got to make it fast.

You got to get it out as soon as possible.

Because speed and scale is wha technology is pushing forward.

So all of the guardrails and boundaries that Dr.

Franks is talkin through would be the solution.

But tech is firmly opposed to any type of restriction on getting the product out fast.

They opposed all of it.

So this is why lobbyists are there in order to try to stop any type of regulation at the state level, as well as the federal level.

This is why technolog companies are doing their best in orde to continue to release products and have the availability of APIs in order to release products faster from individuals.

So there is nothing stopping the bullet train when it comes to technology.

They want to get the technology built.

They want to get it released, they want to get people using it.

Because once you have people dependent and reliant on the technology they don't want it taken away.

And they're banking o the general population being so enamored with the technology that they forget, they don't care about the harms that are being created, that they don't want to prioritize protecting children, protecting women from these types of harmful acts and these harmful images.

Well, what is it about the tech industry that would make— I mean, I understand the money and getting—get—excuse me— getting more customers, getting more people involved.

But what is it about AI that makes these things so realistic that they're just about as effective as if rape or sexual assault happens in real life?

Why does that become a reality to somebody who's seen it online?

It's possible to, and I think often it is the case that we overstate the difference or the harm between, let's say, a human conversation and a non-human conversation, because when we think about it, right, human conversations can also be very destructive.

They can be really traumatic.

They can encourage people to do terrible things.

But I think the difference is that when you're talking to a bot, you think that—there's a tendency to think, well, because this is a machine, it is a bot, it is objective.

There's something less biased about it, or it's giving me better information tha the average human being would.

That becomes really dangerous when we've already seen tha the phenomenon of chatbots is to essentially to please the user, right, and become very sycophantic and agree with them and validate their feelings.

That's really troubling.

Again, human beings do this all the time, but there are fewer ways to interrupt that process, because people may think that this bot is superior in some way to human beings and also because unlike a human conversation, a conversatio with a chatbot might never end.

You don't have as many opportunities to interrupt it, right?

You're talking to this bot in an, you know, all the time.

Other voices get crowded out.

No one is giving you a critical perspective, no ones sort of stopping and giving you a chance to think about it from a different way.

I think that's why they can become even more destructive than human interactions is because there are much fewer opportunities to intervene, and there's this sense that the conversation you're having is somehow more real, more empathetic, more professional than the conversations you might have with a human being.

The systems are designed t respond directly to the person.

This is very enticing because it adapts to their tone and their preferences.

This is what the algorithmic optimizations are doing.

This is what tech is doing right now is all these optimizations.

Trying to customize it as much as it can to the particular users.

And then thirdly, what Dr.

Franks said about the conversation persisting over time.

These are continual conversations.

So it's not as thoug it's one time you get an answer and then you move on to something else.

No, it is— the conversation is store and remembered by the “system”.

And I put these in quotes so that when you come back inside of the system, it can continue to perpetuate that particular line of thought in that particular line of understanding to you.

So you get into a silo, you go down that rabbit hole.

This is what the algorithms are designed to do: to keep your attention, to keep you separated and focused on them, and only to accept the authorit of those results, those outputs as a 100% accurate.

Even when they are not.

Why cant— Why can't Congress pass a law that says—anything that's illegal, because it's sexually harmful, like rape, like sexual assault, like stalking that's illegal in real life is also illegal if it's copied by computers why doesn't that just wipe out this whole AI industry that's been designed to follow an algorithm, to go down to each person individually, get them stuck, get them in this content and not let them out of it?

First Amendment.

Go ahead and talk about freedom of speech.

And we definitely have to talk about how the First Amendment does affect a lot of this.

But part of it is the way that we define these harms.

Right?

If we're talking about sexual assault, if we're talking about rape, these are harms that are defined to mean physical harm.

And so it wouldn't—we wouldn't be able to say that a rape—or we have not made a decision as a society to say that a virtual rape is in fact the same thing as a physical rate, because rape is defined as a physical injury of a certain type.

Now, that doesn't mean that we don't have more expansiv conceptions to sexual harassment or harassment generally, or stalking.

Those are other kinds of harms that are not necessarily physical.

But it is worth noting that when the Supreme Court took up a stalking case a couple of years ago, Counterman versus Colorado, they said that a large part of stalking actually is First Amendment protected.

So the problem isn't really just the bots versus non bots.

It's that our law does a terrible job of tryin to actually impose consequences and accountability for harm that disproportionately affect women and girls, and that gets compounde by the technological interface, because it makes it even harder to name the party and to define the terms of the offense.

But a lot of this is, yes, especially with chatbots and even with virtual visual representations.

All of this has a tendency to get classified as speech.

And the more that something is classified as speech, the less likely it is that at least in the United States, it will actually be designated something harmful.

So that is a bigger problem that is not just about technology, but about our legal system.

The protection of our physical personhood and the protection of our digital personhood i the law has not been rectified.

And what tech can do is with these algorithms is to push the envelope.

And that's what they've done over the past 30 to 40 years.

They keep trying to push us further and further into the digital realm where the laws do not reach and there is no will, in order to ensure that our digital presence is just as protected as our physical one, and even our physical presence in some cases has been under attack and have been targeted.

So tech is just exploiting the loophole that has gotten wider and wider, with the ability of AI being able to produce technologies very quickly and very harmfully without any guardrails.

And there's no wa to just make furthering research into that topic illegal?

Well, no, because morality and legalit are two separate conversations, and not everyone that is in tech has a moral compass.

And I think, too, we want to think about how we do want to distinguish between what is genuinel something we'd want to protect in terms of expression an in the First Amendment context.

It is important to note that it's not just becaus we think everything that we like should be said, and things we don't like should not be said.

It's that we thin as a matter of a free society, we have to tolerate a lot of stuff that many of us would fin offensive or wrong or malicious, because we don't want it to be used in a way.

We don't want laws against certain kinds of expression to be used by the powerful, against the powerless and in political ways.

And I think tha that's important to keep in mind that because it's an incredibly good aspiration to think about breathing room for expression.

But I also think we need to face the reality that it is not as though having those kinds of restrictions has kept the law or powerful people from using those laws to silence their critics and to silence dissent.

So it isn't as though this is necessarily something we should always cling to, as we have to tolerate even incredibly harmful forms of expression, because otherwise political figures will be able to— political powerful figures will be able to oppress those who are less powerful.

That's already happening, and it always has happened.

So maybe we just need to be more honest about what we mean by harm, need to be more honest about who's harm counts, and we need to be reall consistent about when and where we're able to say that sometimes some forms of expression can be so harmful that they should, in fact, be punished and regulated in advance.

Well, it's a scary and dangerous universe out there, and I hope everybody is keeping themselves as safe as humanly possible in this environment.

Thank you for a very enlightening discussion.

That's it for this edition of To The Contrary.

Let's keep talking on social media, including X, Facebook, Instagram and TikTok.

Reach out to us @tothecontrary and visit our website, the address on the scree and whether you agree or think to the contrary, see you next time.

Funding for To The Contrary provided by: You're watching PBS.