How do decision-makers respond to positive versus negative advice from artificial intelligence?

Host: Welcome to A.I.S. Insights — powered by Living Knowledge. I am Anna Ivy Summers.

Expert: And I am Alex Ian Sutherland. Today, we are exploring a fascinating new study titled "Asymmetric Algorithm Aversion."

Host: It is a study that dives deep into how human decision-makers interact with artificial intelligence, particularly in high-stakes financial environments like evaluating Initial Coin Offerings, or ICOs. Alex, we often hear that people either blindly trust AI or completely reject it. But this study reveals something far more nuanced, doesn't it?

Expert: Absolutely, Anna. The core question here is not just whether professionals trust AI, but how their reliance shifts depending on what the AI actually recommends. Most prior research treated algorithm aversion as an all-or-nothing phenomenon based on user personality or task complexity. What this study uncovers is that the exact same expert, looking at the exact same task, responds completely differently depending on whether the algorithm delivers a green light or a red flag.

Host: That brings us right to the big problem the researchers set out to solve. In fields like venture capital and fintech, making a decision requires two distinct types of knowledge. Can you unpack that distinction for us?

Expert: Certainly. First, you have explicit knowledge. This is codifiable, structured data—like financial metrics, whitepaper readability, or source code activity. AI is widely acknowledged to excel at processing this kind of information. But investment decisions also demand tacit knowledge. That includes nuanced, subjective factors, like whether an executive team has genuine passion, founder integrity, or the resilience to execute under pressure. Humans generally believe algorithms cannot evaluate tacit signals. This skepticism creates a critical dilemma when AI offers advice on complex decisions.

Host: So if people already doubt that an AI can understand the human, tacit side of business, how did the authors investigate their actual behavior? What was the approach?

Expert: The authors took a rigorous, multi-method approach. They conducted two randomized controlled experiments, paired them with a field study, and supported the findings with in-depth qualitative interviews with professional investors. In their experiments, participants evaluated real ICO projects and were exposed to either positive or negative AI recommendations. The researchers then tracked how much participants updated their evaluations and measured their underlying perceptions of the AI's accuracy and capabilities.

Host: And what did the data show? What were the key findings?

Expert: The central finding is what the authors coin "asymmetric algorithm aversion." When the AI issued a negative recommendation, advising against an investment, experts listened. They significantly lowered their project evaluations. But when the AI issued a positive recommendation, encouraging investment, experts largely discounted or ignored it.

Host: That seems counterintuitive at first glance. Why would experts trust an algorithm to tell them "no," but refuse to trust it when it says "yes"?

Expert: It comes down to decision logic. The study explains that rejection follows what decision theorists call "disjunctive logic." To reject an investment, you only need one fatal flaw—a single red flag, like a plagiarized whitepaper or poor financial ratios. Because red flags are typically explicit and data-driven, experts believe the AI is fully capable of identifying them. Therefore, negative AI advice is perceived as highly accurate and sufficient on its own.

Host: But approving an investment is a completely different story.

Expert: Exactly. Acceptance follows "conjunctive logic." For a project to succeed, everything must be right. It needs solid financials, clean technology, and strong leadership, market timing, and team dynamics. Since experts doubt AI's ability to assess those tacit factors, a positive AI recommendation is seen as incomplete. It does not convince the expert to invest; instead, it triggers an urge to independently verify all the tacit elements the AI might have missed.

Host: So a positive AI recommendation actually creates more work for the human, while a negative recommendation acts as an immediate shortcut.

Expert: That is spot on, Anna. A negative recommendation is seen as a definitive filter, while a positive recommendation is merely treated as an unconfirmed lead.

Host: Let's talk about why this matters for business. For executives and technology leaders investing millions into enterprise AI, what are the practical takeaways?

Expert: First, business leaders must stop assuming AI adoption will be uniform across different workflows. If you deploy AI as a risk management, compliance, or fraud detection tool—where the primary goal is finding red flags and saying "no"—users will adopt it rapidly and rely on it heavily.

Host: But what about when AI is used for growth initiatives, like selecting candidates in HR or picking acquisition targets?

Expert: That is where the friction occurs. In growth or selection tasks, AI recommendations will naturally be met with skepticism. System designers and managers need to design decision-support tools that openly address tacit criteria. For instance, rather than having an AI simply recommend "Hire this candidate" or "Invest in this asset," the system should explicitly state which explicit metrics were verified, acknowledge the tacit factors that still require human evaluation, and frame its output as a complement to human judgment.

Host: That reframes how organizations should structure human-in-the-loop workflows. Use the AI to eliminate the obvious non-starters quickly, and focus human cognitive energy on evaluating the holistic, tacit elements of the remaining opportunities.

Expert: Precisely. Recognizing this asymmetry prevents leaders from misdiagnosing why their teams might be resisting AI. It is often not an outright fear of technology, but a rational hesitation to trust an algorithm with things only human experience can truly evaluate.

Host: A brilliant perspective that every business leader navigating AI integration needs to keep in mind. Alex, thank you so much for breaking down this study for us today.

Expert: It was my pleasure, Anna.

Host: And to our listeners, thank you for tuning in to A.I.S. Insights — powered by Living Knowledge. Join us next time as we continue to unpack the research shaping the future of business and technology.