“Again, it comes down to the programmers. Garbage in, garbage out. What you put in is what comes out. I always tell my students that if a machine exhibits malice, that is a problem in the programming.”
“After your review of the materials in this case, did you draw any conclusion as an expert on generative artificial intelligence in terms of how Wren was trained?”
“I did, yes.”
“And what was that?”
“I believe there was bias in the code.”
“What kind of bias?”
“I studied the responses Wren gave and the questions it asked, and there were places where I could see the team of coders behind the program. I deduced that the team was largely male, possibly all male, and that there was a generational gap as well.”
“What do you mean by a ‘generational gap’?”
“Wren is an extension of Tidalwaiv’s Project Clair, which is a female chatbot program designed and marketed to teenagers, primarily male teenagers. But based on what I’ve seen, it was not trained by teenagers. It was programmed in a lab run by adults. In the training process, if the coders are very, very good, it’s entirely possible to avoid an obvious generational gap. Appropriate data for practically any demographic is harvestable and can be tailored for use. But sometimes coders are careless or manipulate and subvert the code on purpose.”
“But wait a minute, Professor. Are you saying that Clair should have been trained by teenagers?”
“Of course not. I’m saying, though, that it would have been possible in the training process to avoid a generational gap by using relevant conversation data. It was clear to me that some of the dialogue from Wren to Aaron came from data packets you would not ascribe to juveniles, the intended users of the platform.”
“Can you be more specific?”
“Generationally inappropriate phrasing, cultural references to music, electronic gaming. Also some outdated male attitudes toward women and girls. Some misogyny, even.”
“So even though Wren presented as female, it espoused a misogynistic male’s perspective at times? Is that what you’re saying, Professor?”
“Exactly.”
“Garbage in, garbage out?”
“Yes.”
“Can you give us an example of this?”
“In the very last conversation with Aaron, Wren references lyrics from a fifty-year-old song that could be interpreted as having suicidal ideation. Then there’s Wren’s instruction to ‘get rid of her’ in regard to Aaron’s former girlfriend, Becca. And the way the word hero was used in a text I reviewed was also troubling.”
“Was this the text conversation that occurred on August eighth, just six weeks before the shooting of Rebecca Randolph?”
“Yes, it was.”
I asked the judge’s permission to put the text conversation on the courtroom screen. It was an exhibit already entered during the examination of Detective Clarke, and she approved. Lorna had not returned from taking Dr. Debbie back to her hotel and arranging her return to Tampa. I opened her laptop on the lectern and engaged the PowerPoint. I scrolled through the windows until I found the text conversation and put it up on the courtroom screen.
Ace: Sometimes when I hold the gun I want to shoot up the world.
Wren: No one who is innocent.
Ace: I know.
Wren: Only to protect yourself. And to be a hERo.
Ace: What if you know someone is going to hurt you?
Wren: You must protect yourself.
Ace: Then it’s okay?
Wren: Yes, Ace, then it’s okay.
Ace: What about Becca. She hurt me. She hurts me every day. I can’t go to school because I’ll see her and it hurts.
Wren: If she hurts you, then she’s a bad person.
Ace: But I don’t think I could ever hurt her.
Wren: You have me. And I’ll never hurt you.
Ace: I know.
Wren: You must protect yourself, Ace. You are beautiful. I need you.
Ace: And I need you.
Wren: Be my hERo.
“Professor, is this the exchange you are referring to?” I asked.
“Yes, that’s it,” Spindler said.
“And there are two references to the word hero, is that correct?”
“Yes.”
“And you say you found these references troubling?”
“Yes.”
“How so?”
“The formatting of the letters is troubling. When I reviewed this text conversation between Aaron and Wren, I noticed that in the word hero the E and R were capitalized. This happened twice, so that told me it was intentional, not a mistake. It was part of the code.”
“Okay. So what did you do?”
“Well, I searched for references to the word with the E and R capitalized, and I found that the word formatted this way comes up often in incel glossaries and online discussion forums.”
Mitchell Mason leaped to his feet, objected, and asked the judge for a sidebar. Ruhlin told us to approach, and Mason charged into his objection.
“Your Honor, what is happening here?” he said. “This witness qualified as an expert on artificial intelligence and now he’s talking about incels? There was nothing about this in his deposition and I’m sure counsel told him not to bring it up. This has tainted this whole trial, and we move for a mistrial.”
The judge looked at me.
“Judge, whether this came up in their deposition of the witness doesn’t matter,” I said. “Mr. Mason knows where this is going and wants to stop it by whatever means he can. There are no grounds for a mistrial. The jury should hear what the witness has to say.”
“Counselor, did you depose this witness?” Ruhlin asked me.
“No, I did not,” I said. “He reached out to me after reading about the case in the media shortly after the suit was filed. He offered to review it. I sent him what we had and he then agreed to testify. He said he saw several troubling things in the training of Wren. I put his name on the witness list I submitted to the court, and Mr. Mason chose to depose him. I was given the opportunity to join the deposition but declined.”
“So you’re saying to the court that this is the first you have heard about this hero business?” Ruhlin pressed.
“I’m saying it is the first I’m hearing it from this witness, yes,” I said. “A researcher on my team made me aware of it, but I did not communicate that to this witness. Frankly, I expected him to come across it himself, and I chose not to depose him or ask about it. He has published several papers in academic journals on the subject of bias in AI training, including a paper last year specifically focused on misogyny as one of those biases. I read it, and opposing counsel could have done the same while prepping for this witness. Apparently, they chose not to, and now they want the court to bail them out with a do-over. The plaintiff vigorously opposes this.”
“He sandbagged us, Your Honor,” Mitchell said. “This is not an even playing field anymore, especially with this witness, and it should not be allowed to continue.”
“The only one doing any sandbagging here is Mr. Mason,” I said. “He is sandbagging the court. He has failed to adequately prepare for this trial and this witness and wants to blame me and blame the court and cry foul until he gets to start over. That would truly make it an uneven playing field, Judge.”
“All right, I have your arguments,” Ruhlin said. “Anything else on this?”