Torleif Markussen Lunde

Torleif Markussen Lunde

Auto-generated profile, based on publicly available information

Torleif Markussen Lunde

Vestlandets Innovasjonsselskap, Postboks, Bergen, Norway; University of Bergen, Bergen, Norway

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 February 2026 in medRxiv
Verified
Preprint
Open access
PDF file
JAJanan Arslan
Janan Arslan
KBKurt Benke
Kurt Benke
SOSebastián Andrés Cajas Ordóñez
Sebastián Andrés Cajas Ordóñez
RCRowell Castro
Rowell Castro
LCLeo Anthony Celi
Leo Anthony Celi
PVPritika Vig
Pritika Vig
Background: We present BODHI (Balanced, Open-minded, Diagnostic, Humble, and Inquisitive), an engineering framework for curiosity-driven and humble clinical decision support AI. Despite growing capabilities, large language models (LLMs) often express inappropriate confidence, conflating statistical pattern recognition with genuine medical understanding. BODHI addresses this through a dual-reflective architecture that: (1) decomposes epistemic uncertainty into task-specific dimensions, and (2) constrains model responses using virtue-based stance rules derived from a Virtue Activation Matrix. Methods: We validate the framework through controlled evaluation on 200 clinical vignettes from HealthBench Hard, assessing GPT-4o-mini and GPT-4.1-mini across 5 random seeds (1,800 total observations). Statistical analysis included bootstrap resampling, paired t-tests, and effect size computation (Supplementary Materials S3) Findings: BODHI significantly improved overall clinical response quality (GPT-4.1-mini: +17.3pp, p < 0.0001, Cohen’s d = 0.50; GPT-4o-mini: +7.4pp, p < 0.0001, Cohen’s d = 0.22) while achieving very large effect sizes on curiosity (context-seeking rate: Cohen’s d = 16.38 and 19.54) and humility (hedging: d = 5.80 for GPT-4.1-mini) metrics. Crucially, 97.3% of GPT-4.1-mini responses and 73.5% of GPT-4o-mini responses included appropriate clarifying questions, compared to 7.8% and 0.0% at baseline, demonstrating the framework’s effectiveness in eliciting information-gathering behavior. Interpretation: These findings suggest LLMs can be reliably constrained to operate within epistemic boundaries when provided with structured uncertainty decomposition and virtue-aligned response rules, offering a pathway toward safer clinical AI deployment.

7 August 2026 in MIT Science Policy Review
Verified
Journal Article
Closed access
RARawan Abulibdeh
Rawan Abulibdeh
JAJanan Arslan
Janan Arslan
SOSebastián Andrés Cajas Ordóñez
Sebastián Andrés Cajas Ordóñez
LCLeo Anthony Celi
Leo Anthony Celi
TLTorleif Markussen Lunde
Torleif Markussen Lunde
Generative artificial intelligence (AI) is rapidly transforming how scientific knowledge is produced, reviewed, and disseminated. In response, journals and publishing organizations have begun issuing policies to govern AI use in scholarly publishing. However, it remains unclear whether existing governance frameworks meaningfully address the risks AI introduces across the full publication pipeline. We conducted a narrative review of journal policies, publisher guidance, and recent analyses of AI governance in scientific publishing, complemented by direct examination of submission guidelines from high-impact, open-access, and regional medical journals. Our findings show that while most journals have converged on a narrow legal consensus (prohibiting AI authorship and requiring disclosure), yet governance remains fragmented and incomplete. Policies disproportionately target text generation by authors, while leaving critical domains under-regulated, including AI-assisted data analysis, peer review practices, enforcement mechanisms, and equity implications for researchers and reviewers globally. To synthesize these findings, we introduce the AI governance readiness levels, a five-level framework for assessing how well-equipped journals are to govern AI across the research and publication process. We further describe PRAIDE (Preparation, Representation, Attribution, Integrity checks, Dissemination, and Evaluation) as an illustrative architecture that integrates existing policies, integrity safeguards, and post-publication oversight into a coherent governance model. We argue that effective AI governance in scientific publishing cannot be achieved through static rules or journal-centric control alone. Instead, it requires a shift toward shared, adaptive oversight of the scientific publishing system, recognizing that responsibility for governing AI in science is collective, continuous, and inseparable from the public trust in research.

29 July 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
SCSebastian Cajas
Sebastian Cajas
KDKacper Dobek
Kacper Dobek
KKKrzysztof Krawiec
Krzysztof Krawiec
TLTorleif Markussen Lunde
Torleif Markussen Lunde
RARawan Abulibdeh
Rawan Abulibdeh
LCLeo Anthony Celi
Leo Anthony Celi
Creativity requires both novelty and contextual appropriateness, properties that depend on the capacity to oscillate between generative exploration and evaluative constraint. This review argues that engineering creativity in human-AI systems calls for an explicit architectural separation between these two modes. Following Boden's taxonomy, we define creativity as combinational, exploratory, or transformational novelty evaluated relative to a domain or community, distinguishing it from autonomous computational novelty and anchoring the analysis at the system level. Through a PRISMA-guided systematic review of 1,472 records (43 included studies) spanning neurosymbolic AI, cognitive architectures, psychedelic neuroscience, uncertainty quantification, and ensemble methods, we identify converging support for a dual-process architecture. We propose a dual-engine framework coupling an uncertainty-guided divergent engine (drawing on psychedelic-inspired models of prior relaxation) with a neurosymbolic convergent layer. Human collaborators are integrated throughout as active co-participants rather than terminal evaluators. This reframes creativity as a relational property of the human-AI dyad.