Forbes Insight Team reports:
“The business world has been much more focused on the upside, not the downside, of these technologies.” Codifying ethical standards or enforceable regulations is incredibly complex. And the challenges in using AI ethically are not equal across industries (such as) the programmed intelligence they will use to occasionally make life-and-death decisions on behalf of humans; the risk of bias if an applicant is denied a loan due to a low credit score or slapped with exorbitant insurance premiums; directing people to content that will keep them clicking on ads; “who is deciding who constitutes a target? Who gets to decide what an enemy looks like?”
An algorithm can’t choose where, when, or how it’s used, including whether it’s used for good or bad intentions. This puts the burden for the ethical use of artificial intelligence (AI) squarely on human shoulders.
But are companies taking up the mantle of responsibility? Keith Strier, Global Artificial Intelligence Leader at Ernst & Young, isn’t convinced. “The business world has been much more focused on the upside, not the downside, of these technologies,” Strier says. “They’re focused on, ‘Let’s get this built. Show me the money.’”
Meanwhile, universities and think tanks, such as the Partnership for Artificial Intelligence, the Future of Life Institute, OpenAI, and the AI Now Institute, are actively trying to establish ethical guardrails for AI and urging both governments and business leaders to adopt them.
But codifying ethical standards or enforceable regulations is incredibly complex. And the challenges in using AI ethically are not equal across industries. Notably, healthcare AI has emerged as a minefield of ethics quandaries, from potential misdiagnoses by flawed algorithms to gene editing—that’s why we’ve covered that issue earlier in Forbes AI.
Beyond healthcare, however, are many industries where AI technology is raising equally pressing ethical questions. Here is a look at four of the biggest sectors where ethics and AI are colliding quickly.
Autonomous Transportation
One of the most complex aspects of developing autonomous vehicles is the programmed intelligence they will use to occasionally make life-and-death decisions on behalf of human passengers.
The research behind this concept isn’t new: the “Trolley Problem,” for instance, is a well-known philosophical experiment in which the conductor of an out-of-control streetcar must choose between staying on the track, which will kill five bystanders; or switching tracks and killing just one person.
Making a moral calculation between potential victims is hard enough for humans. For autonomous vehicles, it will eventually be a matter of coding. An algorithm must be designed to automatically make that choice. Self-driving cars will need to wrestle with many variations of the “trolley problem.”
According to Meredith Whittaker, co-founder and co-director of the AI Now Institute at NYU, this is only the tip of the ethical iceberg. Accountability and liability are open and pressing issues that society must address as autonomous vehicles take over roadways.
“Who ultimately bears responsibility if you’re looking at a black box that only the company, who has a vested interest in perpetuating the use of these technologies, is able to audit?” she asks. “These are urban planning questions. They’re regulatory questions. And they’re questions around the role of the public in decision making that impacts their lives and environments.”
Two 2018 fatalities during autonomous vehicle road tests, in Arizona and California, only elevate the urgency of these questions. In the case of a California pedestrian killed by an autonomous taxi, investigators found that the company’s technology had failed catastrophically in an easily preventable way.
Months after launching probes into the accidents, the National Highway Traffic Safety Administration signaled that existing regulations for autonomous vehicles may face changes—to weaken them.
The rationale behind that decision was to remove roadblocks to technology innovation, so that U.S.-based automakers don’t get left behind as competitors from other countries beat them to market with self-driving cars.
“Germany, Japan, and China are very much out front on this,” Strier explains, citing Germany’s recently unveiled national AI policy, which eased regulations on autonomous vehicle development. “From a
global competitive perspective, U.S. regulators are keen to enable companies to have freedom in the sandbox to develop these technologies.”
Looser regulations sidestep the ethical challenges that carmakers will face inevitably, but for now it appears that both enterprise and government have set those questions aside. With more than $80 billion invested in developing self-driving vehicles in recent years, it appears that too much money is at stake to tap the brakes for ethical debate.
“The train has left the station,” Strier says. “Federal governments around the world are trying not to stand in the way.”
Financial Services & Insurance
Small business loans. Home mortgage rates. Insurance premiums. For financial institutions and insurers, AI software is increasingly automating these decisions, saving them money by speeding
application and claims processing and detecting fraud. Automation in the financial services sector could save companies $512 billion by 2020, according to a 2018 Capgemini study.
But the risk of bias is rampant. If an applicant is denied a loan due to a low credit score or deemed a high risk and slapped with exorbitant insurance premiums, the algorithm making those assessments is, for a majority of the time, opaque.
“There are a lot of dangers here,” says Whittaker. “You’re looking at a scenario in which these companies, whose ultimate duty is to their shareholders, are going to be increasingly making assumptions about people’s private lives, about their habits, and what that may mean about their risk profile.”
For racial, gender, and ethnic minorities, biased AI can have potentially life-changing impact. A 2018 study conducted at UC Berkeley found that consumer-lending technology discriminates against minority applicants.
“There are all sorts of ways in which we’re seeing these automated systems effectively become the gatekeepers for determining who gets resources and who doesn’t,” Whittaker warns.
Ironically, tech companies are trying to fight AI bias by using other AI as watchdogs. For example, one tech company created an algorithm to “correct” biased datasets and produce unbiased results. And in 2016, a consortium of researchers designed AI to detect gender and racial discrimination in algorithms, a tool that the Consumer Finance Protection Bureau (CFPB) sought to adopt to test loan decisions for bias.
But major U.S. financial services and insurance companies have pushed back on the CFPB’s anti-bias efforts, saying they put them at a competitive disadvantage to fintech startups. In 2018, lawmakers successfully squashed a CFPB policy aimed at ending racial discrimination in auto lending.
Still, Strier argues, the issue of AI fairness in financial services is on governments’ radar globally, and regulations will emerge to fight bias over time. What role technology will play in enforcing those regulations remains unclear, however.
“Everyone is worried about bias,” says Strier. “It’s a pervasive problem—and a science problem. There are emerging technological methods coming into view, but there’s no clear-cut answer on how to avoid bias. You’ve got a disease for which there’s no obvious cure yet.”
Journalism And “Fake News”
The concept of fake news—deliberate misinformation disguised as journalism that is spread largely through social media—has become an international burden. In March 2017, news broke that a U.K. firm had gamed the data-sharing protocols of a major social media platform in order to influence voters during the U.S. presidential election. The uproar from the ensuing scandal still echoes, and the phrase “fake news” has spread, undercutting public trust in the media.
“This is really an ad-tech story,” Whittaker says. “Massive platforms like Facebook and Twitter work by directing people to content they might like that will keep them clicking on ads. Journalism has become just one input that is buffeted by whatever these algorithmic whims are that are ultimately calibrated to earn these companies a profit, not to where it is most salutary for an informed public.”
Can regulation or legislation stem the tide of this corrosive misinformation?
“We’re going to have years of emerging regulations in different parts of the world trying out different ways to deal with this,” he predicts. “It’s a tradeoff: We want the free flow of information, but we’ve basically created a super highway for bad stuff.”
And with “deepfake” video now emerging, more bad stuff is on the way. Thanks to tools like FakeApp, which is based on open source code from Google, anyone can digitally manipulate video to create a seemingly realistic-looking record of an event that never occurred.
“It’s like Photoshop on steroids,” says Whittaker. “I think about these technologies in the context of old-fashioned dirty political tricks we’ve seen. Not everyone has to believe a fake video for it to have a profound impact on our democratic processes.”
In the U.S., legislators have expressed concerns about deepfake video, but no legislation has emerged and, for now, Congress has let Facebook skate on promises to better police itself.
Elsewhere around the globe, countries from Germany and France to Russia and Indonesia have introduced new laws to crack down on the spread of misinformation on social media, but these laws have raised ethical concerns themselves—namely, that they may be misused to muzzle free speech. In Malaysia, for example, a journalist found guilty of spreading fake news faces up to six years in prison.
“This is a pervasive challenge of our time,” Strier says. “There’s a lot of discussion and methods being talked about, but no one has solved it. Right now, the computational propagandists have the upper hand.”
Military
In 2018, Google employees took a stand. Learning that their company was supplying AI tech to the U.S. Air Force for “Project Maven” which could be used for deadly drone strikes, more than 3,000 workers signed a letter of protest to CEO Sundar Pichai. Bowing to the pressure, the company pulled out of the military contract.
The ethics of tech companies partnering with the U.S. military are fraught. Do the personal moral codes of employees trump the security interests of a country engaged in an “AI arms race” with rogue nations, geopolitical foes, and terrorists?
For academics and think tanks like the Future of Life Institute, it’s a no-brainer. They have launched a worldwide campaign calling on countries to ban the development of autonomous weapons.
But that hasn’t deterred the U.S. Defense Department from boosting its AI spending, and a good chunk of that $2 billion investment is going toward undisclosed partnerships with tech companies.
For Open AI’s Whittaker, one of the biggest ethical dilemmas is transparency. The public has given tech companies mountains of personal data with the tacit belief that they will guard it, not weaponize it.
“These companies are protected by corporate secrecy,” she notes. “It is completely probable that there are similar projects at other companies, potentially Google, that no one knows anything about because they’re protected on one side by corporate secrecy and on the other side by military secrecy protocols.”
And that opacity should be ethically troubling to society at large, she argues.
“Who is deciding who constitutes a target? Who gets to decide what an enemy looks like?” Whittaker asks. “Should we have a say when our data that we entrusted to tech companies is used to train AI systems that are used in weapons?”
Complicating the ethical equation is the question of whether the U.S. should have the means to defend itself against possible AI-based attacks from adversaries or cyber terrorists. Americans are evenly divided on this issue. In a 2018 survey by the Brookings Institute, 30% of respondents said the U.S. should develop AI technologies for warfare, 39% didn’t, and 31% were undecided.
Tech companies are increasingly finding themselves forced to take a stand by their own employees. Like Google’s disaffected workers, a global coalition of Microsoft employees have publicly rebuked CEO Satya Nadella for signing a $479 million contract with the U.S. Army to sell it Hololens technology to train U.S. soldiers for battle—and even to use in combat.
Nadella has publicly defended the contract as a “principled decision” to protect frontline troops. But he also assured his rattled staff that Microsoft would “continue to have a dialogue” about these issues.
As far as ethical issues surrounding the use of AI are concerned, that conversation, in boardrooms and offices around the world, is far from over.
0 comments:
Post a Comment