Liability for AI: public policy considerations

Liability for AI is the subject of a lively debate. Whether new liability rules should be introduced or not and how these rules should be designed hinges on the function of liability rules. Mainly, they create incentives for risk control, varying with their requirements – especially negligence versus strict liability. In order to do so, they have to take into account who is actually able to exercise control. In scenarios where a clear allocation of risk control is no longer possible, social insurance might step in.This article discusses public policy considerations concerning liability for artificial intelligence (AI). It first outlines the major risks associated with current developments in information technology (IT) (1.). Second, the implications for liability law are discussed. Liability rules are seen conceptualized as an instrument for risk control (2.). Negligence liability and strict liability serve different purposes making strict liability the rule of choice for novel risks (3.). The key question is, however, who should be held liable (4.). Liability should follow risk control. In future scenarios where individual risk attribution is no longer feasible social insurance might be an alternative (5). Finally, the innovation function of liability rules is stressed, affirming that appropriate liability rules serve as a stimulus for innovation, not as an impediment (6.).

Improving automatic data extraction from financial statements with clustering analysis

The financial statement analysis is a fundamental part of the credit risk attribution process, producing documents that are valuable sources of information about companies’ economic and financial wealth. Large volumes of that type of document demand automatic data extraction, and locators drive the tools for that task. However, due to the lack of regulation, there is not a standard layout for such documents, which originates a variety of document structures. Such variety burdens the feature extraction tools, reducing their performance. Clustering analysis overcomes such burden by finding the best document clusters, allowing the development of fine-tuned locators for each cluster based on their main characteristics, which is the main objective of this work. We applied state-of-the-art clustering techniques, RNG-HDBSCAN*, FOSC and MustaCHE, over financial statements documents to assess their clusters and main structures, separate outliers, and analyze their main features. The result allows the specialists to define proper locators for each cluster, increasing the performance of the data extraction tools.

A New Approach to Risk Attribution and Its Application in Credit Risk Analysis

How can risk of a company be allocated to its divisions and attributed to risk factors? The Euler principle allows for an economically justified allocation of risk to different divisions. We introduce a method that generalizes the Euler principle to attribute risk to its driving factors when these factors affect losses in a nonlinear way. The method splits loss contributions over time and is straightforward to implement. We show in an example how this risk decomposition can be applied in the context of credit risk.

P203 Oral glucocorticoid use is associated with hypertension in patients with RA

Background Oral glucocorticoids (GC) are frequently prescribed to patients with rheumatoid arthritis (RA), however GC use is associated with several potential side effects. Hypertension is cited as a possible side effect, but few studies have specifically investigated GC-associated hypertension in patients with RA with conflicting results. The aim of this study was to determine whether GCs were associated with an increased risk of incident hypertension in a cohort of patients with RA. Methods A retrospective cohort of patients with incident RA and no hypertension at RA diagnosis were identified from UK primary care electronic health records (Clinical Practice Research Datalink). GC prescriptions were used to determine time-varying GC use and dose, categorised as: no use, >0-4.9 mg/day, 5-7.4 mg/day, 7.5-14.9 mg/day, ≥15mg/day. A 3-month risk attribution model was used where patients continued to remain at risk for 3 months after the end of prescriptions. Hypertension was identified if a patient had either: 1) 2 consecutive systolic blood pressure (BP) measurements >140mmHg within a year, 2) 2 consecutive diastolic BP measurements >90mmHg within a year or 3) antihypertensive prescriptions on at least two occasions and a Read code for hypertension. Unadjusted and adjusted Cox proportional hazards (PH) regression models were fitted to determine if there was an association between GC use and hypertension. Results There were 17,760 patients with incident RA and no hypertension. The cohort had a mean age of 56.3 ± 12.7 years and were predominantly female (68%). 7,421 (42%) were prescribed GCs during follow-up. There were 6,243 cases of incident hypertension. The Cox PH model indicated that recent GC use was associated with a 17% increased hazard of hypertension (hazard ratio: 1.17 (95% CI 1.10 to 1.24)). When categorised by dose, only doses above 7.5mg were significantly associated with hypertension (Table 1). Conclusion In this large cohort of patients with RA and without hypertension, recent GC use was associated with incident hypertension. Doses ≥7.5mg were associated with hypertension while the association with lower doses was inconclusive. Clinicians need to consider cardiovascular risk when prescribing GCs and ensure BP is regularly monitored. Disclosures R.E. Costello None. M. Jani None. B.B. Yimer None. W.G. Dixon None.

Comparing Negative Attitudes Toward the Unemployed Across European Countries in 2008 and 2016: The Role of the Unemployment Rate and Job Insecurity

We explore how negative attitudes toward the unemployed are related to countries’ overall and long-term (LT) unemployment rate. Insights from the risk attribution and position theory are combined with the in-group–out-group conflict model, derived from migration literature. Multilevel analyses are performed on two waves (2008 and 2016) of the European Social Survey. Negative attitudes toward the unemployed are measured via the item “most unemployed people do not really try to find a job.” Results show that in countries with a high LT unemployment rate and/or an increase in this rate, people are more likely to hold negative attitudes toward the unemployed. Moreover, this is more pronounced among people in secure job conditions (those with a permanent contract and/or perceiving job security).