Making AI Safe May be an Impossible Task

When it comes to creating safe AI and regulating this technology, these great minds have little clue what they’re doing. They don’t even know where to begin.

I met with Michael Page, the Policy and Ethics Advisor at OpenAI.

Beneath the glittering skyscrapers of the self-proclaimed “city of the future,” he told me of the uncertainty that he faces. He spoke of the questions that don’t have answers, and the fantastically high price we’ll pay if we don’t find them.

The conversation began when I asked Page about his role at OpenAI. He responded that his job is to “look at the long-term policy implications of advanced AI.” If you think that this seems a little intangible and poorly defined, you aren’t the only one. I asked Page what that means, practically speaking. He was frank in his answer: “I’m still trying to figure that out.” 

Page attempted to paint a better picture of the current state of affairs by noting that, since true artificial intelligence doesn’t actually exist yet, his job is a little more difficult than ordinary.

He noted that, when policy experts consider how to protect the world from AI, they are really trying to predict the future.

They are trying to, as he put it, “find the failure modes … find if there are courses that we could take today that might put us in a position that we can’t get out of.” In short, these policy experts are trying to safeguard the world of tomorrow by anticipating issues and acting today. 

The problem is that they may be faced with an impossible task.

Page is fully aware of this uncomfortable possibility, and readily admits it. “I want to figure out what can we do today, if anything. It could be that the future is so uncertain there’s nothing we can do,” he said.

asked for a concrete prediction of where humanity and AI will together be in a year, or in five years, Page didn’t offer false hope: “I have no idea,”

However, Page and OpenAI aren’t alone in working on finding the solutions. He therefore hopes such solutions may be forthcoming: “Hopefully, in a year, I’ll have an answer. Hopefully, in five years, there will be thousands of people thinking about this,” Page said.

Source: Futurism

Facebooktwittergoogle_plusredditpinterestlinkedinmailFacebooktwittergoogle_plusredditpinterestlinkedinmail

The tech bias: why Silicon Valley needs social theory

Photo by Ramin Talaie/Corbis/Getty Image

In the summer of 2017, a now infamous memo came to light. Written by James Damore, then an engineer at Google, it claimed that the under-representation of women in tech was partly caused by inherent biological differences between men and women.

That Google memo is an extreme example of an imbalance in how different ways of knowing are valued.

Silicon Valley tech companies draw on innovative technical theory but have yet to really incorporate advances in social theory.

Social theorists in fields such as sociology, geography, and science and technology studies have shown how race, gender and class biases inform technical design.

So there’s irony in the fact that employees hold sexist and racist attitudes, yet ‘we are supposed to believe that these same employees are developing “neutral” or “objective” decision-making tools’, as the communications scholar Safiya Umoja Noble at the University of Southern California argues in her book Algorithms of Oppression (2018).

If tech companies are serious about building a better society, and aren’t just paying lip service to justice for their own gain, they must attend more closely to social theory.

If social insights were easy, and if practice followed readily from understanding, then racism, poverty and other debilitating systems of power and inequality would be a thing of the past.

New insights about society are as challenging to produce as the most rarified scientific theorems – and addressing pressing contemporary problems requires as many kinds of knowers and ways of knowing as possible.

Source: aeon



Facebooktwittergoogle_plusredditpinterestlinkedinmailFacebooktwittergoogle_plusredditpinterestlinkedinmail

DeepMind Ethics and Society hallmark of a change in attitude

The unit, called DeepMind Ethics and Society, is not the AI Ethics Board that DeepMind was promised when it agreed to be acquired by Google in 2014. That board, which was convened by January 2016, was supposed to oversee all of the company’s AI research, but nothing has been heard of it in the three-and-a-half years since the acquisition. It remains a mystery who is on it, what they discuss, or even whether it has officially met.

DeepMind Ethics and Society is also not the same as DeepMind Health’s Independent Review Panel, a third body set up by the company to provide ethical oversight – in this case, of its specific operations in healthcare.

Nor is the new research unit the Partnership on Artificial Intelligence to Benefit People and Society, an external group founded in part by DeepMind and chaired by the company’s co-founder Mustafa Suleyman. That partnership, which was also co-founded by Facebook, Amazon, IBM and Microsoft, exists to “conduct research, recommend best practices, and publish research under an open licence in areas such as ethics, fairness and inclusivity”.

Nonetheless, its creation is the hallmark of a change in attitude from DeepMind over the past year, which has seen the company reassess its previously closed and secretive outlook. It is still battling a wave of bad publicity started when it partnered with the Royal Free in secret, bringing the app Streams to active use in the London hospital without being open to the public about what data was being shared and how.

The research unit also reflects an urgency on the part of many AI practitioners to get ahead of growing concerns on the part of the public about how the new technology will shape the world around us.

Source: The Guardian



Facebooktwittergoogle_plusredditpinterestlinkedinmailFacebooktwittergoogle_plusredditpinterestlinkedinmail

Why we launched DeepMind Ethics & Society

We believe AI can be of extraordinary benefit to the world, but only if held to the highest ethical standards.

Technology is not value neutral, and technologists must take responsibility for the ethical and social impact of their work.

As history attests, technological innovation in itself is no guarantee of broader social progress. The development of AI creates important and complex questions. Its impact on society—and on all our lives—is not something that should be left to chance. Beneficial outcomes and protections against harms must be actively fought for and built-in from the beginning. But in a field as complex as AI, this is easier said than done.

As scientists developing AI technologies, we have a responsibility to conduct and support open research and investigation into the wider implications of our work. At DeepMind, we start from the premise that all AI applications should remain under meaningful human control, and be used for socially beneficial purposes. 

So today we’re launching a new research unit, DeepMind Ethics & Society, to complement our work in AI science and application. This new unit will help us explore and understand the real-world impacts of AI. It has a dual aim: to help technologists put ethics into practice, and to help society anticipate and direct the impact of AI so that it works for the benefit of all. 

If AI technologies are to serve society, they must be shaped by society’s priorities and concerns.

Source: DeepMind


Facebooktwittergoogle_plusredditpinterestlinkedinmailFacebooktwittergoogle_plusredditpinterestlinkedinmail

Can machines learn to be moral?  #AI

AI works, in part, because complex algorithms adeptly identify, remember, and relate data … Moreover, some machines can do what had been the exclusive domain of humans and other intelligent life: Learn on their own.

As a researcher schooled in scientific method and an ethicist immersed in moral decision-making, I know it’s challenging for humans to navigate concurrently the two disparate arenas. 

It’s even harder to envision how computer algorithms can enable machines to act morally.

Moral choice, however, doesn’t ask whether an action will produce an effective outcome; it asks if it is good decision. In other words, regardless of efficacy, is it the right thing to do? 

Such analysis does not reflect an objective, data-driven decision but a subjective, judgment-based one.

Individuals often make moral decisions on the basis of principles like decency, fairness, honesty, and respect. To some extent, people learn those principles through formal study and reflection; however, the primary teacher is life experience, which includes personal practice and observation of others.

Placing manipulative ads before a marginally-qualified and emotionally vulnerable target market may be very effective for the mortgage company, but many people would challenge the promotion’s ethicality.

Humans can make that moral judgment, but how does a data-driven computer draw the same conclusion? Therein lies what should be a chief concern about AI.

Can computers be manufactured with a sense of decency?

Can coding incorporate fairness? Can algorithms learn respect? 

It seems incredible for machines to emulate subjective, moral judgment, but if that potential exists, at least four critical issues must be resolved:

  1. Whose moral standards should be used?
  2. Can machines converse about moral issues?
  3. Can algorithms take context into account?
  4. Who should be accountable?

Source: Business Insider David Hagenbuch



Facebooktwittergoogle_plusredditpinterestlinkedinmailFacebooktwittergoogle_plusredditpinterestlinkedinmail