At Rise of AI 2023 in Berlin, Dr. Katharina von Knop, founder and CEO of Digital Trust Analytics, argues that fear of AI is a neuropsychological problem that has to be addressed through interface and system design, not just messaging. She explains that in-person eye contact activates mirror neurons that build trust, which is why video calls feel exhausting, and that trust itself is fragile, energy-saving for the brain, and must be earned continuously. She describes how the amygdala reacts to unfamiliar or overly complex information, such as AI, with the same threat response triggered by seeing a spider, and how most human decisions are made unconsciously by the brainstem rather than the prefrontal cortex, with the conscious mind processing only a small fraction of available information. She presents a Maslow-style layered model of what trustworthy AI needs beyond factual problem-solving, including cybersecurity, transparency and accessibility, and criticizes GDPR consent notices, using Instagram's roughly 4,683-word declaration as an example of a legally compliant text designed in a way that discourages people from reading it. She closes with a Q&A on decision-making, cultural differences in fear, and why people fear rare risks like AI more than common ones like medical malpractice.
Key takeaways
- Von Knop argues eye contact triggers mirror neurons that build trust, which is why video calls feel exhausting compared to in-person meetings.
- She says the amygdala reacts to complex, unfamiliar information such as AI with the same threat response triggered by seeing a spider.
- She states the brainstem drives about 98 percent of human decisions and cannot be influenced by the prefrontal cortex.
- She proposes a layered, Maslow-style model for trustworthy AI running from basic problem-solving up to cybersecurity, transparency and controllability.
- She cites Instagram's roughly 4,683-word GDPR declaration as an example of consent text designed in a way that discourages people from reading it.
Detailed notes
Why In-Person Contact Builds Trust
Von Knop opens by explaining why the audience came in person: eye contact activates a specific brain region and mirror neurons, causing people to synchronize eye blinks, which is how trust begins to form. She says humans are best at decoding mimic and body language, which is why video calls are exhausting, since the brain keeps trying and failing to decode facial and body cues through a camera and screen. She defines trust as an evolutionary mechanism that drives relationships and company performance, boosting innovation, efficiency and results while reducing direct and transaction costs and the number of mistakes. From a neuropsychological perspective, trust lets people waive a full risk-chances analysis by expecting a system to behave as anticipated and to serve their interests. She stresses that trust is highly fragile, never guaranteed, and needs to be earned continuously, which matters more now given geopolitical tension, cyberattacks and growing data intensity.
Fear, the Amygdala and Unconscious Decision-Making
She describes how a threat like a spider is processed twice: slowly through the visual cortex, which analyzes pixels, lines, shapes and eventually recognizes the object, and much faster through the amygdala, which triggers an immediate reaction such as jumping back. The amygdala is also highly active when people face information that is too complex to decode, which she says is always the case with AI, making people who are not fully in control of a system subconsciously afraid. She outlines the human decision-making process: senses (which she says may number anywhere from 6 to 21 depending on the source) feed information through filter databases shaped by the current situation, personal biography and early childhood experience, into the brainstem, which she says drives 98 percent of decisions and cannot be influenced by the prefrontal cortex. She cites a commonly used estimate that the brain processes a very large amount of sensory information unconsciously each second but only about 50 bits per second consciously, and notes that around 184 known biases further shape perception, filtering and decision-making.
A Layered Model for Trustworthy AI
Using a Maslow-style pyramid, she lays out what digital solutions need beyond the basic business model layer of solving a problem, saving time or scarce resources. The next layer for AI requires correct, non-biased data, transparency and efficient databases and processes. Above that sits what she calls the license to do business: cybersecurity, GDPR compliance, machine-readable norms, safety, ethics and data policy, followed by responsible AI and, critically, controllability and accessibility. She illustrates German risk-averse behavior with the example of panic-buying toilet paper at the start of the pandemic, arguing this reflects a historically shaped need for control that, if channeled well, is an advantage for building AI systems that emphasize reliability, transparency, predictability, accountability and auditability.
GDPR as a Failed Trust Example
She uses Reddit's and Instagram's GDPR consent notices as an example of how legal compliance can still fail to build trust, because the text is written in dense, technical language that costs too much brain energy to process, causing users to skip reading it; she cites Instagram's declaration at 4,683 words. She contrasts this with video, which the brain is evolutionarily well-suited to decode quickly. She closes her prepared remarks by listing entrepreneurial opportunities in trustworthy AI, including continuously automated GDPR review, automated quality and risk management systems, trust-by-design content management, AI ethics technologies, and tools to increase transparency in AI decision architectures.
Q&A Highlights
Asked how organizations can survive the next decade, she recommends structured decision-making: defining criteria in a spreadsheet and having team members respond anonymously, which strips out hierarchy bias and gives middle management a factual basis for innovation decisions. Asked whether fear is a good driver, she says a small amount of fear raises adrenaline and cortisol and helps the brain perform, describing her own morning routine of light hunger and exercise to trigger this. Discussing why societies fear different things, she contrasts Germany's historically rooted fear of losing control, shaped by experience with two dictatorships, with fear of burglary in China, and notes that people are often more afraid of rare risks like AI or radiation than common ones like medical malpractice, because fear responds to what feels uncontrollable rather than to actual statistical risk. On declining trust in science and progress, she says fear plays a legitimate role in driving progress when it is not allowed to block action, citing fear of falling behind the US and China in AI as an example of fear that can motivate rather than paralyze.
Quotes
Trust is highly fragile, never take it for granted, needs to be earned day by day.
Katharina von Knop
Our brain just 50 bits per second, this is not that much.
Katharina von Knop
Having a little bit of fear encourages the level of adrenaline and cortisol, having given us a little bit of fear makes our machine working.
Katharina von Knop
Questions answered in this talk
Why do video calls feel more exhausting than in-person meetings?
According to Katharina von Knop at Rise of AI 2023, the brain constantly tries to decode mimic and body language, which humans are best equipped to read in person, and this decoding fails through a camera and screen, making video calls draining.
Why are people afraid of AI even though it has rarely harmed them personally?
According to Katharina von Knop at Rise of AI 2023, the amygdala reacts to information that is too complex to fully understand, such as AI, with a threat response similar to fear of a spider, and people fear a lack of control over unfamiliar systems more than statistically common risks like medical malpractice.
Why do GDPR consent notices fail to build user trust?
According to Katharina von Knop at Rise of AI 2023, notices like Instagram's roughly 4,683-word GDPR declaration are written in dense, technical language that costs too much brain energy to process, so users tend to skip reading them entirely, which undermines rather than builds trust.