AI-assisted fact checker published an inexcusably inept fact check on fetal pain
Do fetuses experience pain during an abortion?
Zebra Fact Check occasionally spot checks “verified” signatories of the International Fact-Checking Network’s Code of Principles. Our evidence strongly suggests the IFCN so far cannot begin to deliver on its claims that it verifies compliance with its code. Also, our evidence suggests mainstream fact checkers often demonstrate incompetence or else a unserious attitude toward reporting the truth.
Our latest example comes from LogicallyAI, a relative newcomer to the fact-checking industry. LogicallyAI, as its name suggests, uses artificial intelligence into its fact-checking methodology. Zebra Fact Check looked for a fact check on abortion to see if LogicallyAI was susceptible to the types of errors we’ve seen other fact checkers commit.
We found a fact check claiming fetuses do not feel pain from an abortion.
That claim piqued our interest, for the medical community uses “fetus” to refer to a developing human being right up through the moment of birth. What evidence might a fact checker use to justify such a claim?
The fact check consists of 11 sentences and three sources. Nearly every item deserves attention, but we’ll focus on the central evidence.
“A 2005 study published in the Journal of the American Medical Association concluded that a fetus could experience pain somewhere between 29 and 30 weeks after conception.”
Surprisingly, the first piece of academic evidence from the fact check allows for the possibility of fetal pain, directly contradicting its own finding that a fetus cannot experience pain. So, does this source have a problem, that LogicallyAI reasonably discounted the 2005 study?
“Researchers concluded that although fetuses are capable of producing a biological response to pain after the second trimester, the functional capacity of pain does not exist until 29 weeks.”
The above sentence directly follows on the heels of the first one we quoted. Like that one, this sentence suggests a fetus might feel pain after 29 weeks. It fits poorly with the fact check’s conclusion, and we’re already halfway through the fact check.
“The British Medical Journal has reported that pain can be felt only once the baby is born and not when still in the womb as a fetus.”
The claim from the British Medical Journal, unlike the one from the Journal of the American Medical Association, suitably matches the conclusion of the fact check. But why did LogicallyAI trust that source over the other one?
“The research mentioned that a fetus cannot feel pain unless stimulated into wakeful activity by separating it from the placenta.”
This line fails to answer our question, but it gives us something specific to look for in the cited research. We visited the website linked in the fact check’s source list. Our landing place lacked any use of the word “placenta.” Still, it was arguably supportive of the fact check’s conclusion:
The paper concludes that the basic physical mechanisms we need to feel pain develop in a fetus from about the 26th week of pregnancy. Peripheral free nerve endings, which act as sensors for pain, reach full maturity between 23 and 25 weeks, and form a complete link with the thalamus and cortex by about 26 weeks. Around the same stage the thalamus and cortex develop important features of maturity.
The article argues that this biological response to a noxious or potentially dangerous stimulus, which is produced by almost all animals, is not sufficient for the experience of pain.
We judge the 2006 article far less certain about fetal pain than the fact check suggests. Note “The article argues” and not “The evidence shows.” The deck of the article reinforces the tentative nature of its argument: “The idea that a fetus can feel pain is not supported by evidence according to a new clinical review published in the British Medical Journal.”
LogicallyAI (illogically) uses a fallacious appeal to ignorance. Lack of evidence a fetus experiences pain does not lead to the sure conclusion a fetus does not experience pain.
Yet the fact check’s conclusion was unequivocal.
More Nails for the Coffin?
As mentioned above, the fact check cited only three sources. We’ve touched on two of them, and the third was The New York Times, apparently in support of the fact check’s mention of the political climate surrounding abortion in the United States. The two journal citations were somewhat dated (2005 and 2006), prompting us to wonder whether they (particularly the British citation) represented mainstream scholarship.
One of our earliest search engine hits was Reconsidering fetal pain, a 2020 paper published in the British Medical Journal. Yes, that’s the same British Medical Journal that published the 2006 paper expressing doubt about the experience of fetal pain.
Embarrassingly for LogicallyAI, the 2020 paper was co-authored by the author of the 2006 paper, Stuart Derbyshire. In the 2020 paper, the authors recount evidence undercutting the basis for the 2006 paper’s conclusion. The 2020 paper directly undercuts the conclusion of the 2006 paper:
Arguably, there never was a consensus that fetal pain is not possible before 24 weeks. Many papers discussing fetal pain have speculated a lower limit for fetal pain under 20 weeks’ gestation.22–25 We note in passing that vote counting and consensus is perhaps not the best way to decide scientific disputes. Regardless of whether there ever was a consensus, however, it is now clear that the consensus is no longer tenable.
Why would a fact checker rely on a 2006 article when that article’s author reversed position in a journal article published in 2020? It makes no sense in terms of competent fact-checking. But it may represent the type of fact-checking error found in AI-guided fact-checking.
“Scientific research has proven that the pain response begins only after a baby is born, and hence we mark the claim as false.”
LogicallyAI’s conclusion flies in the face of the evidence. Science cannot prove anything based on an absence of evidence. The line represents and promulgates a basic misunderstanding of the scientific method. And that’s without even considering how the fact check ignored later evidence countering its conclusion.
Regret the Error?
Given the enormity of the error and our ongoing effort to track the International Fact-Checking Network’s accountability measures, we elected to submit a correction request to LogicallyAI.
The image to the left shows the filled-in complaint form. The image to the right shows confirmation of the submitted form. Though the form requires an email address, LogicallyAI sends no email confirmation from what we can tell.
The correction request gives us another specific test of the IFCN accountability system if LogicallyAI fails to live up to its policy on corrections.
If you notice something published through our journalistic, research, or educational function which you believe to be inaccurate, misleading, or unfair, please submit a correction request through the form at the end of the page. Any complaints will be raised to our editorial leads and responded to within 48 hours. Any complaint found to have substantive merit will be publicly corrected, and the correction will be given equal prominence to the article in question. We offer anybody who was the subject of criticism in our reporting the right to reply, provided a prima facie case can be made that our criticisms can be fairly addressed. Reasons for any refusal to grant a right to reply will be published on our website.
Promising a response without 48 hours seems ambitious. Our screen capture of the complaint submission confirmation has a time stamp of 1:32 p.m. on Aug. 14, 2023.
The clock is ticking. We will update this item with our observations of LogicallyAI’s response.
Updated shortly after publishing to fix some dodgy formatting (our last two lines were swallowed by the preceding quotation formatting). Hat tip to Media Bias Fact Check’s own Dave Van Zandt for pointing out we neglected to link to the fact check we reviewed. That’s fixed with this Aug. 15, 2023 update.
Update 4:07 p.m. Aug. 16, 2023: Here’s a link to a very recently archived version of the flawed fact check. We submitted the correction request over 48 hours ago, and we see no change to the story and no explanation for declining to fix the story either on the website or in our email inbox. We’ll have more to say after more than 72 hours have elapsed.
“Any complaints will be raised to our editorial leads and responded to within 48 hours.”
Update Aug. 24, 2023: Though Logically claims it will offer a response to any complaint within 48 hours, Zebra Fact Check has received no communication from Logically, and the flawed fact check appears as it did when we first encountered it. We took this update as an opportunity to fix a typo: “submt”=>”submit”