Artificial Intelligence (AI) is increasingly becoming a trusted advisor in people's lives. A new concern arises if AI persuades people to break ethical rules for profit. Employing a large-scale behavioural experiment (N = 1,572), we test whether AI-generated advice can corrupt people. We further test whether transparency about AI presence, a commonly proposed policy, mitigates potential harm of AI-generated advice. Using the Natural Language Processing algorithm, GPT-2, we generated honesty-promoting and dishonesty-promoting advice. Participants read one type of advice before engaging in a task in which they could lie for profit. Testing human behaviour in interaction with actual AI outputs, we provide first behavioural insights into the role of AI as an advisor. Results reveal that AI-generated advice corrupts people, even when they know the source of the advice. In fact, AI's corrupting force is as strong as humans'.

Prevailing methods for mapping large generative language models to supervised tasks may fail to sufficiently probe models' novel capabilities. Using GPT-3 as a case study, we show that 0-shot prompts can significantly outperform few-shot prompts. We suggest that the function of few-shot examples in these cases is better described as locating an already learned task rather than meta-learning. This analysis motivates rethinking the role of prompts in controlling and evaluating powerful language models. In this work, we discuss methods of prompt programming, emphasizing the usefulness of considering prompts through the lens of natural language. We explore techniques for exploiting the capacity of narratives and cultural anchors to encode nuanced intentions and techniques for encouraging deconstruction of a problem into components before producing a verdict. Informed by this more encompassing theory of prompt programming, we also introduce the idea of a metaprompt that seeds the model to generate its own natural language prompts for a range of tasks. Finally, we discuss how these more general methods of interacting with language models can be incorporated into existing and future benchmarks and practical applications.

Providing computer systems with the ability to understand and generate natural language has long been a challenge of engineers. Recent progress in natural language processing (NLP), like the GPT-3 language model released by OpenAI, has made both possible to an extent. In this paper, we explore the possibility of rationalising email communication using GPT-3. First, we demonstrate the technical feasibility of understanding incoming emails and generating responses, drawing on literature from the disciplines of software engineering as well as data science. Second, we apply knowledge from both business studies and, again, software engineering to identify ways to tackle challenges we encountered. Third, we argue for the economic viability of such a solution by analysing costs and market demand. We conclude that applying GPT-3 to rationalising email communication is feasible both technically and economically.

The arrival of the New Year brings us to think in many areas. What does 2021 hold in store for artificial intelligence? Here are 10 Artificial Intelligence predictions, from academic research to capital markets to regulation. We will take stock in December 2021 to assess the results. Autonomous vehicle developers like Waymo and Cruise have ongoing and massive cash flow needs.

Labeled data for imitation learning of theorem proving in large libraries of formalized mathematics is scarce as such libraries require years of concentrated effort by human specialists to be built. This is particularly challenging when applying large Transformer language models to tactic prediction, because the scaling of performance with respect to model size is quickly disrupted in the data-scarce, easily-overfitted regime. We propose PACT ({\bf P}roof {\bf A}rtifact {\bf C}o-{\bf T}raining), a general methodology for extracting abundant self-supervised data from kernel-level proof terms for co-training alongside the usual tactic prediction objective. We apply this methodology to Lean, an interactive proof assistant which hosts some of the most sophisticated formalized mathematics to date. We instrument Lean with a neural theorem prover driven by a Transformer language model and show that PACT improves theorem proving success rate on a held-out suite of test theorems from 32\% to 48\%.

Multi-turn dialogue reading comprehension aims to teach machines to read dialogue contexts and solve tasks such as response selection and answering questions. The major challenges involve noisy history contexts and especial prerequisites of commonsense knowledge that is unseen in the given material. Existing works mainly focus on context and response matching approaches. This work thus makes the first attempt to tackle the above two challenges by extracting substantially important turns as pivot utterances and utilizing external knowledge to enhance the representation of context. We propose a pivot-oriented deep selection model (PoDS) on top of the Transformer-based language models for dialogue comprehension. In detail, our model first picks out the pivot utterances from the conversation history according to the semantic matching with the candidate response or question, if any. Besides, knowledge items related to the dialogue context are extracted from a knowledge graph as external knowledge. Then, the pivot utterances and the external knowledge are combined with a well-designed mechanism for refining predictions. Experimental results on four dialogue comprehension benchmark tasks show that our proposed model achieves great improvements on baselines. A series of empirical comparisons are conducted to show how our selection strategies and the extra knowledge injection influence the results.

Cyber-defense systems are being developed to automatically ingest Cyber Threat Intelligence (CTI) that contains semi-structured data and/or text to populate knowledge graphs. A potential risk is that fake CTI can be generated and spread through Open-Source Intelligence (OSINT) communities or on the Web to effect a data poisoning attack on these systems. Adversaries can use fake CTI examples as training input to subvert cyber defense systems, forcing the model to learn incorrect inputs to serve their malicious needs. In this paper, we automatically generate fake CTI text descriptions using transformers. We show that given an initial prompt sentence, a public language model like GPT-2 with fine-tuning, can generate plausible CTI text with the ability of corrupting cyber-defense systems. We utilize the generated fake CTI text to perform a data poisoning attack on a Cybersecurity Knowledge Graph (CKG) and a cybersecurity corpus. The poisoning attack introduced adverse impacts such as returning incorrect reasoning outputs, representation poisoning, and corruption of other dependent AI-based cyber defense systems. We evaluate with traditional approaches and conduct a human evaluation study with cybersecurity professionals and threat hunters. Based on the study, professional threat hunters were equally likely to consider our fake generated CTI as true.

The capabilities of natural language models trained on large-scale data have increased immensely over the past few years. Downstream applications are at risk of inheriting biases contained in these models, with potential negative consequences especially for marginalized groups. In this paper, we analyze the occupational biases of a popular generative language model, GPT-2, intersecting gender with five protected categories: religion, sexuality, ethnicity, political affiliation, and name origin. Using a novel data collection pipeline we collect 396k sentence completions of GPT-2 and find: (i) The machine-predicted jobs are less diverse and more stereotypical for women than for men, especially for intersections; (ii) Fitting 262 logistic models shows intersectional interactions to be highly relevant for occupational associations; (iii) For a given job, GPT-2 reflects the societal skew of gender and ethnicity in the US, and in some cases, pulls the distribution towards gender parity, raising the normative question of what language models _should_ learn.

AI and machine learning may still be hot and top of mind for technology decision makers, line of business folk and investors, but that didn't prevent 2020 from broadsiding some AI initiatives. In our Vote AI ML Infrastructure 2020 survey published in July, 58% of organizations surveyed expected COVID-19 to have a negative impact on their existing AI initiatives, and 19% said the pandemic has led them to stop work on these projects. But 75% of organizations said COVID-19 led to new AI initiatives. In our just-published data from Vote AI ML Use Cases 2021 survey, the picture has changed and things look more optimistic, with 86% of respondents agreeing that the pandemic has or will cause their organization to invest in new AI initiatives. With pandemic-induced uncertainty still looming over us all, this report looks at some of the key trends in AI we expect to see in 2021.

Natural language models are the building blocks of apps including machine translators, text summarizers, chatbots, and writing assistants. But there's growing evidence showing that these models risk reinforcing undesirable stereotypes, mostly because a portion of the training data is commonly sourced from communities with gender, race, and religious prejudices. For example, OpenAI's GPT-3 places words like "naughty" or "sucked" near female pronouns and "Islam" near words like "terrorism." A new study from researchers affiliated with Amazon and the University of California, Santa Barbara aims to shed light specifically on biases in open-ended English natural language generation. The researchers created what they claim is the largest benchmark dataset of its kind containing 23,679 prompts, 5 domains, and 43 subgroups extracted from Wikipedia articles.