How persuasive is AI-generated propaganda?

Can large language models, a form of artificial intelligence (AI), generate persuasive propaganda? We conducted a preregistered survey experiment of US respondents to investigate the persuasiveness of news articles written by foreign propagandists compared to content generated by GPT-3 davinci (a large language model). We found that GPT-3 can create highly persuasive text as measured by participants' agreement with propaganda theses. We further investigated whether a person fluent in English could improve propaganda persuasiveness. Editing the prompt fed to GPT-3 and/or curating GPT-3's output made GPT-3 even more persuasive, and, under certain conditions, as persuasive as the original propaganda. Our findings suggest that propagandists could use AI to create convincing content with limited effort.

The Use of Bayesian Hierarchical Logistic Regression in the Development of a Modular Viral Inactivation Claim.

Low pH inactivation of enveloped viruses has historically been shown to be an effective viral inactivation step in biopharmaceutical manufacturing. To date, most statistical analyses supporting modular low pH viral inactivation claims have used descriptive statistical analyses, which in many cases do not allow for probabilistic characterization of future experimental log10 reduction values (LRVs). Using Bayesian hierarchical logistic regression modeling, probability statements regarding the likelihood of successful low pH viral inactivation based on only certain process parameter settings can be derived. This type of analysis also permits statistical modeling in the presence of historical data from different experiments and right-censored data, two issues that have not as yet been satisfactorily dealt with in the literature. The characterization of the probability of successful inactivation allows creation of a modular claim stating future LRVs will be greater than or equal to some critical value, based on only certain process parameter settings of the viral inactivation unit operation. This risk-based approach, when used in conjunction with traditional descriptive statistics, facilitates coherent and cogent decision-making about modular viral clearance LRV claims.LAY ABSTRACT: Viral contamination of biologically derived drug products is a safety concern for both regulatory agencies and drug manufacturers. Validation of the removal and inactivation of model viruses is required to ensure the safety of patients receiving these drugs, and dedicated steps, including viral filtration and chemical inactivation, are often added to manufacturing processes to provide additional clearance and inactivation capabilities. One of these steps, low pH inactivation, exposes enveloped viruses to a low pH environment to reduce the potential of the virus to infect host cells. Because the viral inactivation capability of this well-understood unit operation has been demonstrated for years across many different biological drugs, many companies have begun investigating the use of the modular viral clearance claim for the low pH inactivation step. Modular claims ensure, without experimentation, that a certain level of reduction of virus will occur if specific parameters are used in the manufacturing process, allowing manufacturers to save both time and resources in the early developmental phases of biologically derived drugs. A novel type of statistical analysis is outlined in this article that when used in addition to previously used analyses allows drug manufacturers to estimate a more valid level of virus reduction in modular viral clearance claims.

Five financial incentives to revive the Gulf of Mexico dead zone and Mississippi basin soils.

A central challenge in the Mississippi River Basin is how to continue to support profitable agricultural production, provide water supply, flood control, transportation, and other benefits, while reducing the current burden of environmental degradation. Several practices have been shown to reduce nutrient runoff and water pollution, and improve soil fertility, while often yielding profits for farmers. Yet many of these beneficial practices remain underutilized. Participants at an expert workshop identified five candidate financial mechanisms that could increase adoption of these beneficial farming practices in four focal Midwest states in the next five years: crop insurance premium subsidies, transformation of the private service provider business model, expansion and targeting of 2019 U.S. Farm Bill funding, development of new state funds, and direction of post-disaster federal funds towards habitat restoration, particularly in floodplains. This study provides rough approximations of the change in nutrient runoff and greenhouse gas (GHG) emissions, the annualized costs, and the nutrient and GHG reductions per dollar likely to result from deployment of each financial mechanism. Based upon these approximations, the adoption of these programs could reduce annual nitrate flows at the outlet of the Ohio and Upper Mississippi River Basins by 25%, surpassing the intermediate 2025 target (20% reduction) and achieving more than half of the long-term target (45% reduction) set by the Mississippi River/Gulf of Mexico Hypoxia Task Force. These approximations also illustrate that these five mechanisms could provide the same GHG reductions (∼43 Tg CO2e yr-1) as taking 12 coal-fired energy plants offline. The total cost of these five financial mechanisms is estimated at ∼$2.6 billion, or 64 g of nitrates and ∼17 kg of CO2e per dollar spent. These proposed solutions all face political, financial, cultural or institutional challenges, but with industry support, creative political action, and continued communication of both private and public benefits, they can create meaningful nutrient reductions and rebuild soils by 2022.