ETHICS & TRANSPARENCY IN AI. Nguyễn Hùng Sơn

Transcription

1 ETHICS & TRANSPARENCY IN AI Nguyễn Hùng Sơn

2 CONTENTS 1. Good and unfair examples 2. Ethical aspects in AI 3. Transparency and Interpretability 4. Conclusions

3 EXAMPLES

4 Good examples Google's DeepMind: made correct diagnoses 94.5 per cent of the time in a trial with Moorfields Eye Hospital, looking at 50 common eye problems. Another is helping us understand how the brain works Technology using artificial intelligence will have the capability to tackle issues from environmental disasters to financial crises, from crime, terrorism and war, to famine, poverty, ignorance, inequality, and appalling living standards, Climate change: AI has already been used to sift through hundreds of bird sounds to estimate when songbirds arrived at their Arctic breeding grounds. This kind of analysis will allow researchers to understand how migratory animals are responding to climate change. An AI trained by looking at hundreds of pictures of coral helped researchers to discover a new species this year, and the technique will be used to analyse coral's resistance to ocean warming.

5 Bad Examples ML-based systems for employment recommendation Loan application approving systems

6 Autonomous vehicle who is responsible if they make a mistake? how the car could be programmed? - to act in the safest way for the passenger - to protect the people in the other vehicle who is responsible for the fate of people involved in the car crash? At the beginning of 2018: ETHICAL KNOB full altruist full egoist

7 The Trolley Problem

8 More Ethical Military Machines? Isaac Asimov s Three Laws of Robotics (1) A robot may not injure a human being, or, through inaction, allow a human being to come to harm. (2) A robot must obey the orders given it by human beings except where such orders would conflict with the First Law. (3) A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

9 Bad examples The potential problems that come with artificial intelligence include a lack of transparency about what goes into the algorithms

10 THE ETHICAL ASPECTS OF ARTIFICIAL INTELLIGENCE

11 UNEMPLOYMENT: What s at risk of being automated? From legal writing to detecting fraud, composing art to conducting research, AI is learning how to automate non-routine jobs, too. A recent Mckinsey report estimates that up to 800 million jobs could be lost worldwide to automation by For the first time ever, we will start competing with machines on a cognitive level. The scariest part? They will ultimately have the capacity to be much, much smarter than us. Many economists are concerned that as a society, we won t be able to adapt and will ultimately get left behind.

12 INEQUALITY: What happens in a future without jobs? with the help of AI, a company can massively reduce its human workforce. So, its overall revenue will go to fewer people. And those in charge of AI-driven companies will proportionally earn a much higher wage. That wealth gap is already widening. In 2008, Microsoft was the only tech company that made the top ten most valuable globally; Apple came in next at 39, Google at 51. In 2018: look at top 5

13 INEQUALITY: proposition of solution Universal Basic Income (UBI) mean all citizens receive a set income, regardless of their occupation, financial history, housing and demographics. The solution has lauded by thought leaders including Richard Branson, Elon Musk, Bill Gates, and Mark Zuckerberg; and a 2018 Gallup poll found that 48 % of Americans agreed. 80% of those in favor of UBI expect that the businesses who benefit from AI should pay for it.

14 RELATIONSHIPS: How do machines trigger reward centers in the human brain? In 2015, a bot called Eugene Goostman won the Turing Test for the first time in history. Technology has already become a powerful way of streamlining human behavior and sparking action. So instead of causing dependency and addiction, could it nudge people towards better, more altruistic behaviors? Or perhaps AI will simply perform these behaviors itself? While people have a finite capacity in terms of the patience and affection they can expend on others, AI has unlimited resources that it can channel towards building and nurturing relationships.

15 SINGULARITY: How do we stay in control of an intelligent system? Why are humans at the top of the food chain? We aren t bigger, faster or stronger than many animals. No, our dominance is down to our intelligence. We can overcome pythons and lions and sharks because we re capable of creating resources to control them with: physical tools such as cages, and cognitive tools such as training. But what happens when we create AI that is smarter than us? This concept is called singularity : the point at which humans are no longer the most intelligent beings on earth. when we succeed in building Artificial General Intelligence (AGI), whereby a system would be capable of recursive self-improvement, ultimately leading to Artificial Super Intelligence (ASI) è an intelligence explosion

16 HOW DO WE DEFINE HUMANE TREATMENT OF AI? Moral status boils down to a two-way criteria: sentience and sapience. Sentience means the capacity to experience sensations like pain and suffering. Sapience refers to a set of capacities associated with higher intelligence, such as self- awareness and responding to reason.

17 RACIST ROBOTS: How do we eliminate AI bias?

18 Voluntary regulation and governance

19 Google s declaration of ethical principles, Blog post from Sundar Pichai: Google would not design AI for: Technologies that cause or are likely to cause overall harm. Weapons or other technologies whose principal purpose is to cause or directly facilitate injury to people. Technology that gathers or uses information for surveillance violating internationally accepted norms. Technologies whose purpose contravenes widely accepted principles of international law and human rights. 7 guidelines: a big win for ethical AI principles AI should be socially beneficial. It should avoid creating or reinforcing bias. Be built and tested for safety. Be accountable. Incorporate privacy design principles. Uphold high standards of scientific excellence. Be made available for use.

20 British Academy and Royal Society developed a set of principles to guide data governance. These principles state that all systems of data governance across the varied ways data is managed and used should: Protect individual and collective rights and interests. Ensure that trade-offs affected by data management and data use are made transparently, accountably and inclusively. Seek out good practices and learn from success and failure. Enhance existing democratic governance.

21 TRANSPARENCY AND EXPLAINABILITY

22 IDENTIFICATION OF THE PROBLEM

23 Transparency principles Academics, journalists, advocates, technologists, industry groups, and even the Obama Administration have raised concerns about the potential harms of unchecked, biased automation. Fairness Explanation Auditability Reliability

24 CONCLUSIONS

25 AGI IS STILL A LONG WAY AWAY At the moment, AI that is equivalent or superior to human intellect is programmed to a specific, restricted domain, such as searching the internet or playing a game. Misidentification with machine intelligence leads to false ethical evaluations of AI s potentials and threats. What could lead us to So, the AI is really good at that one thing but nothing else. over-identify with machines? Quite simply, a misconception of human life that puts the capabilities of language, mathematics and reason as its key characteristics.

26 The unknown As machines become faster, smarter and more capable, our lives become more efficient and therefore, more prosperous. AI is a boundless landscape with unimaginable power to transform the world we live in. It s exciting. It s unfamiliar. But with great change comes potential danger. History has given us many warnings of how the widespread, rapid adoption of new technologies can result in public uprisings, fear and anxiety. When will we have reached the point of no return, at which technological development becomes irreversible? it s as soon as technology has expectations of us and not the other way around. That might seem like an impossible outcome right now. But if we ve learnt anything about exponential technology, it s that it creeps up on us One moment it s a speck in the distant future, the next we can t remember life without out it. Hãy thận trọng tiến lên! And as Mary Shelley wrote: Nothing is so painful to the human mind as a great and sudden change. Our world is on the cusp of being transformed forever.