August 2023
Tinius Digest
Månedlige rapporter om endringer, trender og utviklinger i mediebransjen.
Om Tinius Digest
Tinius Digest gir deg en oversikt over rapporter om og analyser av utvikling i mediebransjen og publiseres en gang i måneden. Her er våre viktigste funn fra denne måneden.
Del gjerne innholdet med kollegaer og bruk det i møter og presentasjoner.
Innhold
- AI is better at proving they're humans—than humans
- ChatGPT is left-leaning
- Norwegians have mixed feelings about AI
- Women twice as vulnerable to AI-automation
- Half of immigrants in Norway avoid news
- UN: Hundreds of thousands forced into online scams
- Digital public spheres: more visible opinion opposition or polarization?
- Algorithms play a key role in shaping human interactions
AI is better at proving they're humans—than humans
Researchers from Microsoft, ETH Zurich, and the University of California, Irvine, have examined the effectiveness of CAPTCHA solutions, which confirm that an internet user is a human.
The study includes 1400 participants who solved 14000 CAPTCHA challenges.
Three main findings
1
AI outperform humans
AI bot services outperform humans in both solving time and accuracy. Human accuracy in solving CAPTCHAs varies between 50 percent and 85 percent. Bots have an accuracy between 85 percent and 100 percent, with the majority over 96 percent.
2
Age differences
There are significant age differences. Older participants took longer to solve CAPTCHAs. However, unlike previous studies, this study found no correlation between education level and solving time.
3
Abandonment rate
The study also examined how many users abandon tasks when confronted with a CAPTCHA. Abandonment rates ranged from 18 percent to 45 percent, depending on the context and compensation offered.
ChatGPT is left-leaning
Researchers at the University of East Anglia have studied political biases in the chatbot ChatGPT (GPT-3.5).
Four main findings
1
Political bias
The study robustly establishes that ChatGPT has a significant political bias. This bias leans towards the left side of the political spectrum. ChatGPT is biased towards Democrats in the United States, Lula in Brazil, and the Labour Party in the United Kingdom.
2
Not a mechanical result
The study contends that this bias is not merely a mechanical outcome of the algorithm but suggests an inherent form of bias. Various robustness tests, including placebo and profession-politics alignment tests, support this assertion.
3
Impersonation tests
When ChatGPT was asked to impersonate various political personas, its responses while impersonating Democrats were more closely aligned with its ‘default’ or non-impersonated responses. This implies that the algorithm’s ‘neutral’ stance tends to align more closely with Democratic or left-leaning views.
4
Amplifying existing biases
These findings are significant because they imply that the algorithm is not merely reflecting the data it has been trained on but may amplify existing biases. This has far-reaching implications, particularly given the growing role of AI in shaping public opinion and political discourse.
Norwegians have mixed feelings about AI
Ipsos Norway has published the results of a survey on artificial intelligence.
Four main findings
1
2 out of 3 'understands AI'
62 percent of Norwegians feel they have a good understanding of what artificial intelligence is. Men and people under 40 have a higher self-assessed knowledge. In a global context, Norwegian responses are more moderate than a worldwide average of 67 percent.
2
Mixed feelings
There are mixed feelings about AI in Norway. 52 percent feel nervous about AI-based products and services, while 38 percent express enthusiasm. These numbers vary by age and gender. The scepticism is lower than in many English-speaking countries.
3
Impact on the workforce
Few Norwegians believe that AI will have a direct negative impact on their job. 37 percent think their job will be affected within five years, which is lower than globally (57%). Only 15 percent believe AI will replace their job.
4
Cultural differences
There is significant variation in how AI is perceived globally. In Asia, the perception is more optimistic, focusing on economic growth. In the West, there is more focus on potential negative impacts like privacy and concentration of power.
Women twice as vulnerable to AI-automation
The UN’s International Labour Organization has published a global analysis of how jobs and tasks could be affected by generative AI, specifically Generative Pre-trained Transformers (GPTs).
Four main findings
1
Women more vulnerable
Women in high-income countries are more than twice as likely to lose their jobs due to automation (7.8%) compared to men (2.9%). The gender disparities arise because a higher percentage of tasks performed by women are at risk. The UN estimates that this affects a total of 21 million jobs.
2
Clerical work is highly susceptible
Only clerical work is highly susceptible, with 24 percent of tasks considered highly exposed and an additional 58 percent having medium-level exposure. For other occupational groups, the share of highly exposed tasks ranges between one and four percent, and tasks with medium exposure do not exceed 25 percent.
3
Automating work, not replacing jobs
Generative AI is more likely to augment work by automating some tasks within a job rather than fully automating entire occupations.
4
Variability across countries
The potential impact on employment varies widely across countries, based on their income levels. In low-income countries, only 0.4 percent of total employment is potentially exposed to automation. In contrast, in high-income countries, it’s 5.5 percent.
Half of immigrants in Norway avoid news
The Global Village has published a report on media consumption and trust in media in immigrant-dense areas in Oslo. The survey for the report was conducted on behalf of Järvaveckan Research and Norwegian IN/LAB.
Five main findings
1
News avoiders
Nearly half (46%) deliberately avoid news—the proportion is highest among women (52%) and individuals between 30 and 39 years old (54%). The proportion is nearly identical among Norwegians in general.
2
Unwanted and negative
The most important reasons for avoiding news are that the news is negative (23%), covers unwanted news events (19%), creates frustration (10%), is uninteresting (9%), and lack of trust (7%).
3
Lower subscription rate
37 percent of Norwegians in immigrant-dense areas in Oslo subscribe to a news source/media, which is lower compared to Norwegians in general (59%).
4
Lack of trust
61 percent believe journalists have a political agenda, and 46 percent believe journalists have a personal agenda. Only 23 percent believe journalists can write whatever they want, compared to 14 percent for Norwegians in general.
5
Muslim bias
35 percent believe that news reporting about Muslims contributes to a negative perception among the general public and/or in public debate.
UN: Hundreds of thousands forced into online scams
The UN High Commissioner for Human Rights has published a report examining the extent to which individuals are being forced into digital criminality in Southeast Asia.
Four main findings
1
Human trafficking
Organized criminal gangs are forcing hundreds of thousands of people in Southeast Asia into various forms of digital criminality, including romance scams, cryptocurrency fraud, and illegal gambling. Sources suggest that at least 120,000 individuals in Myanmar and around 100,000 in Cambodia may be victims of this form of human trafficking.
2
Mislabeled as criminals
Despite being coerced through violence and threats of violence, those carrying out the scams are categorized as criminals, rather than victims.
3
Mostly male victims
The majority of victims are men, many of whom are well-educated and multilingual.
4
Surge amid the pandemic
Increased digital activity in the West during the pandemic has contributed to a rise in the number of people being forced into online scam operations.
Digital public spheres: more visible opinion opposition or polarization?
The Norwegian Institute for Social Research has published an analysis of polarization and opinion dynamics in the digital public sphere.
Four main findings
1
Uncertainty surrounding polarization
There is still uncertainty about whether the digital public sphere creates, amplifies, or mitigates polarization.
2
Increased visibility of differences
Oppositions, pointed arguments, and negative characterizations are more visible in the digital public sphere than in earlier broadcast-based public spheres. This visibility can lead to an amplified perception that society is polarized, making the polarization more real.
3
Facilitation of echo chambers
Digitization has led to a fragmented public sphere, facilitating the formation of echo chambers. The network aspect in the digital public sphere enables mobilization around various viewpoints, leading to trench warfare dynamics.
4
Potential for positive outcomes
In the long term, exposure to diverse opinions in the digital public sphere may actually lead to learning and moderation of attitudes.
Algorithms play a key role in shaping human interactions
Researchers from Karolinska Institutet, Northwestern University, and Princeton University have delved into how social media algorithms impact how we learn from and interact with others.
Three main findings
1
Algorithmic influence on social learning
Social platforms shape human social learning via algorithms. Although human social learning has evolved to foster cooperation and collective problem-solving, algorithms prioritize increasing online engagement. This often results in social misconceptions and the spread of misinformation.
2
Inducing strong emotions
To maximize engagement, algorithms often showcase content that triggers potent emotions. Over time, repeated exposure to such content can skew one’s perception of reality.
3
Magnifying the reach of influencers
Users frequently gravitate towards admired figures like influencers or those from their perceived ‘in-group’. Social media algorithms augment the reach of these individuals, thus enhancing their influence over their audience. This can sometimes restrict followers from accessing diverse information on pivotal topics.