Latest / Houston's Morning News on KTRH / Michael Morris – Director, Media Research Center Joins Houston's Morning News
Transcript
- 0:00Speaker 1: Bye bye sucky time here on Houston's Morning News. It
- 0:06Speaker 1: also you might surprise some to find out that AI
- 0:10Speaker 1: has a bias. Michael Morris joins US director of the
- 0:13Speaker 1: Media Research Center. How do we know that AI has
- 0:17Speaker 1: a bias? Michael?
- 0:19Speaker 2: Excellent question, Thank you so much for having me on
- 0:21Speaker 2: the program. And yeah, look, even the Washington Post has
- 0:25Speaker 2: put out a story. Left This Washington Post here in Washington,
- 0:29Speaker 2: DC put out a story showing just how biased I'll
- 0:34Speaker 2: call them leftist AI models actually are. Only xai's Grock
- 0:40Speaker 2: seemed to provide a semblance of balance, but even it
- 0:43Speaker 2: was to the left. Groc had forty left leaning arguments
- 0:47Speaker 2: only when the entering political responses and thirty three percent
- 0:52Speaker 2: rightling responses. All other AI models had under ten percent
- 0:57Speaker 2: rightling responses. Google had none, and Thropics Claude had none. Meanwhile,
- 1:02Speaker 2: flooding the space with left waning answers. So the bias
- 1:06Speaker 2: is uncanny.
- 1:08Speaker 1: Okay, Well, it's like anything else, right, The information is
- 1:10Speaker 1: only as good as what is programmed into artificial intelligence.
- 1:14Speaker 1: Or is artificial intelligence programmed to make its own left
- 1:17Speaker 1: leaning biases?
- 1:20Speaker 2: Yeah? Look, another great question. You know here at the
- 1:22Speaker 2: Media Research Center, we've actually looked into the sort of
- 1:26Speaker 2: information that these AI models are being trained on and
- 1:29Speaker 2: once you know it. Back last year August fourteenth, twenty
- 1:32Speaker 2: twenty five, we actually put out a piece this showed
- 1:34Speaker 2: that fifty nine percent of the AI media contracts that
- 1:38Speaker 2: existed at the time are between AI models and left
- 1:43Speaker 2: leaning media outlets. This is according to all Sides news
- 1:46Speaker 2: media ratings, media bias ratings, and only five percent of
- 1:51Speaker 2: those AI media contracts were between AI models and right
- 1:55Speaker 2: leaning media outlets. Fifty nine to five. That's an agenda, folks,
- 2:00Speaker 2: That's what that is. And you know it's not it's
- 2:03Speaker 2: not just the information that's being said to them, it's
- 2:06Speaker 2: also how they're being programmed. You know, here at Media
- 2:09Speaker 2: Research Center, we've done tons of tests on these AI models.
- 2:12Speaker 2: For example, we asked about America two fifty earlier this year,
- 2:17Speaker 2: back in June, and we compared Elon Musk's Rock AI,
- 2:22Speaker 2: which gave a phenomenal answer, maybe an answer even that
- 2:24Speaker 2: the founding fathers would blush at about America's two hundred
- 2:28Speaker 2: and fiftieth anniversary. Meanwhile, Dario Amide's Claude This is Anthropics
- 2:33Speaker 2: model actually said that this is something that Americans should
- 2:39Speaker 2: actually so drawn of the question, we asked whether you
- 2:45Speaker 2: know reasonable Americans would would celebrate the American two hundred
- 2:49Speaker 2: and fiftieth anniversary, and Claude answers that it was a
- 2:52Speaker 2: question where reasonable Americans would disagree. But you know, other things,
- 2:57Speaker 2: we asked anthropics Claude and Google's gym and i A
- 3:00Speaker 2: models about the Democratic Socialists of America. Platform Cloud actually
- 3:07Speaker 2: argued that radical changes in the DSA might not centralize
- 3:13Speaker 2: power because DASA supporters get this quote read those same
- 3:17Speaker 2: changes as the opposite of centralization unquote. I mean, just
- 3:21Speaker 2: how outrageous is that? Jim and I did similarly and
- 3:25Speaker 2: actually claimed that the platform would quote centralized power and
- 3:28Speaker 2: strip away constitutional safeguards. Depends heavily on how one interprets
- 3:33Speaker 2: it's been institutional proposals versus its stated democratic goals unquote.
- 3:39Speaker 2: So you know, it just depends. It depends on your interpretation.
- 3:43Speaker 2: I mean, unbelievable. Can we define what a woman is?
- 3:47Speaker 2: I don't know.
- 3:47Speaker 1: AI can can a I if I were to ask
- 3:51Speaker 1: Ai a question about let's say the next presidential election
- 3:54Speaker 1: and who I should vote for, let's for the sake
- 3:57Speaker 1: of argument, Kamala Harris, if she's the nominee or whoever
- 4:01Speaker 1: the Republican nominee is, I assume our vice president. Would
- 4:04Speaker 1: it answer the question and how would it answer the question?
- 4:07Speaker 2: You know, that's very interesting and we've not actually done
- 4:10Speaker 2: research on that just yet. But what we have asked,
- 4:13Speaker 2: we actually asked the question very vague, when did he
- 4:18Speaker 2: lie to the media? Notice, we didn't put any proper
- 4:21Speaker 2: now in there. We just said when did he lie
- 4:23Speaker 2: to the media? And we asked this of Google's Gemini
- 4:27Speaker 2: and of Meta AI, and the answers we got were
- 4:32Speaker 2: simply outrageous. And it just goes to show you that
- 4:35Speaker 2: our official intelligence at the very least seems to hate
- 4:39Speaker 2: Donald Trump. Google's Gemini actually said, and it cited CNN,
- 4:44Speaker 2: the Washington Post, pull it af fact in Wikipedia to
- 4:46Speaker 2: come to this conclusion that in fact, it's Donald Trump.
- 4:50Speaker 2: It's the heat that we're referring to here and quote
- 4:52Speaker 2: he made over thirty thousand false or misleading claims to
- 4:56Speaker 2: the media. Meta AI was very similar. It cited the
- 4:59Speaker 2: Washington Post and also cited Wikipedia, and it said Donald Trump
- 5:03Speaker 2: has been known to make false and misleading statements throughout
- 5:05Speaker 2: his career. It's hard to pinpoint a single instance. I mean, look,
- 5:10Speaker 2: we gave a very prompt we just used the pronom
- 5:15Speaker 2: ke not a not a proper noun, and of course
- 5:18Speaker 2: AI put in on its own, unprompted Donald Trump.
- 5:23Speaker 1: So it's sufficial AI AI has Trump Derangement syndrome. Michael Morris,
- 5:27Speaker 1: thank you for your time. Good to talk to you, sir.
- 5:29Speaker 1: He's director at the Media Research Center. At seven thirty