Tropical peatlands contain around one-sixth of the global peat carbon stock. Decomposition is a key determinant of tropical peat persistence, but there is a scarcity of data on decomposition in tropical peatlands. To further understand decomposition in tropical peatlands, we conducted an 8-year field experiment in a primary peat swamp forest in Brunei. We tracked mass loss and the organic matter composition of Shorea albida wood buried at multiple depths over 8 years, including blocks buried with and without termite exclusion mesh. The proportion of time wood blocks spent above the water table explained the majority of the variation in wood decomposition over time. Carbon loss from wood that spent <1% of the time under the water table was 32.1%-86.5% higher on average than from wood that spent 30%-100% of the time under the water table. We estimate that termites enhanced wood decomposition by similar to 2% per year. Despite significant decomposition, we did not observe a strong shift in wood organic matter composition. To contextualize our results, we synthesized past work on wood decomposition across tropical peatlands. We found that burial in waterlogged peat soils slows decomposition across tropical peatlands and that decomposition is also strongly influenced by peatland trophic status. Overall, our results affirm that waterlogging is the key to tropical peat persistence. Our study highlights the vulnerability of tropical peat carbon stocks to lowered water tables by either drainage or prolonged dry spells, as well as the promise of peatland rewetting to mitigate carbon losses from disturbed peatlands. Plain Language Summary Tropical peatlands store over 100 gigatonnes of carbon in their organic-rich peat soils. This soil carbon persists due to waterlogged conditions that discourage the decomposition (or breakdown) of organic plant material in peat soils. Wood is a large component of organic matter in tropical peat soils, but to date few studies have investigated the controls on wood decomposition in these ecosystems. We used an 8-year field experiment in which we buried wood from a native tree at many depths in a peat swamp in northem Borneo to assess how the proportion of time wood spends in waterlogged conditions impacts decomposition rates. We found that wood that spent at least part of the year under water decomposed >30% less than wood that was never under water. By compiling previous studies, we found that while decomposition is impacted by site-specific factors such as pH, burial in waterlogged peat soils reduces wood decomposition across the tropics. Our results underscore the vulnerability of tropical peat carbon to drainage and drought and support that efforts to restore degraded tropical peatlands by raising the groundwater level have the potential to reduce carbon emissions from peat decomposition.

Abstract Eusocial insects fascinate researchers with their sophisticated communication systems and sensory specialisations. Ants and termites have coexisted in a long-standing predator–prey arms race, offering insight into the interplay between ecology and evolution. The subterranean termite Coptotermes acinaciformis can detect the predatory ant Iridomyrmex purpureus through footstep-induced vibrations, triggering defensive responses. Ants produce noisier walking signatures than termites, while the inquiline termite Macrognathotermes sunteri walks more quietly than its host, suggesting species-specific vibroacoustic strategies. Using statistical analysis of video-tracked motion and footstep vibrations in confined arenas across six ant and ten termite species, we show that C. acinaciformis , despite its body size, moves more smoothly than ants, which alternate between directed and erratic paths. Inquiline termites, by contrast, displayed erratic movements. Ants consistently produced stronger vibrations closely linked to body mass, while Highly Comparative Time Series Analysis revealed termite motions approaching chaotic dynamics. Notably, while C. acinaciformis and I. purpureus produced distinct vibrational signatures, M. sunteri ’s signals overlapped with its host, consistent with vibroacoustic mimicry. Although the ecological nature of this association remains unresolved, our findings underscore the central role of vibrational cues in shaping interspecific dynamics and highlight vibroacoustic communication as an underappreciated driver of social insect ecology and evolution.

Eusocial insects like termites and ants use diverse communication methods, including pheromones, sound, and vibrations. Termites, blind and with fewer glands, rely heavily on vibrations for foraging, communication, and predator avoidance. Their appendages detect subtle signals amid noise, though the underlying physiological mechanisms remain largely unknown and understudied. We explore the role of termite legs and antennae as sensory probes. These appendages receive the vibration signals, which are detected by the leg's subgenual organ and the Johnston's organ in the antenna, and which in turn convert these mechanical environmental signals into nerve impulses sent to the nervous system. We compare these appendages in termites and ants, two eusocial, subterranean insect groups that share ecological traits but differ in trophic roles, with ants being major predators of termites. Termite legs and antennae have lower slenderness ratios (legs: 19-35 vs. 48; antennae: 23-32 vs. 61). Wasps and bees fall in between. Assuming similar material properties, termite legs likely have lower stiffness and higher natural frequencies, enhancing vibration sensitivity. The subgenual organ's position near the head may further improve detection. These morphological traits suggest termites could be better adapted for sensing a broader range of vibrations than ants. However, more specimens and species of Isoptera and Formica need to be tested to validate this claim fully. Comparing the legs of termites with ants, we found that termite tibiae amplify lower-frequency vibrations (~0-2.25 kHz), while ants show an amplification at higher frequencies (1.9-3.1 kHz). This suggests the vibrational sensitivity of termites is better adapted to wood-borne signals, which corresponds to their food, whereas ants, as generalist foragers, are tuned for diverse terrains, including light structures, such as twigs, leaves, and other plant matter. Considered together, our findings suggest that termite legs may function as an integrated auditory complex.

Tropical savannas typically experience high fire frequencies, with prescribed fire commonly used as a management tool. Termites play an important role in the ecological functioning of tropical savannas, yet we have a limited understanding of how fire affects these important ecosystem engineers. To account for the effects of fire management on ecosystem structure and function, we need to understand the links between fire management and termite communities. This study used a long-term (18-year) fire experiment in a tropical savanna near Darwin, northern Australia, to investigate the effects of different fire regimes on termite species composition, abundance and activity. We measured termite abundance and activity using a combination of baiting and reduced transect survey methods and compared these with fire activity (summarised fire frequency and intensity) and woody cover. Termite species richness was similar across all fire treatments, and the level of fire activity had a minimal effect on species composition, which was more strongly influenced by woody cover. Wood-feeding termite abundance and the consumption of wood baits were negatively correlated with fire activity and positively correlated with woody cover. Soil/wood interface-feeding termites showed no correlation with fire activity but a positive correlation with woody cover. Significant negative mediation effects of fire activity through woody cover were detected on the abundance of wood- and soil/wood interface feeders and wood and straw bait consumption. Grass-feeding termites were encountered too infrequently to draw conclusions about their correlation with fire activity and woody cover; however, straw bait consumption was positively correlated with fire activity. Synthesis and applications. The effects of fire on termite abundance and activity are primarily indirect, mediated through changes in vegetation structure. As high fire activity is associated with reduced woody cover, maintaining regimes of frequent, high-intensity fires over the long term has the potential to affect ecosystem function. While minimising the occurrence of high-intensity, late dry season fires is consistent with fire management goals in these savannas, care is still required to avoid the negative consequences of high fire frequencies.

Eusocial insects have attracted significant research interest due to their unique physiology and ability to detect and respond to various signals, enabling efficient and often highly specialised communication within their colonies. Ants and termites have been locked into a predator-prey arms race for millions of years. Recent research reveals that termites can detect predatory ants by sensing their footsteps. It is, therefore, of great interest to determine the characteristics and differences of the vibrations induced by the walking of ants and termites. Here, we conduct a statistical analysis of six ants and ten termite species' video motion and footstep response vibrations. The motion analysis reveals that although termites are as heavy and active as ants, they move less erratically and quickly fall into a pattern following each other. The median speed relative to the body length shows a moderate negative correlation in ants, but a weak correlation in termites. By isolating single-step vibration response using nonlinear filtering and an enveloping function, the average amplitude and decay times are determined. Compared with termites, ants are found to excite more harmonics and exhibit more power in their vibrations which correlates with their average body weight, the average response amplitudes, and decay times but this is not observed for termites.

Anthelmintic residues in livestock dung can adversely affect beneficial organisms. Targeted selective treatment (TST) of a reduced proportion of livestock with anthelmintics can slow resistance development in gastrointestinal nematodes by providing residue-free dung which could also benefit non-target organisms. We tested effects of TST on survival and reproduction of the dung beetle Onthophagus taurus (Scarabaeidae) in a factorial glasshouse experiment (Experimental treatments: five TST levels, 0.00, 0.25, 0.50, 0.75, 1.00 x four ivermectin concentrations, 125, 250, 375, 500 ppb). Each mesocosm comprised a 60 L bin containing sand, four dung pats and six pairs of adult beetles (F0 generation). No effects of TST level and ivermectin concentration on mortality of F0 adults after one week were observed. F0 adult brood ball production was affected by TST level, particularly at high ivermectin concentrations. Brood ball production increased as more untreated pats became available, with greater increases at higher ivermectin concentrations. We tested for evidence of a reported attraction of dung beetles to ivermectin-treated dung using a novel glitter-marker to trace the origin of dung used in brood balls. Where mesocosms contained both dung types, the proportion of brood balls created from untreated dung showed no statistical difference from the null expectation based on untreated dung availability in the mesocosm. Emergence of F1 adults was affected by the increase in TST, with this effect dependent on concentration. Treatments with concentrations of 250-500 ppb had the lowest emergence rates (ca. 5-20 % in mesocosms where all dung pats were treated) but emergence rates increased with TST level, reaching 68-88 % emergence where no dung pats were treated with ivermectin. Ivermectin-induced mortality occurred predominantly at egg and first instar stages. TST can provide refuges for dung beetles offering a strategy for livestock producers to maintain livestock welfare whilst benefiting from ecosystem services provided by important insects.

Since 1968, the Australian Dung Beetle Project has carried out field releases of 43 deliberately introduced dung beetle species for the biological control of livestock dung and dung-breeding pests. Of these, 23 species are known to have become established. For most of these species, sufficient time has elapsed for population expansion to fill the extent of their potential geographic range through both natural and human-assisted dispersal. Consequently, over the last 20 years, extensive efforts have been made to quantify the current distribution of these introduced dung beetles, as well as the seasonal and spatial variation in their activity levels. Much of these data and their associated metadata have remained unpublished, and they have not previously been synthesized into a cohesive dataset. Here, we collate and report data from the three largest dung beetle monitoring projects from 2001 to 2022. Together, these projects encompass data collected from across Australia, and include records for all 23 species of established dung beetles introduced for biocontrol purposes. In total, these data include 22,718 presence records and 213,538 absence records collected during 10,272 sampling events at 546 locations. Most presence records (97%) include abundance data. In total, 1,752,807 dung beetles were identified as part of these data. The distributional occurrence and abundance data can be used to explore questions such as factors influencing dung beetle species distributions, dung beetle biocontrol, and insect-mediated ecosystem services. These data are provided under a CC-BY-NC 4.0 license and users are encouraged to cite this data paper when using the data.

Termites are important ecosystem engineers in many ecosystems globally. Hence, surveys of termite species composition, abundance and activity can be important for understanding ecosystem function-especially in biomes where they tend to be abundant, such as tropical savannas. However, comprehensively surveying termites can be challenging due to their cryptic nature and varied feeding and nesting habits, which strongly influence the effectiveness of different survey methods. Baiting and active searches of reduced transects are two methods commonly used to sample termites, and while these methods have been evaluated in the savannas of South Africa, this has not occurred in the extensive tropical savannas of northern Australia. Thus, this study evaluated the effectiveness of baits and reduced transects to assess termite species richness and activity across 18 x 1 ha experimental plots in a tropical savanna near Darwin, Australia. Surveys in each plot consisted of two 60 x 2 m transects and a 9 x 3 baiting grid of alternating buried wood and paper baits and surface straw baits. Baits were checked three times: at 4-, 7- and 10-week intervals following placement. Upon survey completion, the sampling effort, efficacy and costs of each method were compared. Reduced transects detected all 32 species recorded in this study, representing four feeding groups (from undecayed wood to highly decayed organic material in the soil). Baiting detected 20 species, but failed to detect some of the species that fed on decayed materials. Paper baits, checked only twice (at 4 and 10 weeks following placement), were required to detect all species sampled at both wood and paper baits. Therefore, overall baiting costs could be reduced (without data loss) by using paper baits only and reducing the number of bait checks. Compared with baiting using all three bait types, reduced transects detected the most species and had the lowest per-species cost. Consequently, reduced transect surveys are the most effective method in these northern Australian savannas when assessing species composition. However, if the abundance of species that feed on undecayed wood or levels of termite activity are being assessed, then reduced baiting is a more appropriate method.

The origin of the German cockroach,Blattella germanica, is enigmatic, in part because it is ubiquitous worldwide in human-built structures but absent from any natural habitats. The first historical records of this species are from ca. 250 years ago (ya) from central Europe (hence its name). However, recent research suggests that the center of diversity of the genus is Asian, where its closest relatives are found. To solve this paradox, we sampled genome-wide markers of 281 cockroaches from 17 countries across six continents. We confirm thatB. germanicaevolved from the Asian cockroachBlattella asahinaiapproximately 2,100 ya, probably by adapting to human settlements in India or Myanmar. Our genomic analyses reconstructed two primary global spread routes, one older, westward route to the Middle East coinciding with various Islamic dynasties (~1,200 ya), and another younger eastward route coinciding with the European colonial period (~390 ya). While Europe was not central to the early domestication and spread of the German cockroach, European advances in long-distance transportation and temperature-controlled housing were likely important for the more recent global spread, increasing chances of successful dispersal to and establishment in new regions. The global genetic structure of German cockroaches further supports our model, as it generally aligns with geopolitical boundaries, suggesting regional bridgehead populations established following the advent of international commerce.

The ecosystem services provided by dung beetles are well known and valued. Dung beetles bury dung for feeding and breeding, and it is generally thought that the process of burying dung increases nutrient uptake by plant roots, which promotes plant growth. Many studies have tested the effects of dung beetles on plant growth, but there has been no quantitative synthesis of these studies. Here we use a multi-level meta-analysis to estimate the average effect of dung beetles on plant growth and investigate factors that moderate this effect. We identified 28 publications that investigated dung beetle effects on plant growth. Of these, 24 contained the minimum quantitative data necessary to include in a meta-analysis. Overall, we found that dung beetles increased plant growth by 17%; the 95% CI for possible values for the true increase in plant growth that were most compatible with our data, given our statistical model, ranged from 1% to 35%. We found evidence that the dung beetle-plant growth relationship is influenced by the plant measurement type and the number of beetles accessing the dung. However, beetles did not increase plant growth in all quantitative trials, as individual effect sizes ranged from -72% to 806%, suggesting important context-dependence in the provision of ecosystem services.

The higher classification of termites requires substantial revision as the Neoisoptera, the most diverse termite lineage, comprise many paraphyletic and polyphyletic higher taxa. Here, we produce an updated termite classification using genomic-scale analyses. We reconstruct phylogenies under diverse substitution models with ultraconserved elements analyzed as concatenated matrices or within the multi-species coalescence framework. Our classification is further supported by analyses controlling for rogue loci and taxa, and topological tests. We show that the Neoisoptera are composed of seven family-level monophyletic lineages, including the Heterotermitidae Froggatt, Psammotermitidae Holmgren, and Termitogetonidae Holmgren, raised from subfamilial rank. The species-rich Termitidae are composed of 18 subfamily-level monophyletic lineages, including the new subfamilies Crepititermitinae, Cylindrotermitinae, Forficulitermitinae, Neocapritermitinae, Protohamitermitinae, and Promirotermitinae; and the revived Amitermitinae Kemner, Microcerotermitinae Holmgren, and Mirocapritermitinae Kemner. Building an updated taxonomic classification on the foundation of unambiguously supported monophyletic lineages makes it highly resilient to potential destabilization caused by the future availability of novel phylogenetic markers and methods. The taxonomic stability is further guaranteed by the modularity of the new termite classification, designed to accommodate as-yet undescribed species with uncertain affinities to the herein delimited monophyletic lineages in the form of new families or subfamilies. Here, the authors produce an updated termite classification with genomic scale analyses, highlighting thirteen family-level lineages and resilience of their classification to future termite research.

1. Insect biological control agents may be released in relatively few locations; thus, the rate of spread is critical to predicting the spatial distribution of the agent.2. Exotic dung beetles have been introduced into Australia since the 1960s to bury dung, reduce bush fly populations, increase pasture productivity and improve livestock health. 3. We use a stochastic cellular automata model to predict the historical occupancy and abundance of introduced dung beetles in the South West agricultural region of Western Australia. The model predicts the spread of four species of dung beetles in six-month time steps for 10 years following initial release. Our model includes release locations, species-specific ecological parameters, dung resource availability and weather variables.4. The average rate of spread varied between species from a high of ca. 79 km/year (Euoniticellus intermedius) to a low of 28 km/year (Onthophagus taurus). Thus, after 10 years, the area of spread varied between 99,175 and 34,175 km(2) and abundance varied from 88,100 to 4,189 beetles/km(2).5. These findings provide an estimate of the spread patterns of dung beetles. The model can be used to guide future dung beetle release programmes in Australia and elsewhere.

The estimation of leaf litter turnover is often limited to early-stage decomposition using unrepresentative models and litter types. In tropical secondary forests, particularly exotic-dominated novel forests, the characterisation of litter turnover remains poor. This study estimated the annual turnover of in-situ leaf litter across four forest successional types in Singapore using a Weibull residence time model. Litter turnover and nutrient dynamics diverged between young secondary and old-growth forests. In particular, within novel forests, annual phosphorus return via leaf litterfall was three times that of primary forests, while the mass loss of in-situ leaf litter was highest among all forest successional types, estimated at 92.8% annually with a mean residence time of 176 days, resulting in a litter pool size a third that of primary forests. Our findings suggest that tree species composition and species-specific effects shaped the observed variations in litter turnover and nutrient dynamics across forest successional types and forest stands, whereas tree species richness, canopy structure, soil nutrient levels, and microclimate were found to be non-predictors. Taken together, our study provides an insight into litter turnover in human-modified tropical landscapes increasingly characterised by novel forests, potentially leading to a reduction in surface litter and soil organic carbon pools.

The current move towards digital pathology enables pathologists to use artificial intelligence (AI)‐based computer programmes for the advanced analysis of whole slide images. However, currently, the best‐performing AI algorithms for image analysis are deemed black boxes since it remains – even to their developers – often unclear why the algorithm delivered a particular result. Especially in medicine, a better understanding of algorithmic decisions is essential to avoid mistakes and adverse effects on patients. This review article aims to provide medical experts with insights on the issue of explainability in digital pathology. A short introduction to the relevant underlying core concepts of machine learning shall nurture the reader's understanding of why explainability is a specific issue in this field. Addressing this issue of explainability, the rapidly evolving research field of explainable AI (XAI) has developed many techniques and methods to make black‐box machine‐learning systems more transparent. These XAI methods are a first step towards making black‐box AI systems understandable by humans. However, we argue that an explanation interface must complement these explainable models to make their results useful to human stakeholders and achieve a high level of causability, i.e. a high level of causal understanding by the user. This is especially relevant in the medical field since explainability and causability play a crucial role also for compliance with regulatory requirements. We conclude by promoting the need for novel user interfaces for AI applications in pathology, which enable contextual understanding and allow the medical expert to ask interactive ‘what‐if’‐questions. In pathology, such user interfaces will not only be important to achieve a high level of causability. They will also be crucial for keeping the human‐in‐the‐loop and bringing medical experts' experience and conceptual knowledge to AI processes.

Automated image analysis and artificial intelligence (AI) are a growing market in digital pathology. While various proprietary pathology systems exist, there are no fully vendor-agnostic integration approaches for AI apps. This makes it difficult for vendors of AI solutions to integrate their products into the multitude of non-standard software systems in pathology. The EMPAIA Consortium (EcosysteM for Pathology Diagnostics with AI Assistance) develops an open and decentralized platform allowing AI-based apps of different vendors to be integrated with existing lab IT infrastructures. This is intended to lower the barriers to entry for AI vendors and provide pathologists with access to advanced AI tools. The EMPAIA platform is based on web technologies that can be deployed both on-premises and in the cloud. There are open-source reference implementations for core platform services that can be integrated with or replaced by proprietary alternatives as long as they conform to open API specifications. Apps can be obtained through a central marketplace so pathologists can use them in their daily workflow. In this paper, we provide an overview of the EMPAIA platform architecture. We identify critical use cases and requirements for AI-based software platforms in pathology and explain how these are fulfilled by the EMPAIA platform. Finally, we evaluate the efficiency of routing image data through the platform.

Current methods for identifying resource preferences in dung beetles are based on above‐ground trials. Although useful, these methods do not directly investigate resource provisioning of offspring below ground, missing an important part of dung beetle ecology. We tested the potential for UV‐fluorescent and non‐fluorescent coloured glitters to be used as markers for tracing the origin of dung incorporated into brood balls (dung shaped by parental beetles enclosing an egg), and so aid in a more complete understanding of resource use in dung beetles. We tested the effect of glitter addition on brood ball production in two species of tunnelling dung beetles, Onthophagus taurus and Euoniticellus fulvus . There was no effect of glitter addition on brood ball production during no‐choice tests for each species: both species made a similar number of brood balls, regardless of glitter presence or glitter colour. In a separate choice trial, O. taurus showed no preference for dung when presented with four dung pats containing four unique glitter colours. Here we show that glitter can be used as an effective marker of individual brood ball origin in tunnelling dung beetles. This method provides a useful tool for tracking below ground resource use and larval provisioning experiments in dung beetles.

In the past, when scientists encountered and studied 'new' environmental phenomena, they rarely considered the existing knowledge of First Peoples (also known as Indigenous or Aboriginal people). The scientific debate over the regularly spaced bare patches (so-called fairy circles) in arid grasslands of Australian deserts is a case in point. Previous researchers used remote sensing, numerical modelling, aerial images and field observations to propose that fairy circles arise from plant self-organization. Here we present Australian Aboriginal art and narratives, and soil excavation data, that suggest these regularly spaced, bare and hard circles in grasslands are pavement nests occupied by Drepanotermes harvester termites. These circles, called linyji (Manyjilyjarra language) or mingkirri (Warlpiri language), have been used by Aboriginal people in their food economies and for other domestic and sacred purposes across generations. Knowledge of the linyji has been encoded in demonstration and oral transmission, ritual art and ceremony and other media. While the exact origins of the bare circles are unclear, being buried in deep time and Jukurrpa, termites need to be incorporated as key players in a larger system of interactions between soil, water and grass. Ecologically transformative feedbacks across millennia of land use and manipulation by Aboriginal people must be accounted for. We argue that the co-production of knowledge can both improve the care and management of those systems and support intergenerational learning within and across diverse cultures.

Pesticide use on tropical crops has increased substantially in recent decades, posing a threat to biodiversity and ecosystem services. Amphibians and reptiles are common in tropical agricultural landscapes, but few field studies measure pesticide impacts on these taxa. Here we combine 1-year of correlative data with an experimental field approach from Indonesia. We show that while pesticide application cannot predict amphibian or reptile diversity patterns in cocoa plantations, our experimental exposure to herbicides and insecticides in vegetable gardens eliminated amphibians, whereas reptiles were less impacted by insecticide and not affected by herbicide exposure. The pesticide-driven loss of a common amphibian species known to be a pest-control agent (mainly invertebrate predation) suggests a strong indirect negative effect of pesticides on this service. We recommend landscape-based Integrated Pest Management and additional ecotoxicological studies on amphibians and reptiles to underpin a regulatory framework and to assure recognition and protection of their ecosystem services.

Insects breathe using one or a combination of three gas exchange patterns; continuous, cyclic and discontinuous, which vary in their rates of exchange of oxygen, carbon dioxide and water. In general, there is a trade-off between lowering gas exchange using discontinuous exchange that limits water loss at the cost of lower metabolic rate. These patterns and hypotheses for the evolution of discontinuous exchange have been examined for relatively large insects (>20 mg) over relatively short periods (<4 h), but smaller insects and longer time periods have yet to be examined. We measured gas exchange patterns and metabolic rates for adults of a small insect pest of grain, the red flour beetle, Tribolium castaneum (Coleoptera: Tenebrionidae), using flow-through respirometry in dry air for 48 h. All adults survived the desiccating measurement period; initially they used continuous gas exchange, then after 24 h switched to cyclic gas exchange with a 27% decrease in metabolic rate, and then after 48 h switched to discontinuous gas exchange with increased interburst duration and further decrease in metabolic rate. The successful use of the Qubit, a lower cost and so more common gas analyser, to measure respiration in the very small T. castaneum , may prompt more flow-through respirometry studies of small insects. Running such studies over long durations may help to better understand the evolution of respiration physiology and thus suggest new methods of pest management.

The increasing prevalence of digitised workflows in diagnostic pathology opens the door to life-saving applications of artificial intelligence (AI). Explainability is identified as a critical component for the safety, approval and acceptance of AI systems for clinical use. Despite the cross-disciplinary challenge of building explainable AI (xAI), very few application- and user-centric studies in this domain have been carried out. We conducted the first mixed-methods study of user interaction with samples of state-of-the-art AI explainability techniques for digital pathology. This study reveals challenging dilemmas faced by developers of xAI solutions for medicine and proposes empirically-backed principles for their safer and more effective design.