Artificial intelligence guru Jack Clark has written the longest, most interesting Twitter thread on AI policy that I've ever read. After a brief initial introductory tweet on August 6, Clark went on to post an additional 79 tweets in this thread. It was a real tour de force. Because I'm currently finishing up a new book on AI governance, I decided to respond to some of his thoughts on the future of governance for artificial intelligence (AI) and machine learning (ML). Clark is a leading figure in the field of AI science and AI policy today. He is the co-founder of Anthropic, an AI safety and research company, and he previously served as the Policy Director of OpenAI. So, I take seriously what he has to say on AI governance matters and really learned a lot from his tweetstorm. But I also want to push back on a few things. Specifically, several of the issues that Clark raises about AI governance are not unique to AI per se; they are broadly applicable to many other emerging technology sectors, and even some traditional ones. Below, I will refer to this as my "general critique" of Clark's tweetstorm. On the other hand, Clark correctly points to some issues that are unique to AI/ML and which really do complicate the governance of computational systems.

Like the European Union (EU)'s General Data Protection Regulation (GDPR) that entered into force in 2016, the upcoming Artificial Intelligence (AI) Act will have extraterritorial scope and global impact. Considering the AI Act's broad scope and the financial risks relating to non-compliance, businesses must prepare for these future regulatory changes now and proactively take the initiatives to comply with best practices early on, according to a new whitepaper by Swiss data services company Unit8. The paper, titled Upcoming AI Regulation: What to expect and how to prepare, delves into the EU's forthcoming AI Act, providing insights into the future development of AI regulation in Europe and the potential implications for organizations worldwide. The European Commission (EC) unveiled a proposal for a legal framework on AI in April 2021, seeking to address risks of specifically created by AI applications, proposing a list of high risk applications, setting clear requirements for AI systems for high risk applications and defining specific obligations for AI users and providers of high risk applications. The proposed rules also propose a conformity assessment method for AI systems, propose enforcement after an AI system is placed in the market, and propose a governance structure at European and national level.

A few short years ago, personal digital assistants like Amazon's Alexa, Apple's Siri, and Google Assistant sounded futuristic. Now, the future is here and this future is embedded, augmented, and ubiquitous. Digital assistants can be found in your office, home, car, hotel, phone, and many other places. They have recently undergone a massive transformation and run on operating systems that are fueled by artificial intelligence (A.I.). They observe and collect data in real-time and have the capability to pull information from different sources such as smart devices and cloud services and put the information into context using A.I. to make sense of the situation.

It started with a single tweet in November 2019. David Heinemeier Hansson, a high-profile tech entrepreneur, lashed out at Apple's newly launched credit card, calling it "sexist" for offering his wife a credit limit 20 times lower than his own. The allegations spread like wildfire, with Hansson stressing that artificial intelligence – now widely used to make lending decisions – was to blame. "It does not matter what the intent of individual Apple reps are, it matters what THE ALGORITHM they've placed their complete faith in does. And what it does is discriminate. While Apple and its underwriters Goldman Sachs were ultimately cleared by US regulators of violating fair lending rules last year, it rekindled a wider debate around AI use across public and private industries. Politicians in the European Union are now planning to introduce the first comprehensive global template for regulating AI, as institutions increasingly automate routine tasks in an attempt to boost efficiency and ...

A governance paradigm called "responsible AI" describes how a particular organization handles the ethical and legal issues around artificial intelligence (AI). Liable AI projects are primarily motivated by the need to clarify who is responsible if something goes wrong. The data scientists and software engineers who create and implement an organization's AI algorithmic models are responsible for developing appropriate, reliable AI standards. This indicates that each organization has different requirements for the procedures needed to stop prejudice and ensure transparency. Supporters of responsible AI believe that a widely accepted governance framework of AI best practices will make it simpler for organizations worldwide to ensure that their AI programming is human-centered, interpretable, and explainable, much like ITIL provided a common framework for delivering IT services.

Non-autoregressive text to speech (TTS) models such as FastSpeech can synthesize speech significantly faster than previous autoregressive models with comparable quality. The training of FastSpeech model relies on an autoregressive teacher model for duration prediction (to provide more information as input) and knowledge distillation (to simplify the data distribution in output), which can ease the one-to-many mapping problem (i.e., multiple speech variations correspond to the same text) in TTS. However, FastSpeech has several disadvantages: 1) the teacher-student distillation pipeline is complicated and time-consuming, 2) the duration extracted from the teacher model is not accurate enough, and the target mel-spectrograms distilled from teacher model suffer from information loss due to data simplification, both of which limit the voice quality. In this paper, we propose FastSpeech 2, which addresses the issues in FastSpeech and better solves the one-to-many mapping problem in TTS by 1) directly training the model with ground-truth target instead of the simplified output from teacher, and 2) introducing more variation information of speech (e.g., pitch, energy and more accurate duration) as conditional inputs. Specifically, we extract duration, pitch and energy from speech waveform and directly take them as conditional inputs in training and use predicted values in inference. We further design FastSpeech 2s, which is the first attempt to directly generate speech waveform from text in parallel, enjoying the benefit of fully end-to-end inference. Experimental results show that 1) FastSpeech 2 achieves a 3x training speed-up over FastSpeech, and FastSpeech 2s enjoys even faster inference speed; 2) FastSpeech 2 and 2s outperform FastSpeech in voice quality, and FastSpeech 2 can even surpass autoregressive models. Audio samples are available at https://speechresearch.github.io/fastspeech2/.

Prior work on language models (LMs) shows that training on a large number of diverse tasks improves few-shot learning (FSL) performance on new tasks. We take this to the extreme, automatically extracting 413,299 tasks from internet tables - orders of magnitude more than the next-largest public datasets. Finetuning on the resulting dataset leads to improved FSL performance on Natural Language Processing (NLP) tasks, but not proportionally to dataset scale. In fact, we find that narrow subsets of our dataset sometimes outperform more diverse datasets. For example, finetuning on software documentation from support.google.com raises FSL performance by a mean of +7.5% on 52 downstream tasks, which beats training on 40 human-curated NLP datasets (+6.7%). Finetuning on various narrow datasets leads to similar broad improvements across test tasks, suggesting that the gains are not from domain adaptation but adapting to FSL in general. We do not observe clear patterns between the datasets that lead to FSL gains, leaving open questions about why certain data helps with FSL.

The Patent Act requires an "inventor" to be a natural person, the US Court of Appeals for the Federal Circuit said, rejecting computer scientist Stephen Thaler's bid for patents on two inventions he said his DABUS system created. Thaler said in an email Friday that DABUS, which stands for "Device for the Autonomous Bootstrapping of Unified Sentience," is "natural and sentient." His attorney Ryan Abbott of Brown Neri Smith & Khan said the decision "ignores the purpose of the Patent Act" and has "real negative social consequences." He said they plan to appeal. The US Patent and Trademark Office declined to comment on the decision.

Thaler had asked for patents on behalf of his AI system Court affirms ruling that patent'inventor' must be human being Court affirms ruling that patent'inventor' must be human being The Patent Act requires an "inventor" to be a natural person, the U.S. Court of Appeals for the Federal Circuit said, rejecting computer scientist Stephen Thaler's bid for patents on two inventions he said his DABUS system created. Thaler said in an email Friday that DABUS, which stands for "Device for the Autonomous Bootstrapping of Unified Sentience," is "natural and sentient." His attorney Ryan Abbott of Brown Neri Smith & Khan said the decision "ignores the purpose of the Patent Act" and has "real negative social consequences." He said they plan to appeal. The U.S. Patent and Trademark Office declined to comment on the decision.

The prominent computer scientist discusses the quest for ethical artificial intelligence.