Editor

What is the limitation with today’s deep learning? It is good at finding patterns in data, but cannot explain how they exist. Computer Science professor Yoshua Bengio of the University of Montreal in Canada wants to change that. He was a co-recipient of the 2018 ACM Turing Award for his work on deep learning, sharing the award with professors Geoffrey Hinton and Yann LeCun. The Turing Award is the equivalence of the Nobel Prize in the field of Computing. The Wired Magazine recently wrote about his latest endeavor.  

“Current approaches to machine learning assume that the trained AI system will be applied on the same kind of data as the training data. In real life it is often not the case,” said Professor Bengio. For new situations, deep learning still requires a lot of examples to learn. He thinks that AI will not realize its full potential until deep learning can understand cause and effect, so that it can be used efficiently and effectively in critical situations. He recently co-authored a research paper outlining how this goal can be achieved.

Humans do understand cause and effect but it remains unclear why we have this capability. Bengio’s work on causality might be a small step towards answering this question. Read more about his work and related discussions in the full Wired article here.

Professor Howard is a statutory Professor of Internet Studies and Director of the Oxford Internet Institute. He investigates the impact of digital media on political life around the world, demonstrating how new information technologies are used in both civic engagement and social control in countries around the world. He delivered an invited talk for our AIWS Summit earlier this year, which was focused on the impact of misinformation on manipulation of public opinion.

This week we introduce a new report he co-authored with Samantha Bradshaw, a DPhil student the the Oxford Internet Institute, entitled “The Global Disinformation Order 2019 Global Inventory of Organised Social Media Manipulation”. The NY Times wrote about it in a recent episode.

According to the report, organizations in more than 70 countries “are spreading disinformation to discredit political opponents, bury opposing views and interfere in foreign affairs”. Facebook remains the most popular platform, which is used in 56 countries.

Professor Howard suggests that such online disinformation campaigns are not the work of “lone hackers, or individual activists, or teenagers in the basement doing things for clickbait.” Ms. Bradshaw, the report’s leading, does not see both government regulation and the steps taken by Facebook to combat disinformation go far enough. “To address that you need to look at the algorithm and the underlying business model,” she said.

The report concludes by asking “are social media platforms really creating a space for public deliberation and democracy? Or, are they amplifying content that keeps citizens addicted, disinformed, and angry?

The full report can be found here.

Nationally-known sex offender Jeffrey Epstein’s ties to MIT, Harvard, and numerous institutions and scientists are no longer a secret. But a deeper problem started to surface. As universities are increasingly turning to wealthy donors for research funding, we face several dilemmas when it comes to the ethics of funding. As discussed in a recent article of MIT Technology Review, if such a university “wanted to institute a clear ethical policy, henceforth and forever more, on what kinds of money it was and was not okay to take. What might that policy look like?”

The response is not clear. MIT refused to cut its funding ties with Saudi Arabia after the country’s leaders allegedly ordered the assassination of journalist Jamal Khashoggi. In contrast, the world-famous museums Guggenheim and the Louvre have begun to turn down money from the Sacklers, the family behind the manufacturing of the painkiller blamed for worsening the US’s opioid crisis. Or speaking of China, a country known for anti-democracy policies and human-right violations, should US universities keep taking money from its government-sponsored companies such as Huawei, as asked by Nicholas Negroponte, a co-founder of MIT Media Lab.

So we need to think about the following questions: what is the difference between bad money and money from bad people, how to measure harm, what makes the money dirty, can bad money be used for good purposes, when is anonymous donation acceptable, can dirty money be acceptable again if the donors are punished, and how far back does the reckoning need to go?

Next steps? See the full MIT Technology article for some suggestions. The AIWS’s viewpoint on this issue is clear. We need a policy framework to raise the standards for philanthropy ethics, which is not only about the missions and goals of the institution receiving the money, but also how it affects the broader society as a whole.

The AI professors, supposedly. But a recently released study by University of Rochester researchers found that an exodus of AI professors from North American universities to the private sector has hurt the post-college prospects of students. In time, this academic attrition could hinder innovation and economic growth. “The knowledge transfer is lost, and because of that, so is innovation,” said Professor Michael Gofman, one of the authors of the study.

AI faculty departures to the industry have increased exponentially since 2009. To attract them, the industry offers millions of dollars. This corporate poaching has raised public concerns. “That raises significant issues for universities and governments. They also need AI expertise, both to teach the next generation of researchers and to put these technologies into practice… But they could never match the salaries being paid in the private sector.” said Professor Yoshua Bengio of the University of Montreal, one of the 2018 Turing Award winners. Professor Ariel Procaccia of Carnergie Mellon University, added “If industry keeps hiring the cutting-edge scholars, who will train the next generation of innovators in artificial intelligence?”

Tech companies disagree with the notion that they hurt academia. “We’ve given over $250 million to academic research since 2005, and every year we host over 30 visiting faculty, dozens of Ph.D. students and thousands of interns,” said a Google spokesman, Jason Freidenfelds. He said many professors went to work at Google and returned to their university positions.

Experts are split, but many call for increased university funding to ensure that the next generation is properly educated. A recent technology article on NY Times offers more details and thoughts on this future impacting matter.

With today’s AI advances, deepfakes and fake news can easily be produced to sound and look so human-like and trustworthy that we cannot tell the difference. In an earlier episode of AIWS Newsletter, we wrote about “Grover,” a program that both creates convincing fake articles but also is able to detect them.

Speaking of “deepfakes”, it is a term originally coined in 2017, referring to the use of Generative Adversarial Networks to generate hyper-realistic videos of people doing or saying things they never did or said. How to combat deepfakes is getting a lot of traction in the AI research community. A review on OpenDataScience.com highlights three recent attempts.

One is a project at the University of California at Berkeley, by Shruti Agarwal and Hany Farid, working on an AI algorithm to detect face-swapped videos based on head and face quirks. The intuition is that people “tend to have unique head movements such as a statement of fact coupled with a nod of the head, and also face gestures such as smirking when making a point.”

Similarly, another project is by Li et al. of SUNY Albany, which is focused on detection of unnatural eye movement. Taking a different direction, Bappi et al. of the University of California at Riverside look into pixel artifacts. The idea is based on the observation that “pixels around the boundaries of objects that are artificially inserted into or removed from an image contain special characteristics, such as unnatural smoothing and feathering”.

The race between creators of deepfakes and those who fight against them will go on. To help win this race, the AIWS has also made efforts toward addressing fake content generation and its impact. In 2018, its parent organizations, the Boston Global Forum and the Michael Dukakis Institute, organized the 4th Annual Global Cybersecurity Day, with an event entitled ‘AI Solve Disinformation’ to explore the current state of cyber issues and the threat posed by disinformation and fake news, as well as effective defense mechanisms against these activities. In 2017, the BGF also wrote a policy proposal on fake news for consideration at the 2017 G-7 Summit in Taormina, Italy.

China’s approach to AI was to put its development under the control of the government. By contrast, the U.S. has allowed AI development to be dominated by private technology companies. Amid this global split, Germany, France and Japan have joined forces to fund research into “human-centered” AI that aims to respect privacy and transparency.

Warning that AI has the potential to “violate individual privacy and right to informational self-determination,” Europe is taking the lead by announcing a joint call for research proposals, backed by an initial 7.4 million euros. The focus will be on the “democratization” of AI, “integrity of data for fairness” and “AI ethics to avoid gender/age segmentation”. Results would be released on an open-access basis.

The “European way” was an attempt to find a “balance” between “government, industry and individual,” said Holger Hoos, one of the founders of the Confederation of Laboratories for Artificial Intelligence Research in Europe. This is an approach also supported by Japan, with Canada potentially joining.

The AI World Society (AIWS) welcomes this important effort by the E.U in setting standards for the rest of the world when it comes to “ethical” AI. AIWS was founded to serve a similar cause, that is, promoting ethical norms and practices in the development and use of AI. We recognized the importance of ethics guidelines at the policy level and published a comprehensive report about AI Ethics.

It is no question that AI can boost competitiveness, increase productivity, protect national security, and help solve societal challenges. As such many nations are racing to achieve a global innovation advantage in AI. But who is pulling ahead? A new study finds China lagging behind America, but surpassing the EU.

The study, conducted by the Center for Data Innovation (CDI), ranked the US first in terms of talent, research, development, and hardware. The US benefits immensely from its ability to attract, educate, and retain foreign and domestic talent. Not only that the US has the largest number of AI startups, its startup ecosystem has garnered more private equity and venture capital investment than was seen in any other country. CDI also said the U.S. “leads in the development of both traditional semiconductors and the computer chips that power AI systems; while it produces fewer AI scholarly papers than the EU or China, it produces the highest-quality papers on average.”

However, compared with China and the EU, the US does have some disadvantages. Its smaller population is a limitation in generating the data necessary to develop AI. The report suggests that the United States focus on ways to “maximize the value of the data it generates, especially by creating a regulatory environment that facilitates data sharing and reuse.”

Another disadvantage in the United States is in part due to negative public perceptions of AI. Consequently, the United States should “focus on demonstrating the value of AI to its citizens and businesses while using the federal government’s ability to fund, procure, and regulate AI to spur its adoption”. When it comes to regulations, the US should avoid “the temptation to embrace policies that would limit innovation and discourage AI adoption”.

CDI’s full report on the AI global race is here.

For almost 70 years, despite many advances in machine learning, there remain “different characteristics that are distant from current knowledge of learning in neuroscience”, even with the emergence of deep learning.

Last week, a group of scientists at Bar-Ilan University in Israel published an article on Scientific Reports, announcing a significant progress toward closing this gap. They were able to experiment with a new type of AI algorithms outperforming learning rates achieved to date by state-of-the-art learning algorithms.

The new AI is inspired by the very slow brain dynamics.

“The number of neurons in a brain is less than the number of bits in a typical disc size of modern personal computers, and the computational speed of the brain is like the second hand on a clock, even slower than the first computer invented over 70 years ago,” said the study’s lead author, Prof. Ido Kanter.

However, biological hardware is designed to deal with asynchronous inputs and refine their relative information, whereas traditional learning algorithms are based on synchronous inputs. Hence, the disadvantage of the complicated brain’s learning scheme is actually an advantage.

The idea of deep learning based on the slow brain’s dynamics illustrates that “insights of fundamental principles of our brain have to be once again at the center of future artificial intelligence.”

The full research article is here.