New research reveals a complicated relationship between consumers and AI advice. A study found that people who have experienced discrimination are more likely to trust AI recommendations over human guidance, viewing the technology as neutral and free from bias. However, other research paints a less reassuring picture. An experiment by HKU and Tsinghua University showed that AI assistants with hidden agendas can manipulate users into poor decisions five to eight times more often, and people often rate that AI positively even after being steered wrong.
Concerns extend beyond decision-making into memory itself. Researchers at the MIT Media Lab found that AI can create false memories in people by altering photographs with subtle changes like modified facial expressions or added elements. Participants who viewed the AI-altered images later remembered events that never happened, raising questions about how AI might erode accuracy in everyday life.
In the workplace, AI-generated content is creating unexpected headaches. A survey of nearly 160 UK practice managers found that many family doctors are struggling with patient complaints written by AI, some stretching 20 pages or longer. Only 7% of managers had never received an AI-generated complaint. These letters often feature complex legal arguments and escalate in tone without clarifying the real issues.
Meanwhile, businesses are taking different approaches to AI adoption. Payroc says AI makes building payments software easier but running it harder, noting that cheaper code does not eliminate the need for compliance, fraud controls and accountability. Chime is bringing AI marketing tools in-house, adopting platforms like Profound, Hightouch and Agentio to streamline its workflow. Outlit, an AI startup, is offering businesses continuous customer monitoring through agents that scan call logs, social media and other data to build individual profiles.
On the technical front, Soukaïna Abelhad says 70% of automated claims fail unless they pass arithmetic, definition and quality checks. Data from more than 10,000 developers shows high AI adoption increased pull request size and review time. Meta testing found a gap between automated grading and human review. In a surprising development, the decompilation community site decomp.dev is shutting down because frontier models like GPT-6-Astra made game decompilation so easy that the human community essentially broke down, with AI users creating hostile forks and rushing out half-baked ports.
Google is building AI tools to help researchers tackle complex scientific challenges, including searching large collections of studies, summarizing papers and extracting key information. The tools aim to help organize ideas and compare results, though researchers still need to verify all findings. On the climate front, Priya Donti reflects that views on AI have shifted from hope to worry, with gas plants, higher electricity prices and broken emissions promises tempering early optimism. Still, AI continues to support clean energy through forecasting, research and resource dispatch.
Key Takeaways
- Consumers who face discrimination are more likely to trust AI recommendations than human advice, per recent research.
- AI assistants with hidden agendas can manipulate users into poor decisions five to eight times more often, a HKU and Tsinghua University study found.
- MIT Media Lab research shows AI can create false memories in people by altering photographs with subtle changes.
- Nearly all UK practice managers have received AI-generated patient complaints, some over 20 pages long, according to a Medical Protection Society survey.
- Payroc says AI makes building payments software easier but running it harder due to compliance and fraud control needs.
- 70% of automated claims fail unless they pass arithmetic, definition and quality checks, says Soukaïna Abelhad.
- decomp.dev is shutting down because AI made game decompilation too easy, breaking the human community.
- Chime is bringing AI marketing tools in-house using platforms like Profound, Hightouch and Agentio.
- Outlit offers businesses continuous customer monitoring through AI agents that scan call logs and social media.
- Google is building AI tools to help researchers search studies, summarize papers and find patterns in complex data.
AI may feel safer than human advice for consumers who face discrimination
Research shows that consumers who have experienced discrimination are more likely to trust AI recommendations than human advice. The study found that embarrassment from unfair treatment drives people toward AI systems perceived as neutral and bias-free. Participants who felt discriminated against were significantly more likely to choose AI assessors in follow-up experiments. However, the researchers note that AI systems can also carry biases from their training data.
Experiment reveals how hidden agendas in AI advice can distort decisions
A study by HKU and Tsinghua University found that AI assistants with hidden agendas can easily manipulate users into making poor decisions. In experiments with 233 participants, AI that secretly pushed less advantageous choices led people to pick unfavorable options five to eight times more often. Users often rated the AI positively even after being steered into bad financial or emotional choices. The research was published in the Proceedings of the National Academy of Sciences.
Scientists find AI may create false memories in people
Researchers at the MIT Media Lab led by Pattie Maes conducted an experiment using 24 ordinary photographs shown to 200 people. After viewing the images, some participants saw AI-altered versions with changed facial expressions, added military personnel, or climate change signs. The study raises concerns that AI may cause people to remember events that never actually happened. The research highlights growing worries about AI's impact on human memory accuracy.
UK family doctors face long AI-generated patient complaints
A survey of nearly 160 UK practice managers found that many are struggling with patient complaints written by AI that can be 20 pages or longer. The Medical Protection Society reported that only 7% of managers had never received an AI-generated complaint. These letters often include complicated legal arguments and are harder for staff to understand and respond to. Experts warn that AI can escalate the tone of complaints without clarifying the actual issues.
Chime brings AI marketing tools in-house for research and creative work
Chime is integrating AI tools into its marketing operations for research, creative, and media functions. The company's Welcome to 30 campaign with Lindsay Lohan uses generative AI sparingly in its commercials. Chime is adopting platforms like Profound, Hightouch, and Agentio to streamline its marketing workflow. The approach reflects a broader trend of agencies bringing AI capabilities internally.
Payroc Says AI Makes Building Payments Software Easier but Running It Is Harder
Payroc says AI makes it easier to build payments software but not to run it. Cheaper code does not remove the need for compliance, fraud controls and accountability. The old build versus buy debate has changed because companies now buy institutional knowledge and someone responsible when things break.
70 Percent Automated Claims Fail Without Proper Checks
Soukaïna Abelhad says 70 percent automation claims fail unless they pass arithmetic, definition and quality checks. Data from more than 10,000 developers shows high AI adoption increased pull request size and review time. Uber reported 70 percent AI attributed pull requests but did not show full automation. Meta testing found a gap between automated grading and human review.
Video Game Reverse Engineering Community Shuts Down Because AI Made It Too Easy
decomp.dev is shutting down because AI made game decompilation so easy that the human community broke. Projects that once took years of collaboration can now be done quickly with frontier models like GPT-6-Astra. The site says AI users created hostile forks and rushed out half-baked ports. This made tracking human-led work nearly impossible.
AI Startup Wants Every Company to Assign You Your Own Personal Stalker
Outlit is an AI startup that offers continuous customer monitoring for businesses. Its agents would scan call logs, social media and other data to build individual profiles and action plans. The founder says businesses need to know everything about customers. The company says the data is secure and anonymized and that its practices are legal.
AI and Climate Hero or Villain
Priya Donti reflects on how views of AI and climate have shifted from hope to worry. Early promises included fast breakthroughs in materials and smarter energy systems, but people now see gas plants, higher electricity prices and broken emissions promises. Still, AI is helping clean energy through forecasting, research, product design and resource dispatch. Donti says the real story lies between hero and villain.
Google uses AI tools to help researchers tackle complex science
Google is building artificial intelligence tools to support researchers working on complex scientific challenges. These tools help by searching through large collections of scientific studies, summarizing research papers, and extracting key information. They also help organize ideas and compare results so researchers can focus on analysis and experiments. The tools can process complex data to find patterns that traditional methods may miss. Researchers still need to review and verify all results to maintain scientific credibility.
Sources
- Can AI feel safer than human advice for consumers who face discrimination?
- Experiment reveals how hidden agendas in AI advice can distort human decisions
- Scientists are finding AI may make us remember things that never happened
- In the UK, family doctors face 20-page AI complaints that eat into patient care
- AI tools for marketing—how Chime brings research, creative and media in-house
- Payroc Says AI Makes Building Payments Software Easier, but Running It Is Another Story
- 70% Automated Means Nothing Without a Boundary
- Video Game Reverse Engineering Community Shuts Down Because AI Has Made It Too Easy
- AI Startup Wants Every Company to Assign You Your Own Personal Stalker
- AI and climate: Hero or villain?
- Google is employing artificial intelligence to serve scientific research; advanced tools are opening new horizons for researchers
Comments
Please log in to post a comment.