Presentation
DH6 - Effects of Neurodivergence on Deepfake-Video Detection: Mild Cognitive Impairment
DescriptionMisinformation threatens human well-being. Improving the safety of the engineered systems that afford the production and distribution of misinformative content is a task for which the human factors research/practitioner community is well-equipped to address. These risks should especially be attended to by the human factors researchers/practitioners who specialize in the healthcare domain. A misinformed patient may have been led to believe it is in their best interest to reject lifesaving treatment. A misinformed patient may engage in unhealthy behavior at the behest of those who would seek to deceive them. Even more direct harm, such as depression that arises from intentional misinformation that is designed to deplete morale, is an eventuality in today's information environment. These are healthcare issues. These are design issues. These issues are human factors issues.
The type of potentially misinformative technology that is of primary concern to the proposed work is that of the so-called, deepfake video. The term "deepfake" refers to the situation in which deep machine learning techniques are used to create media (in the present context, video) in which a typically human subject is digitally manipulated such that they appear to do, or say, something they never did. In fact, the faked human need not even share likeness with any real person. This technology has already been used to spread harmful misinformation in a political context (Reuters, 2023) and even in a wartime context (Allyn, 2022). The following is not unique to deepfake-style misinformation but, the potential that certain health circumstances could make one
That misinformative technology is a healthcare issue is even further evidenced by the fact that certain clinical diagnoses might be associated with increased susceptibility to being misinformed. Tidler and Catrambone (2021; and another, currently under-review report), have shown that a sample of participant who self-reported having been formally diagnosed with autism spectrum disorder (ASD) displayed greater confidence in their ability to discern deepfake videos from authentic videos then neurotypical counterparts. While confidence in ones judgement is certainly not equivalent to unreliability in ones judgement, overconfidence in ones judgement could, in this context, lead to an inappropriate degree of credulity when one has judged an information source to be authentic. For example, two people might agree that a video they encounter appears to be authentic (not a deepfake), while not having the same degree of certainty in their judgement. If both are then confronted with an opportunity to make some critical decision based on the information that is ostensibly provided by the video, then presumably the evaluator who is less certain of his/her judgement of authenticity will discount the informativity of the video during the decision calculus.
The work that is currently proposed to be presented was an effort to extend the previous, ASD-related findings to see if other neurodivergent populations are similarly different from the neurotypical population in terms of their behavior when attempting to distinguish between deepfake and non-deepfake videos. The instantiation of neurodivergence included in the to be presented study was, what is known as, mild cognitive impairment (MCI). MCI is a condition in which cognitive abilities (e.g., memory) are impaired to varying degrees of subtlety but not to the extent that ones ability to function in their normal activities is substantially diminished (Peterson et al, 1999).
METHOD
To that end, participants (N = 56) were shown a series of videos that included deepfakes and non-deepfakes. For each video the participants were asked to identify the authenticity of each video and to provide a rating of their confidence in their judgement. Three measures were used to assess performance:
- Raw performance scores - which was simply the number of correctly identified videos minus the number of incorrectly defined videos.
- Total confidence - the sum of the confidence ratings that participants provided after judging each video.
- Weighted performance scores - This variable was computed by multiplying the participants raw judgement score for each video (correct = 1, incorrect = -1) by the associated confidence rating for each video, and summing all of the weighted scores.
Thirty of the participants self-reported having been formally diagnosed with MCI and the other twenty-six of self-reported not having MCI.
RESULTS/DISCUSSION/CONCLUSIONS
When compared to the non-MCI control group, the subset of participants who self-reported having been diagnosed with MCI did not perform significantly differently in terms of their raw performance scores or their weighted performance scores. However, the self-reported MCI diagnosis condition was observed to be significantly more confident in their judgements than the control condition (t(54) = 2.427, p = .019).
Interestingly, this is the same pattern of results that was observed in its predecessor study (under review) in which self-reported ASD diagnosed participants were compared to neurotypical participants. Of course, we make no suggestion that there is some shared cognitive mechanism between ASD and MCI but the concern that some neurodivergent populations might be displaying some degree of overconfidence in their judgement of the authenticity of videos is bolstered by these findings. Of course, the pattern of results could just as easily be construed as underconfidence among the neurotypical population. At the very least, these results reinforce the often discussed but often unrealized position that the so-called "generalized human mind" is a poor model upon which to build principles of design. Individual variation in psychological profiles, especially when that variation reaches clinical thresholds, must be given proper consideration.
REFERENCES
Allyn, B. (2022, March 16). Deepfake video of Zelenskyy could be 'tip of the iceberg' in info war, experts warn. NPR. https://www.npr.org/2022/03/16/1087062648/deepfake-video-zelenskyy-experts-war-manipulation-ukraine-russia
Petersen, R. C., Smith, G. E., Waring, S. C., Ivnik, R. J., Tangalos, E. G., & Kokmen, E. (1999). Mild cognitive impairment: clinical characterization and outcome. Archives of neurology, 56(3), 303-308.
Reuters (2023, April 28). Fact Check-Video features deepfakes of Nancy Pelosi, Alexandria Ocasio-Cortez and Joe Biden. Reuters. https://www.reuters.com/article/factcheck-pelosi-deepfake-idUSL1N36V2E0
Tidler, Z. R., & Catrambone, R. (2021). Individual Differences in Deepfake Detection: Mindblindness and Political Orientation. In Proceedings of the Annual Meeting of the Cognitive Science Society (Vol. 43, No. 43).
The type of potentially misinformative technology that is of primary concern to the proposed work is that of the so-called, deepfake video. The term "deepfake" refers to the situation in which deep machine learning techniques are used to create media (in the present context, video) in which a typically human subject is digitally manipulated such that they appear to do, or say, something they never did. In fact, the faked human need not even share likeness with any real person. This technology has already been used to spread harmful misinformation in a political context (Reuters, 2023) and even in a wartime context (Allyn, 2022). The following is not unique to deepfake-style misinformation but, the potential that certain health circumstances could make one
That misinformative technology is a healthcare issue is even further evidenced by the fact that certain clinical diagnoses might be associated with increased susceptibility to being misinformed. Tidler and Catrambone (2021; and another, currently under-review report), have shown that a sample of participant who self-reported having been formally diagnosed with autism spectrum disorder (ASD) displayed greater confidence in their ability to discern deepfake videos from authentic videos then neurotypical counterparts. While confidence in ones judgement is certainly not equivalent to unreliability in ones judgement, overconfidence in ones judgement could, in this context, lead to an inappropriate degree of credulity when one has judged an information source to be authentic. For example, two people might agree that a video they encounter appears to be authentic (not a deepfake), while not having the same degree of certainty in their judgement. If both are then confronted with an opportunity to make some critical decision based on the information that is ostensibly provided by the video, then presumably the evaluator who is less certain of his/her judgement of authenticity will discount the informativity of the video during the decision calculus.
The work that is currently proposed to be presented was an effort to extend the previous, ASD-related findings to see if other neurodivergent populations are similarly different from the neurotypical population in terms of their behavior when attempting to distinguish between deepfake and non-deepfake videos. The instantiation of neurodivergence included in the to be presented study was, what is known as, mild cognitive impairment (MCI). MCI is a condition in which cognitive abilities (e.g., memory) are impaired to varying degrees of subtlety but not to the extent that ones ability to function in their normal activities is substantially diminished (Peterson et al, 1999).
METHOD
To that end, participants (N = 56) were shown a series of videos that included deepfakes and non-deepfakes. For each video the participants were asked to identify the authenticity of each video and to provide a rating of their confidence in their judgement. Three measures were used to assess performance:
- Raw performance scores - which was simply the number of correctly identified videos minus the number of incorrectly defined videos.
- Total confidence - the sum of the confidence ratings that participants provided after judging each video.
- Weighted performance scores - This variable was computed by multiplying the participants raw judgement score for each video (correct = 1, incorrect = -1) by the associated confidence rating for each video, and summing all of the weighted scores.
Thirty of the participants self-reported having been formally diagnosed with MCI and the other twenty-six of self-reported not having MCI.
RESULTS/DISCUSSION/CONCLUSIONS
When compared to the non-MCI control group, the subset of participants who self-reported having been diagnosed with MCI did not perform significantly differently in terms of their raw performance scores or their weighted performance scores. However, the self-reported MCI diagnosis condition was observed to be significantly more confident in their judgements than the control condition (t(54) = 2.427, p = .019).
Interestingly, this is the same pattern of results that was observed in its predecessor study (under review) in which self-reported ASD diagnosed participants were compared to neurotypical participants. Of course, we make no suggestion that there is some shared cognitive mechanism between ASD and MCI but the concern that some neurodivergent populations might be displaying some degree of overconfidence in their judgement of the authenticity of videos is bolstered by these findings. Of course, the pattern of results could just as easily be construed as underconfidence among the neurotypical population. At the very least, these results reinforce the often discussed but often unrealized position that the so-called "generalized human mind" is a poor model upon which to build principles of design. Individual variation in psychological profiles, especially when that variation reaches clinical thresholds, must be given proper consideration.
REFERENCES
Allyn, B. (2022, March 16). Deepfake video of Zelenskyy could be 'tip of the iceberg' in info war, experts warn. NPR. https://www.npr.org/2022/03/16/1087062648/deepfake-video-zelenskyy-experts-war-manipulation-ukraine-russia
Petersen, R. C., Smith, G. E., Waring, S. C., Ivnik, R. J., Tangalos, E. G., & Kokmen, E. (1999). Mild cognitive impairment: clinical characterization and outcome. Archives of neurology, 56(3), 303-308.
Reuters (2023, April 28). Fact Check-Video features deepfakes of Nancy Pelosi, Alexandria Ocasio-Cortez and Joe Biden. Reuters. https://www.reuters.com/article/factcheck-pelosi-deepfake-idUSL1N36V2E0
Tidler, Z. R., & Catrambone, R. (2021). Individual Differences in Deepfake Detection: Mindblindness and Political Orientation. In Proceedings of the Annual Meeting of the Cognitive Science Society (Vol. 43, No. 43).
Event Type
Poster Presentation
TimeTuesday, March 264:45pm - 6:15pm CDT
LocationSalon C
Digital Health
Simulation and Education
Hospital Environments
Medical and Drug Delivery Devices
Patient Safety Research and Initiatives