Weekly Archives - Business Intelligence Info

Tag Archives: Weekly

AI Weekly: Continual learning offers a path toward more humanlike AI

April 10, 2021 Big Data

Join GamesBeat Summit 2021 this April 28-29. Register for a free or VIP pass today.

State-of-the-art AI systems are remarkably capable, but they suffer from a key limitation: statisticity. Algorithms are trained once on a dataset and rarely again, making them incapable of learning new information without retraining. This is as opposed to the human brain, which learns constantly, using knowledge gained over time and building on it as it encounters new information. While there’s been progress toward bridging the gap, solving the problem of “continual learning” remains a grand challenge in AI.

This challenge motivated a team of AI and neuroscience researchers to found ContinualAI, a nonprofit organization and open community of continual and lifelong learning enthusiasts. ContinualAI recently announced Avalanche, a library of tools compiled over the course of a year from over 40 contributors to make continual learning research easier and more reproducible. The group also hosts conference-style presentations, sponsors workshops and AI competitions, and maintains a repository of tutorial, code, and guides.

As Vincenzo Lomonaco, cofounding president and assistant professor at the University of Pisa, explains, ContinualAI is one of the largest organizations on a topic its members consider fundamental for the future of AI. “Even before the COVID-19 pandemic began, ContinualAI was funded with the idea of pushing the boundaries of science through distributed, open collaboration,” he told VentureBeat via email. “We provide a comprehensive platform to produce, discuss and share original research in AI. And we do this completely for free, for anyone.”

Even highly sophisticated deep learning algorithms can experience catastrophic learning or catastrophic interference, a phenomenon where deep networks fail to recall what they’ve learned from a training dataset. The result is that the networks have to be constantly reminded of the knowledge they’ve gained or risk becoming “stuck” with their most recent “memories.”

OpenAI research scientist Jeff Clune, who helped to cofound Uber AI Labs in 2017, has called catastrophic forgetting the “Achilles’ heel” of machine learning and believes that solving it is the fastest path to artificial general intelligence (AGI). Last February, Clune coauthored a paper detailing ANML, an algorithm that managed to learn 600 sequential tasks with minimal catastrophic forgetting by “meta-learning” solutions to problems instead of manually engineering solutions. Separately, Alphabet’s DeepMind has published research suggesting that catastrophic forgetting isn’t an insurmountable challenge for neural networks. And Facebook is advancing a number of techniques and benchmarks for continual learning, including a model that it claims is effective in preventing the forgetting of task-specific skills.

But while the past several years have seen a resurgence of research into the issue, catastrophic forgetting largely remains unsolved, according to Keiland Cooper, a cofounding member of ContinualAI and a neuroscience research associate at the University of California, Irvine. “The potential of continual learning exceeds catastrophic forgetting and begins to touch on more interesting questions of implementing other cognitive learning properties in AI,” Cooper told VentureBeat. “Transfer learning is one example, where when humans or animals learn something previously, sometimes this learning can be applied to a new context or aid learning in other domains … Even more alluring is that continual learning is an attempt to push AI from narrow, savant-like systems to broader, more general ones.”

Even if continual learning doesn’t yield the sort of AGI depicted science fiction, Cooper notes that there are immediate advantages to it across a range of domains. Cutting-edge models are being trained on increasingly larger datasets in search of better performance, but this training comes at a cost — whether waiting weeks for training to finish or the impact of the electricity usage on the environment.

“Say you run a certain AI organization that built a natural language model that was trained over weeks on 45 terabytes of data for a few million dollars,” Cooper explained. “If you want to teach that model something new, well, you’d very likely have to start from scratch or risk overwriting what it had already learned, unless you added continual learning additions to the model. Moreover, at some point, the cost to store that data will be exceedingly high for an organization, or even impossible. Beyond this, there are many cases where you can only see the data once and so retraining isn’t even an option.”

While the blueprint for a continual learning AI system remains elusive, ContinualAI aims to connect researchers and stakeholders interested in the area and support and provide a platform for projects and research. It’s grown to over 1,000 members in the three years since its founding.

“For me personally, while there has been a renewed interest in continual learning in AI research, the neuroscience of how humans and animals can accomplish these feats is still largely unknown,” Cooper said. “I’d love to see more of an interaction with AI researchers, cognitive scientists, and neuroscientists to communicate and build upon each of their fields ides towards a common goal of understanding one of the most vital aspects of learning and intelligence. I think an organization like ContinualAI is best positioned to do just that, which allows for the sharing of ideas without the boundaries of the academic or industry walls, siloed fields, or distant geolocation.”

Beyond the mission of dissemination information about continual learning, Lomonaco believes that ContinualAI has the potential to become a reference points for a more inclusive and collaborative way of doing research in AI. “Elite university and private company labs still work mostly behind close doors, [but] we truly believe in inclusion and diversity rather than selective elitiarity. We favor transparency and open-source rather than protective IP licenses. We make sure anyone has access to the learning resources she needs to achieve her potential.”

For AI coverage, send news tips to Kyle Wiggers — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI channel, The Machine.

Thanks for reading,

Kyle Wiggers

AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: Here’s how enterprises say they’re deploying AI responsibly

April 3, 2021 Big Data

Join Transform 2021 for the most important themes in enterprise AI & Data. Learn more.

Implementing AI responsibly means different things to different companies. For some, “responsible” implies adopting AI in a manner that’s ethical, transparent, and accountable. For others, it means ensuring that their use of AI remains consistent with laws, regulations, norms, customer expectations, and organizational values. In any case, “responsible AI” promises to guard against the use of biased data or algorithms, providing an assurance that automated decisions are justified and explainable — at least in theory.

But some evidence suggests that organizations are implementing AI less responsible than they internally believe. According to a recent Boston Consulting Group survey of 1,000 enterprises, less than half that achieved AI at scale had fully mature, responsible AI implementations, according to the same report.

The lagging adoption of responsible AI belies the value that these practices can bring to bear. A study by Capgemini found that customers and employees will reward organizations that practice ethical AI with greater loyalty, more business, and even a willingness to advocate for them — and in turn, punish those that don’t. The study suggests that there’s both reputational risk and a direct impact on the bottom line for companies that don’t approach the issue thoughtfully.

To get a sense of the extent to which brands are thinking about — and practicing — the tenets of responsible AI, VentureBeat surveyed executives at companies that claim to be using AI in a tangible capacity. Their responses reveal that a single definition of “responsible AI” remains elusive. At the same time, they show an awareness of the consequences of opting not to deploy AI thoughtfully.

Companies in enterprise automation

ServiceNow was the only company VentureBeat surveyed to admit that there’s no clear definition of what constitutes responsible AI usage. “Every company really needs to be thinking about how to implement AI and machine learning responsible,” ServiceNow chief innovation officer Dave Wright told VentureBeat. “[But] every company has to define it for themselves, which unfortunately means there’s a lot of potential for harm to occur.”

According to Wright, ServiceNow’s responsible AI approach encompasses the three pillars of diversity, transparency, and privacy. When building an AI product, the company brings in a variety of perspectives and has them agree on what counts as fair, ethical, and responsible before development begins. ServiceNow also ensures that its algorithms remain explainable in the sense that it’s clear why they arrive at their predictions. Lastly, the company says it limits and obscures the amount of personally identifiable information it collects to train its algorithms. Toward this end, ServiceNow is investigating “synthetic AI” that could allow developers to train algorithms without handling real data and the sensitive information it contains.

“At the end of the day, responsible AI usage is something that only happens when we pay close attention to how AI is used at all levels of our organization. It has to be an executive-level priority,” Wright said.

Automation Anywhere says it established AI and bot ethical principles to provide guidelines to its employees, customers, and partners. They include monitoring the results of any process automated using AI or machine learning so as to prevent them from producing outputs that might reflect racial, sexist, or other biases.

“New technologies are a two-edged sword. While they can free humans to realize their potential in entirely new ways, sometimes these technologies can also, unfortunately, entrap humans in bad behavior and otherwise lead to negative outcomes,” Automation Anywhere CTO Prince Kohli told VentureBeat via email. “[W]e have made the responsible use of AI and machine learning one of our top priorities since our founding, and have implemented a variety of initiatives to achieve this.”

Beyond the principles, Automation Anywhere created an AI committee charged with challenging employees to consider ethics in their internal and external actions. For example, engineers must seek to address the threat of job loss raised by AI and machine learning technologies and the concerns of customers from an “all-inclusive” range of different minority groups. The committee also reevaluates Automation Anywhere’s principles on a regular basis so that they evolve with emerging AI technologies.

Splunk SVP and CTO Tim Tully, who anticipates the industry will see a renewed focus on transparent AI practices over the next two years, says that Splunk’s approach to putting “responsible AI” into practice is fourfold. First, the company makes sure that the algorithms it’s developing and operating are in alignment with governance policies. Then, Splunk prioritizes talent to work with its AI and machine learning algorithms to “[drive] continual improvement.” Splunk also takes steps to bake security into its R&D processes while keeping “honesty, transparency, and fairness” top of mind throughout the building lifecycle.

“In the next few years, we’ll see newfound industry focus on transparent AI practices and principles — from more standardized ethical frameworks, to additional ethics training mandates, to more proactively considering the societal implications of our algorithms — as AI and machine learning algorithms increasingly weave themselves into our daily lives,” Tully said. “AI and machine learning was a hot topic before 2020 disrupted everything, and over the course of the pandemic, adoption has only increased.”

Companies in hiring and recruitment

LinkedIn says that it doesn’t look at bias in algorithms in isolation but rather identifies what biases cause harm to users and works to eliminate this. Two years ago, the company launched an initiative called Project Every Member to take a more rigorous approach to reducing and eliminating unintended consequences in the services it builds. By using inequality A/B testing throughout the product design process, LinkedIn says it aims to build trustworthy, robust AI systems and datasets with integrity that comply with laws and “benefit society.”

For example, LinkedIn says it uses differential privacy in its LinkedIn Salary product to allow members to gain insights from others without compromising information. And the company claims its Smart Replies product, which taps machine learning to suggest responses to conversations, was built to prioritize member privacy and avoid gender-specific replies.

“Responsible AI is very hard to do without company-wide alignment. ‘Members first’ is a core company value, and it is a guiding principle in our design process,” a spokesperson told VentureBeat via email. “We can positively influence the career decisions of more than 744 million people around the world.”

Mailchimp, which uses AI to, among other things, provide personalized product recommendations for shoppers, tells VentureBeat that it trains each of its data scientists in the fields that they’re modeling. (For example, data scientists at the company working on products related to marketing receive training in marketing.) However, Mailchimp also admits that its systems are trained on data gathered by human-powered processes that can lead to a number of quality-related problems, including errors in the data, data drift, and bias.

“Using AI responsibly takes a lot of work. It takes planning and effort to gather enough data, to validate that data, and to train your data scientists,” Mailchimp chief data science officer David Dewey told VentureBeat. “And it takes diligence and foresight to understand the cost of failure and adapt accordingly.”

For its part, Zendesk says it places an emphasis on a diversity of perspectives where its AI adoption is concerned. The company claims that broadly, its data scientists examine processes to ensure that its software is beneficial, unbiased, following strong ethical principles, and securing the data that makes its AI work. “As we continue to leverage AI and machine learning for efficiency and productivity, Zendesk remains committed to continuously examining our processes to ensure transparency, accountability and ethical alignment in our use of these exciting and game-changing technologies, particularly in the world of customer experience,” Zendesk president of products Adrian McDermott told VentureBeat.

Companies in marketing and management

Adobe EVP of general counsel and corporate secretary Dana Rao points to the company’s ethics principles as an example of its commitment to responsible AI. Last year, Adobe launched an AI ethics committee and review board to help guide its product development teams and review new AI-powered features and products prior to release. At the product development stage, Adobe says its engineers use an AI impact assessment tool created by the committee to capture the potential ethical impact of any AI feature to avoid perpetuating biases.

“The continued advancement of AI puts greater accountability on us to address bias, test for potential misuse, and inform our community about how AI is used,” Rao said. “As the world evolves, it is no longer sufficient to deliver the world’s best technology for creating digital experiences; we want our technology to be used for the good of our customers and society.”

Among the first AI-powered features the committee reviewed was Neural Filters in Adobe Photoshop, which lets users add non-destructive, generative filters to create things that weren’t previously in images (e.g., facial expressions and hair styles). In accordance with its principles, Adobe added an option within Photoshop to report whether the Neural Filters output a biased result. This data is monitored to identify undesirable outcomes and allows the company’s product teams to address them by means of updating an AI model in the cloud.

Adobe says that while evaluating Neural Filters, one review board member flagged that the AI didn’t properly model the hairstyle of a particular ethnic group. Based on this feedback, the company’s engineering teams updated the AI dataset before Neural Filters was released.

“This constant feedback loop with our user community helps further mitigate bias and uphold our values as a company — something that the review board helped implement,” Rao said. “Today, we continue to scale this review process for all of the new AI-powered features being generated across our products.”

As for Hootsuite CTO Ryan Donovan, he believes that responsible AI ultimately begins and ends with transparency. Brands should demonstrate where and how they’re using AI — an ideal that Hootsuite strives to achieve, he says.

“As a consumer, for instance, I fully appreciate the implementation of bots to respond to high level customer service inquiries. However, I hate when brands or organizations masquerade those bots off as human, either through a lack of transparent labelling or assigning them human monikers,” Donovan told VentureBeat via email. “At Hootsuite, where we do use AI within our product, we have consciously endeavored to label it distinctly — suggested times to post, suggested replies, and schedule for me being the most obvious.”

SVP of product development at ADP Jack Berkowitz says that that responsible AI at ADP starts with the ethical use of data. In this context, “ethical use of data” means looking carefully at what the goal of an AI system is and the right way to achieve it.

“When AI is baked into technology, it comes with inherently heightened concerns, because it means an absence of direct human involvement in producing results,” Berkowitz said. “But a computer only considers the information you give it and only the questions you ask, and that’s why we believe human oversight is key.”

ADP retains an AI and data ethics board of experts in tech, privacy, law, and auditing that works with teams across the company to evaluate the way they use data. It also provides guidance to teams developing new uses and follows up to ensure the outputs are desirable. The board reviews ideas and evaluates potential uses to determine whether data is executed on fairly and in compliance with legal requirements and ADP’s own standards. If an idea falls short of meeting transparency, fairness, accuracy, privacy, and accountability requirements, it doesn’t move forward within the company, Berkowitz says.

Marketing platform HubSpot similarly says its AI projects undergo a peer review for ethical considerations and bias. According to senior machine learning engineer Sadhbh Stapleton Doyle, the company uses proxy data and external datasets to “stress test” its models for fairness. In addition to model cards, HubSpot also maintains a knowledge base of ways to detect and mitigate bias.

The road ahead

A number of companies declined to tell VentureBeat how they’re deploying AI responsibly in their organizations, highlighting one of the major challenges in the field: Transparency. A spokesperson for UiPath said that the robotic process automation startup “wouldn’t be able to weigh in” on responsible AI. Zoom, which recently faced allegations that its face-detection algorithm erased Black faces when applying virtual backgrounds, chose not to comment. And Intuit told VentureBeat that it had nothing to share on the topic.

Of course, transparency isn’t the end-all-be-all when it comes to responsible AI. For example, Google, which loudly trumpets its responsible AI practices, was recently the subject of a boycott by AI researchers over the company’s firing of Timnit Gebru and Margaret Mitchell, coleaders of a team working to make AI systems more ethical. Facebook also purports to be implementing AI responsibly, but to date, the company has failed to present evidence that its algorithms don’t encourage polarization on its platforms.

Returning to the Boston Consulting Group survey, Steven Mills, chief ethics officer and a coauthor, noted that the depth and breadth of most responsible AI efforts fall behind what’s needed to truly ensure responsible AI. Organizations’ responsible AI programs typically neglect the dimensions of fairness and equity, social and environmental impact, and human-AI cooperation because they’re difficult to address.

Greater oversight is a potential remedy. Companies like Google, Amazon, IBM, and Microsoft; entrepreneurs like Sam Altman; and even the Vatican recognize this — they’ve called for clarity around certain forms of AI, like facial recognition. Some governing bodies have begun to take steps in the right direction, like the EU, which earlier this year floated rules focused on transparency and oversight. But it’s clear from developments over the past months that much work remains to be done.

As Salesforce principal architect of ethical AI practice Kathy Baxter told VentureBeat in a recent interview, AI can result in harmful, unintended consequences if algorithms aren’t trained and designed inclusively. Technology alone can’t solve systemic health and social inequities, she asserts. In order to be effective, technology must be built and used responsibly — because no matter how good a tool is, people won’t use it unless they trust it.

“Ultimately, I believe the benefits of AI should be accessible to everyone, but it is not enough to deliver only the technological capabilities of AI,” Baxter said. “Responsible AI is technology developed inclusively, with a consideration towards specific design principles to mitigate, as much as possible, unforeseen consequences of deployment — and it’s our responsibility to ensure that AI is safe and inclusive. At the end of the day, technology alone cannot solve systemic health and social inequities.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI channel, The Machine.

Thanks for reading,

Kyle Wiggers

AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: With RPA on the rise, security challenges remain

March 20, 2021 Big Data

Join Transform 2021 for the most important themes in enterprise AI & Data. Learn more.

Robotic process automation (RPA) is experiencing a watershed moment. Gartner estimates that 50% of U.S. health providers will invest in RPA over the next three years. Moreover, the overall RPA market is expected to grow by more than 7% annually over the next few years to reach $ 379.87 million by 2027, up from $ 182.8 million in 2019.

Switching repetitive tasks to RPA functions not only eliminates errors, it also garners significant cost savings. That’s because RPA addresses bottlenecks with workflows, data, and documentation while providing audit trails and reducing compliance expenses and risks. RPA can also boost legacy integration and record digitization and enable data-driven decisions and “path-to-cognitive” technologies, according to Technologent CIO and VP Mike McLaughlin.

“The definition of RPA can be simplified as the category of software that automates tasks traditionally performed by a human, using software robots that follow a specific set of rules and interact with existing systems via user interfaces,” McLaughlin explained to VentureBeat via email. “These software robots can replace repetitive tasks, do system integrations, and automate transactions from task level to enterprise level via scheduled orchestration.”

For example, San Jose-based RPA firm Automation Anywhere recently worked with a pharmaceutical company in Europe to accelerate the research and approval of COVID-19 vaccines by augmenting reporting. RPA startup UiPath has also assisted with efforts around the pandemic, for instance helping the U.S. Department of Homeland Security use software bots to perform coronavirus-related data analysis.

Deloitte reports organizations that have implemented and scaled RPA see a return on investment within 12 months. And according to Everest Group, top performers earned nearly 4 times on their RPA investments, while other enterprises earned almost double.

This isn’t to suggest that RPA is without challenges. The credentials enterprises grant to RPA technology are an access point for hackers. When dealing with hundreds to thousands of RPA robots with IDs connected to a network, each could become an attack vessel if identity-centric security practices aren’t applied.

“Without tracking the exact access being granted, organizations are allowing [RPA] workers to leave their most valuable asset out in the open — privileged credentials,” One Identity president and GM Bhagwat Swaroop told VentureBeat. “Bad actors are targeting privileged credentials to gain access and move laterally within the network. With 53% of breaches being linked to misused or overused privileged credentials, the unmonitored and unrestricted access of RPA makes it even more susceptible to a breach than its human counterparts. If the privileged access vulnerabilities of RPA aren’t addressed quickly, I predict there will be a significant amount of RPA breaches that occur over the next year.”

Part of the problem is that many RPA platforms don’t focus on solving security flaws. That’s because they’re optimized to increase productivity and because some security solutions are too costly to deploy and integrate with RPA, Swaroop says.

Of course, the first step to solving the RPA security dilemma is recognizing that there is one. Realizing that RPA workers have identities gives IT and security teams a head start to securing RPA technology prior to its implementation. Swaroop recommends that organizations extend their identity and governance administration (IGA) to focus on the “why” behind a task, rather than the “how.” Through a strong IGA process, companies adopting RPA can implement a zero trust model to manage all identities — from human to machine and application.

“Through IGA processes, enterprises have the correct access management controls to mitigate risks, such as privilege creep, orphaned accounts, and the exposure of passwords and secrets. By eliminating the gaping holes in companies’ current RPA strategies, IT and security teams can ensure cybercriminals can’t compromise the robot or, worse, infiltrate their network,” Swaroop said.

He also suggests putting in place a privileged access management (PAM) setup that can secure and govern RPA systems. PAM systems allow enterprises to secure, control, and audit credentials and privileges RPA technology uses without compromising the return on investment.

“When a digital worker needs privileged access, the robot can retrieve credentials automatically from a PAM system, without any exposure to the bot owners or developers,” Swaroop said. “This not only provides a full audit trail of which digital workers accessed what applications, but also provides individual accountability and proof that no one can obtain the password, in a noncompliant manner, without slowing down robotic operations.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI channel, The Machine.

Thanks for reading,

Kyle Wiggers

AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: Facebook, Google, and the tension between profits and fairness

March 14, 2021 Big Data

Join Transform 2021 for the most important themes in enterprise AI & Data. Learn more.

This week, we learned a lot more about the inner workings of AI fairness and ethics operations at Facebook and Google and how things have gone wrong. On Monday, a Google employee group wrote a letter asking Congress and state lawmakers to pass legislation to protect AI ethics whistleblowers. That letter cites VentureBeat reporting about the potential policy outcomes of Google firing former Ethical AI team co-lead Timnit Gebru. It also cites research by UC Berkeley law professor Sonia Katyal, who told VentureBeat, “What we should be concerned about is a world where all of the most talented researchers like [Gebru] get hired at these places and then effectively muzzled from speaking. And when that happens, whistleblower protections become essential.” The 2021 AI Index report found that AI ethics stories — including Google firing Gebru — were among the most popular AI-related news articles of 2020, an indication of rising public interest. In the letter published Monday, Google employees spoke of harassment and intimidation, and a person with policy and ethics matters at Google described a “deep sense of fear” since the firing of ethics leaders Gebru and former co-lead Margaret Mitchell.

On Thursday, MIT Tech Review’s Karen Hao published a story that unpacked a lot of previously unknown information about ties between AI ethics operations at Facebook and the company’s failure to address misinformation peddled through its social media platforms and tied directly to a number of real-world atrocities. A major takeaway from this lengthy piece is that Facebook’s responsible AI team focused on addressing algorithmic bias in place of issues like disinformation and political polarization, following 2018 complaints by conservative politicians, although a recent study refutes their claims. The events described in Hao’s report appear to document political winds shifting the definition of fairness at Facebook, and the extremes to which a company will go in order to escape regulation.

Facebook CEO Mark Zuckerberg’s public defense of President Trump last summer and years of extensive reporting by journalists have already highlighted the company’s willingness to profit from hate and misinformation. A Wall Street Journal article last year, for example, found that the majority of people in Facebook groups labeled as extremist joined as a result of a recommendation made by a Facebook algorithm.

What this week’s MIT Tech Review story details is a tech giant deciding how to define fairness to advance its underlying business goals. Just as with Google’s Ethical AI team meltdown, Hao’s story describes forces within Facebook that sought to co-opt or suppress ethics operations after just a year or two of operation. One former Facebook researcher, who Hao quoted on background, described their work as helping the company maintain the status quo in a way that often contradicted Zuckerberg’s public position on what’s fair and equitable. Another researcher speaking on background described being told to block a medical-misinformation detection algorithm that had noticeably reduced the reach of anti-vaccine campaigns.

In what a Facebook spokesperson pointed to as the company’s official response, Facebook CTO Mike Schroepfer called the core narrative of Hao’s article incorrect but made no effort to dispute facts reported in the story.

Facebook chief AI scientist Yann LeCun, who got into a public spat with Gebru over the summer about AI bias that led to accusations of gaslighting and racism, claimed the story had factual errors. Hao and her editor reviewed the claims of inaccuracy and found no factual error.

Facebook’s business practices have played a role in digital redlining, genocide in Myanmar, and the insurrection at the U.S. Capitol. At an internal meeting Thursday, according to BuzzFeed reporter Ryan Mac, an employee asked how Facebook funding AI research differs from Big Tobacco’s history of funding health studies. Mac said the response was that Facebook was not funding its own research in this specific instance, but AI researchers spoke extensively about that concern last year.

Last summer, VentureBeat covered stories involving Schroepfer and LeCun after events drew questions about diversity, hiring, and AI bias at the company. As that reporting and Hao’s nine-month investigation highlight: Facebook has no system in place to audit and test algorithms for bias. A civil rights audit commissioned by Facebook and released last summer calls for the regular and mandatory testing of algorithms for bias and discrimination.

Following allegations of toxic, anti-Black work environments, both Facebook and Google have been accused in the past week of treating Black job candidates in a separate and unequal fashion. Reuters reported last week that the Equal Employment Opportunity Commission (EEOC) is investigating “systemic” racial bias at Facebook in hiring and promotions. And additional details about an EEOC complaint filed by a Black woman emerged Thursday. At Google, multiple sources told NBC News last year that diversity investments in 2018 were cut back in order to avoid criticism from conservative politicians.

On Wednesday, Facebook also made its first attempt to dismiss an antitrust suit brought against the company by the Federal Trade Commission (FTC) and attorneys general from 46 U.S. states.

All of this happened in the same week that U.S. President Joe Biden nominated Lina Khan to the FTC, leading to the claim that the new administration is building a “Big Tech antitrust all-star team.” Last week, Biden appointed Tim Wu to the White House National Economic Council. A supporter of breaking up Big Tech companies, Wu wrote an op-ed last fall in which he called one of the multiple antitrust cases against Google bigger than any single company. He later referred to it as the end of a decades-long antitrust winter. VentureBeat featured Wu’s book The Curse of Bigness about the history of antitrust reform in a list of essential books to read. Other signals that more regulation could be on the way include the appointments of FTC chair Rebecca Slaughter and OSTP deputy director Alondra Nelson, who have both expressed a need to address algorithmic bias.

The Google story calling for whistleblower protections for people researching the ethical deployment of AI marks the second time in as many weeks that Congress has received a recommendation to act to protect people from AI.

The National Security Commission on Artificial Intelligence (NSCAI) was formed in 2018 to advise Congress and the federal government. The group is chaired by former Google CEO Eric Schmidt, and Google Cloud AI chief Andrew Moore is among the group’s 15 commissioners. Last week, the body published a report that recommends the government spend $ 40 billion in the coming years on research and development and the democratization of AI. The report also says individuals within government agencies essential to national security should be given a way to report concerns about “irresponsible AI development.” The report states that “Congress and the public need to see that the government is equipped to catch and fix critical flaws in systems in time to prevent inadvertent disasters and hold humans accountable, including for misuse.” It also encourages ongoing implementation of audits and reporting requirements. However, as audits at businesses like HireVue have shown, there are a lot of different ways to audit an algorithm.

This week’s consensus between organized Google employees and NSCAI commissioners who represent business executives from companies like Google Cloud, Microsoft, and Oracle suggests some agreement between broad swaths of people intimately familiar with the deployment of AI at scale.

In casting the final vote to approve the NSCAI report, Moore said, “We are the human race. We are tool users. It’s kind of what we’re known for. And we’ve now hit the point where our tools are, in some limited sense, more intelligent than ourselves. And it’s a very exciting future, which we have to take seriously for the benefit of the United States and the world.”

While deep learning and forms of AI may be capable of doing things that people describe as superhuman, this week we got a reminder of how untrustworthy AI systems can be when OpenAI demonstrated that its state-of-the-art model can be fooled to think an apple with “iPod” written on it is in fact an iPod, something any person with a pulse could discern.

Hao described the subjects of her Facebook story as well-intentioned people trying to make changes in a rotten system that acts to protect itself. Ethics researchers in a corporation of that size are effectively charged with considering society as a shareholder, but everyone else they work with is expected to think first and foremost about the bottom line, or personal bonuses. Hao said that reporting on the story has convinced her that self-regulation cannot work.

“Facebook has only ever moved on issues because of or in anticipation of external regulation,” she said in a tweet.

After Google fired Gebru, VentureBeat spoke with ethics, legal, and policy experts who have also reached the conclusion that “self-regulation can’t be trusted.”

Whether at Facebook or Google, each of these incidents — often told with the help of sources speaking on condition of anonymity — shine light on the need for guardrails and regulation and, as a recent Google research paper found, journalists who ask tough questions. In that paper, titled “Re-imagining Algorithmic Fairness in India and Beyond,” researchers state that “Technology journalism is a keystone of equitable automation and needs to be fostered for AI.”

Companies like Facebook and Google sit at the center of AI industry consolidation, and the ramifications of their actions extend beyond even their great reach, touching virtually every aspect of the tech ecosystem. A source familiar with ethics and policy matters at Google who supports whistleblower protection laws told VentureBeat the equation is pretty simple: “[If] you want to be a company that touches billions of people, then you should be responsible and held accountable for how you touch those billions of people.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers — and be sure to subscribe to the AI Weekly newsletter and bookmark The Machine.

Thanks for reading,

Khari Johnson

Senior AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: These researchers are improving AI’s ability to understand different accents

March 6, 2021 Big Data

Join Transform 2021 for the most important themes in enterprise AI & Data. Learn more.

The pandemic appears to have supercharged voice app usage, which was already on an upswing. According to a study by NPR and Edison Research, the percentage of voice-enabled device owners who use commands at least once a day rose between the beginning of 2020 and the start of April. Just over a third of smart speaker owners say they listen to more music, entertainment, and news from their devices than they did before, and owners report requesting an average of 10.8 tasks per week from their assistant this year compared with 9.4 different tasks in 2019. According to a new report from Juniper Research, consumers will interact with voice assistants on 8.4 billion devices by 2024.

But despite their growing popularity, assistants like Alexa, Google Assistant, and Siri still struggle to understand diverse regional accents. According to a study by the Life Science Centre, 79% of people with accents alter their voice to make sure that they’re understood by their digital assistants. And in a recent survey commissioned by the Washington Post, popular smart speakers made by Google and Amazon were 30% less likely to understand non-American accents than those of native-born users.

Traditional approaches to narrowing the accent gap would require collecting and labeling large datasets of different languages, a time- and resource-intensive process. That’s why researchers at MLCommons, a nonprofit related to MLPerf, an industry-standard set of benchmarks for machine learning performance, are embarking on a project called 1000 Words in 1000 Languages. It’ll involve creating a freely available pipeline that can take any recorded speech and automatically generate clips to train compact speech recognition models.

“In the context of consumer electronic devices, for instance, you don’t want to have to go out and build new language datasets because that’s costly, tedious, and error-prone,” Vijay Janapa Reddi, an associate professor at Harvard and a contributor on the project, told VentureBeat in a phone interview. “What we’re developing is a modular pipeline where you’ll be able to plug in different sources speech and then specify the [words] for training that you want.”

While the pipeline will be limited in scope in that it’ll only create training datasets for small, low-power models that continually listen for specific keywords (e.g. “OK Google” or “Alexa”), it could represent a significant step toward truly accent-agnostic speech recognition systems. By convention, training a new keyword-spotting model would require manually collecting thousands of examples of labeled audio clips for each keyword. When the pipeline is released, developers will be able to simply provide a list of keywords they wish to detect along with a speech recording and the pipeline will automate the extraction, training, and validation of models without requiring any labeling.

“It’s not even really creating a dataset, it’s just training a dataset that comes about as a result of searching the larger corpus,” Reddi explained. “It’s like doing a Google search. What you’re trying to do is find a needle in a haystack — you end up with a subset of results with different accents and whatever else you have in there.”

The 1000 Words in 1000 Languages project builds on existing efforts to make speech recognition models more accessible — and equitable. Mozilla’s Common Voice, an open source and annotated speech dataset, consists of voice snippets and voluntarily contributed metadata useful for training speech engines like speakers’ ages, sex, and accents. As a part of Common Voice, Mozilla maintains a dataset target segment that aims to collect voice data for specific purposes and use cases, including the digits “zero” through “nine” as well as the words “yes,” “no,” “hey,” and “Firefox.” For its part, in December, MLCommons released the first iteration of a public 86,000-hour dataset for AI researchers, with later versions due to branch into more languages and accents.

“The organizations that have a huge amount of speech are often large organizations, but speech is something that has many applications,” Reddi said. “The question is, how do you get this into the hands of small organizations that don’t have the same scale as big entities like Google and Microsoft? If they have a pipeline, they can just focus on what they’re building.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI channel, The Machine.

Thanks for reading,

Kyle Wiggers

AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: Biden calls for $37 billion to address chip shortage

February 28, 2021 Big Data

Join Transform 2021 for the most important themes in enterprise AI & Data. Learn more.

Shortly after a meeting with members of Congress on Wednesday, President Joe Biden signed an executive order that launches a review of supply chain vulnerabilities in the United States. COVID-19 made evident gaps in the U.S. supply chain in medical equipment like face masks and ventilators, but in a ceremony carried live by TV news networks, Biden held up a chip, calling it the “21st century horseshoe nail.”

AI research has received military funding from the outset, and government organizations like DARPA continue to fund AI startups, but a global chip supply shortage caused by COVID-19 has hindered the progress of numerous industries. During his remarks, Biden acknowledged that semiconductor chip shortages impacts products like cars, smartphones, and medical diagnostic equipment. Earlier this month, Ford said the shortage would reduce production by up to 20% in Q1 2021.

Smartphone production is also expected to decline as a result of the chip shortage, and earlier this month, business executives from AMD, Intel, Nvidia, and Qualcomm sent a letter to Biden urging support for the CHIPS for America Act and stating that a chip shortage could interrupt progress for emerging technology areas like AI, 5G, and quantum computing. CHIPS stands for Creating Helpful Incentives to Produce Semiconductors. That bill was introduced in Congress in summer 2020 and called for $ 22 billion in tax credits and research and development funding. The American Foundries Act, also introduced in Congress last summer, called for $ 25 billion. As part of the executive order signing ceremony Wednesday, Biden pledged support for $ 37 billion over an unspecified period described as “short term” and pledged to work with ally nations to address the chip bottleneck. The executive order will also review key minerals and materials, pharmaceuticals, and the kinds of batteries used in electric vehicles.

“We need to prevent the supply chain crisis from hitting in the first place. And in some cases, building resilience will mean increasing our production of certain types of elements here at home. In others, it’ll mean working more closely with our trusted friends and partners, nations that share our values, so that our supply chains can’t be used against us as leverage,” Biden said.

A 2019 U.S. Air Force report put the urgency of the matter in context. That report finds that “90% of all high-volume, leading-edge [semiconductor] production will soon be based in Taiwan, China, and South Korea.” The Semiconductor Industry Association (SIA) finds that 12% of global semiconductor production takes place in the U.S. today.

Analysts who spoke to VentureBeat found a number of factors contributing to the current chip shortage.

Kevin Krewell is a principal analyst at Tirias Research. He attributes the chip shortage to an initial slump followed by unexpected demand increase, not enough advanced semiconductor manufacturers, the fact that more complex semiconductor processes are hard to scale, and that there’s a long lead time on building new semiconductor manufacturing facilities, or “fabs.” Intel and Samsung being slow to get advanced process nodes out in a timely fashion has put more pressure on TSMC to make more chips, but he expects shortages will get addressed as more capacity comes on line and demand returns to more predictable levels.

“The $ 37 billion figure is a small start, but it is a start,” he said. Building a single semiconductor manufacturing facility can cost tens of billions of dollars.

Linley Group senior analyst Mike Demler said a fourth quarter growth in car sales caught auto manufacturers off guard, that high demand for consumer electronics during the pandemic rippled through other industries. He also said that the U.S. semiconductor industry wants to use the shortage to increase domestic semiconductor-manufacturing capacity.

“The semiconductor industry has thrived because of the global supply chain. Greater investment in R&D could help restore US technological leadership in manufacturing technology, but it would take many years to shift the ecosystem,” Demler said.

IDC analyst Mario Morales said the chip shortage is a real thing but that some businesses may be blaming that shortage to distract from deeper underlying business problems or poor planning. For example, Ford may be reducing inventory due to a lack of chips, but Toyota has a stockpile.

“I think some of this is just not very good business continuity planning, and that some of this is a reaction to that. And others I think they’re using this as an excuse, because there is some underperformance from some of these vendors,” he said.

When discussing what caused the chip shortage, analysts VentureBeat interviewed talked primarily about COVID-19 and made virtually no mention of China, but you could potentially say the opposite about national security interests in the U.S., the other driver of interest in domestic chip production. The final report from the National Security Commission on AI is due out next week. That group was formed by Congress a few years ago and is made up of some of the most influential AI and business leaders in the United States today, like soon-to-be Amazon CEO Andy Jassy, Google Cloud AI chief Andrew Moore, and former Google CEO Eric Schmidt.

The report calls for the United States to remain “two generations ahead of China,” with $ 12 billion over the next five years for research, development, and infrastructure. It also supports creation of a national microelectronics research strategy like the kind espoused in the American Foundries Act. The 2021 National Defense Authorization Act created a committee to develop a national microelectronic research strategy.

The report calls for 40% refundable tax credit as well. The CHIPS for America Act also calls for hefty tax credits for semiconductor manufacturers through 2027.

“The dependency of the United States on semiconductor imports, particularly from Taiwan, creates a strategic vulnerability for both its economy and military to adverse foreign government action, natural disaster, and other events that can disrupt the supply chains for electronics,” the draft final report reads. “If a potential adversary bests the United States in semiconductors, it could gain the upper hand in every domain of warfare.”

The draft final report echoes calls from the National Security Commission on Artificial Intelligence (NSCAI) for more public-private partnerships around semiconductors. In testimony before the House Budget committee about how AI will change the economy, NSCAI commissioner and Intelligence Advanced Research Projects Activity (IARPA) director Dr. Jason Matheny said, “It will be very difficult for China to match us if we play our cards right.”

“We shouldn’t rest on our laurels, but if we pursue policies that strengthen our semiconductor industry while also placing the appropriate controls on the manufacturing equipment that China doesn’t have and that China currently doesn’t have the ability to produce itself and is probably a decade away from being able to produce itself, we’ll be in a very strong position,” he said.

A Bloomberg analysis found that Chinese spending on computer chip production equipment jumped 20% in 2020 compared to 2019. Reuters has recorded Chinese chip imports above $ 300 billion for the past three years.

Advanced semiconductor manufacturing facilities can be more expensive than modern day aircraft carriers, and fabs are only part of the equation. IDC’s Morales agreed with Krewell that $ 37 billion is a start, but that becoming a leader in manufacturing could take a decade of investment not just in semiconductor manufacturing plants, but also design, IP, and infrastructure.

“The goal should be to collaborate a lot more with other regions that I would say are more neutral,” Morales said. He added that, based on conversations with manufacturers, he expects an end to chip supply chain shortage issues by Q2 or Q3 2021.

We’ll have to wait a few months to see what the review ordered by the Biden administration prescribes to improve resilience when it comes to chip production, but it seems clear that $ 37 billion may only be the start.

For AI coverage, send news tips to Khari Johnson, Kyle Wiggers, and AI editor Seth Colaner — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI Channel.

Thanks for reading,

Khari Johnson

Senior AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: The challenges of creating open source AI training datasets

February 20, 2021 Big Data

In January, AI research lab OpenAI released Dall-E, a machine learning system capable of creating images to fit any text caption. Given a prompt, Dall-E generates photos for a range of concepts, including cats, logos, and glasses.

The results are impressive, but training Dall-E required building a large-scale dataset that OpenAI has so far opted not to make public. Work is ongoing on an open source implementation, but according to Connor Leahy, one of the data scientists behind the effort, development has stalled because of the challenges in compiling a corpus that respects both moral and legal norms.

“There’s plenty of not-legal-to-scrape data floating around that isn’t [fair use] on platforms like social media, Instagram first and foremost,” Leahy, who’s a member of the volunteer AI research effort EleutherAI, told VentureBeat. “You could scrape that easily at large scale, but that would be against the terms of service, violate people’s consent, and probably scoop up illegal data both due to copyright and other reasons.”

Indeed, creating AI training datasets in a privacy-preserving, ethical way remains a major blocker for researchers in the AI community, particularly those who specialize in computer vision. In January 2019, IBM released a corpus designed to mitigate bias in facial recognition algorithms that contained nearly a million photos of people from Flickr. But neither the photographers nor the subjects of the photos were notified by IBM that their work would be included. Separately, an earlier version of ImageNet, a dataset used to train AI systems around the world, was found to contain photos of naked children, porn actresses, college parties, and more — all scraped from the web without those individuals’ consent.

“There are real harms that have emerged from casual repurposing, open-sourcing, collecting, and scraping of biometric data,” said Liz O’Sullivan, cofounder and technology director at the Surveillance Technology Oversight Project, a nonprofit organization litigating and advocating for privacy. “[They] put people of color and those with disabilities at risk of mistaken identity and police violence.”

Techniques that rely on synthetic data to train models might lessen the need to create potentially problematic datasets in the first place. According to Leahy, while there’s usually a minimum dataset size needed to achieve good performance on a task, it’s possible to a degree to “trade compute for data” in machine learning. In other words, simulation and synthetic data, like AI-generated photos of people, could take the place of real-world photos from the web.

“You can’t trade infinite compute for infinite data, but compute is more fungible than data,” Leahy said. “I do expect for niche tasks where data collection is really hard, or where compute is super plentiful, simulation to play an important role.”

O’Sullivan is more skeptical that synthetic data will generalize well from lab conditions to the real world, pointing to existing research on the topic. In a study last January, researchers at Arizona State University showed that when an AI system trained on a dataset of images of engineering professors was tasked with creating faces, 93% were male and 99% white. The system appeared to have amplified the dataset’s existing biases — 80% of the professors were male and 76% were white.

On the other hand, startups like Hazy and Mostly AI say that they’ve developed methods for controlling the biases of data in ways that actually reduce harm. A recent study published by a group of Ph.D. candidates at Stanford claims the same — the coauthors say their technique allows them to weight certain features as more important in order to generate a diverse set of images for computer vision training.

Ultimately, even where synthetic data might come into play, O’Sullivan cautions that any open source dataset could put people in that set at greater risk. Piecing together and publishing a training dataset is a process that must be undertaken thoughtfully, she says — or not at all, where doing so might result in harm.

“There are significant worries about how this technology impacts democracy and our society at large,” O’Sullivan said.

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers and AI editor Seth Colaner — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI channel, The Machine.

Thanks for reading,

Kyle Wiggers

AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: Techno-utopianism in the workplace and the threat of excessive automation

February 12, 2021 Big Data

Every so often, VentureBeat writes a story about something that needs to go away. A few years back, my colleague Blair Hanley Frank argued that AI systems like Einstein, Sensei, and Watson must go because corporations tend to overpromise results for their products and services. I’ve taken runs at charlatan AI and white supremacy.

This week, a series of events at the intersection of the workplace and AI lent support to the argument that techno-utopianism has no place in the modern world. Among the warning signs in headlines was a widely circulated piece by a Financial Times journalist who said she was wrong to be optimistic about robots.

The reporter describes how she used to be a techno-optimist but in the course of her reporting found that robots can crunch people into their system and force them to work at a robot’s pace. In the article, she cites the Center for Investigative Reporting’s analysis of internal Amazon records that found instances of human injury were higher in Amazon facilities with robots than in facilities without robots.

“Dehumanization and intensification of work is not inevitable,” wrote the journalist, who’s quite literally named Sarah O’Connor. Fill in your choice of Terminator joke here.

Also this week: The BBC quoted HireVue CEO Kevin Parker as saying AI is more impartial than a human interviewer. After facing opposition on multiple fronts, HireVue announced last month it would no longer use facial analysis in its AI-powered video interview analysis of job candidates. Microsoft Teams got similar tech this week to recognize and highlight who’s enjoying a video call.

External auditors have examined the Al used by HireVue and hiring software company Pymetrics, which refers to its AI as “entirely bias free,” but the processes seem to have raised more questions than they’ve answered.

And VentureBeat published an article about a research paper with a warning: Companies like Google and OpenAI have a matter of months to confront negative societal consequences of the large language models they release before they perpetuate stereotypes, replace jobs, or are used to spread disinformation.

What’s important to understand about that paper, written by researchers at OpenAI and Stanford, is that before criticism of large language models became widespread, research and dataset audits found major flaws in large computer vision datasets that were over a decade old, like ImageNet and 80 Million Tiny Images. An analysis of face datasets dating back four decades also found ethically questionable practices.

A day after that article was published, OpenAI cofounder Greg Brockman tweeted what looked like an endorsement of a 90-hour work week. Run the math on that. If you slept seven hours a night, you would have about four hours a day to do anything that is not work — like exercise, eating, resting, or spending time with your family.

Agreed: https://t.co/1rZc1Dhgma pic.twitter.com/vESf1DrBn4

— Greg Brockman (@gdb) February 10, 2021

An end to techno-utopianism doesn’t have to mean the death of optimistic views about ways technology can improve human lives. There are still plenty of people who believe that indoor farming can change lives for the better or that machine learning can accelerate efforts to address climate change.

Google AI ethics co-lead Margaret Mitchell recently made a case for AI design that keeps the bigger picture in mind. In an email sent to company leaders before she was placed under investigation, she said consideration of ethics and inclusion is part of long-term thinking for long-term beneficial outcomes.

“The idea is that, to define AI research now, we must look to where we want to be in the future, working backwards from ideal futures to this moment, right now, in order to figure out what to work on today,” Mitchell said. “When you can ground your research thinking in both foresight and an understanding of society, then the research questions to currently focus on fall out from there.”

With that kind of long-term thinking in mind, Google’s Ethical AI team and Google DeepMind researchers have produced a framework for carrying out internal algorithm audits, questioned the wisdom of scale when addressing societal issues, and called for a culture change in the machine learning community. Google researchers have also advocated rebuilding the AI industry according to principles of anticolonialism and queer AI and evaluating fairness using sociology and critical race theory. And ethical AI researchers recently asserted that algorithmic fairness cannot simply be transferred from Western nations to non-Western nations or those in the Global South, like India.

The death of techno-utopia could entail creators of AI systems recognizing that they may need to work with the communities their technology impacts and do more than simply abide by the scant regulations currently in place. This could benefit tech companies as well as the general public. As Parity CEO Rumman Chowdhury told VentureBeat in a recent story about what algorithmic auditing startups need to succeed, unethical behavior can have reputation and financial costs that stretch beyond any legal ramifications.

Computer scientists trying to find a path to ‘Ethical AI’ but refusing to learn anything about white supremacy, heteropatriarchy, capitalism, ablism, or settler colonialism pic.twitter.com/vPf0aa4gLa

— Sasha Costanza-Chock (@schock) February 8, 2021

The lack of comprehensive regulation may be why some national governments and groups like Data & Society and the OECD are building algorithmic assessment tools to diagnose risk levels for AI systems.

Numerous reports and surveys have found automation on the rise during the pandemic, and events of the past week remind me of the work of MIT professor and economist Daron Acemoglu, whose research has found one robot can replace 3.3 human jobs.

In testimony before Congress last fall about the role AI will play in the economic recovery in the United States, Acemoglu warned the committee about the dangers of excessive automation. A 2018 National Bureau of Economic Research (NBER) paper coauthored by Acemoglu says automation can create new jobs and tasks, as it has done in the past, but says excessive automation is capable of constraining labor market growth and has potentially acted as a drag on productivity growth for decades.

“AI is a broad technological platform with great promise. It can be used for helping human productivity and creating new human tasks, but it could exacerbate the same trends if we use it just for automation,” he told the House Budget committee. “Excessive automation is not an inexorable development. It is a result of choices, and we can make different choices.”

To avoid excessive automation, in that 2018 NBER paper Acemoglu and his coauthor, Boston University research fellow Pascual Restrepo, call for reforms of the U.S. tax code, because it currently favors capital over human labor. They also call for new or strengthened institutions or policy to ensure shared prosperity, writing, “If we do not find a way of creating shared prosperity from the productivity gains generated by AI, there is a danger that the political reaction to these new technologies may slow down or even completely stop their adoption and development.”

This week’s events involve complexities like robots and humans working together and language models with billions of parameters, but they all seem to raise a simple question: “What is intelligence?” To me, working 90 hours a week is not intelligent. Neither is perpetuating bias or stereotypes with language models or failing to consider the impact of excessive automation. True intelligence takes into account long-term costs and consequences, historical and social context, and, as Sarah O’Connor put it, makes sure “the robots work for us, and not the other way around.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers and AI editor Seth Colaner — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI channel, The Machine.

Thanks for reading,

Khari Johnson

Senior AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: What Andy Jassy’s ascension to CEO means for Amazon’s AI initiatives

February 6, 2021 Big Data

This week, Jeff Bezos announced that he’ll step down as CEO at Amazon and transition to an executive chair role during the third quarter of this year. Amazon Web Services (AWS) CEO Andy Jassy will take his place, heading up a company currently valued at around $ 1.6 trillion.

Jassy, who joined Amazon in 1997 and has led AWS since its inception in 2003, believes Amazon’s decision to double down on AI early differentiated it from the competition. In May 2020, Gartner ranked AWS the industry leader in terms of vision and ability to execute on AI developer services. Beyond AWS, product recommendations on Amazon are powered by AI, as well as Alexa, Prime Air, Amazon Go, and the pick paths used in distribution centers to find products and fulfill orders.

So how might Jassy’s elevation to CEO impact Amazon’s AI initiatives? Interviews in recent years suggest Jassy is enthusiastic about cloud services tailored to the needs of machine learning practitioners, particularly for large enterprise applications. Controversially, Jassy has also said customers, not Amazon itself, are responsible for curbing their usage of potentially problematic AI technologies like facial recognition.

In a conversation with Silicon Angle in December, Jassy said he expects the majority of applications to be infused with AI in the next five to 10 years. While he endorses the idea of catering to expert machine learning practitioners who know how to train, tune, and deploy AI models, he asserts that AWS, more than rivals like Google Cloud Platform and Microsoft Azure, has aimed to “democratize” data science by lowering the barriers to entry.

“There just aren’t that many expert machine learning practitioners. And so it never gets extensive in most enterprises if you don’t make it easier for everyday developers and data scientists to use machine learning,” Jassy told Silicon Angle. He stressed the importance of “top-layer” AI services that transcribe audio, translate text, and more via APIs, without requiring customers to develop custom models. But he said the most important thing Amazon has done to make AI more accessible is building fully managed services.

“Enterprises have so much data that they want to use predictive algorithms to get value added,” Jassy said during a keynote at the Goldman Sachs Technology and Internet Conference in San Francisco last February.

SageMaker is one example of these fully managed services. Launched in 2016, it’s designed to let developers build, test, and maintain AI models from a single dashboard. Amazon says SageMaker, which gained nine new capabilities in December — following the launch of SageMaker Studio, an integrated development environment for machine learning — now has tens of thousands of customers.

It’s a safe bet that investments in services akin and complementary to SageMaker will accelerate with Jassy at the helm. So too, most likely, will the buildout of backend tools Amazon uses to solve challenges like call analytics.

“As Clay Christensen, author of The Innovator’s Dilemma, said … people hire products and services to do a job. They don’t really care what you do under the covers, but they’re hiring a product to do a job,” Jassy told Silicon Angle. “[Some people] don’t actually want to hire machine learning [experts]. They want to have an easier way to get automatic call analytics on all of their calls … And what we’re finding is that increasingly we’re using machine learning as the source to get those jobs done, but without people necessarily knowing that that’s what we’re doing behind the scenes.”

This work might be of a controversial nature. In an interview at Recode’s 2019 Code Conference, Jassy defended the company’s facial recognition service, Rekognition, while calling for the federal government to introduce national guidelines. (In September 2019, Recode reported that Amazon was writing its own facial recognition laws to pitch to lawmakers.) “Just because tech could be misused doesn’t mean we should ban it and condemn it,” he said, adding that Amazon would provide its facial recognition tech to governments, excepting those that violate the law or infringe on civil liberties.

Last year, Amazon declared a halt on the sale of facial recognition to police departments for 12 months but did not necessarily extend that restriction to federal law enforcement agencies. Prior to the moratorium, the company reportedly attempted to sell its facial recognition tech to U.S. Immigration and Customs Enforcement (ICE), and police in Orlando, Florida and other cities have trialed it.

A number of academics have called Jassy’s stance on facial recognition technology, which runs counter to that of many Amazon shareholders, problematic at best. Anima Anandkumar, the principal scientist for artificial intelligence at Amazon, told PBS Frontline that facial recognition isn’t “battle-tested” to work in the types of challenging conditions where law enforcement might use it (e.g., with low-light, grainy, or low-quality images). And dating back to 2018, AI researchers Joy Buolamwini, Timnit Gebru, and Deborah Raji have found that facial recognition software from companies like Amazon work best for white men and worst for women with dark skin. Amazon has publicly dismissed their coauthored work, the Gender Shades project.

Given this history, it seems unlikely that Jassy will extend the moratorium on facial recognition sales when it expires in July. He’s also unlikely to curtail the law enforcement relationships that Ring, Amazon’s smart home division, has fostered since its acquisition by Amazon in 2018. Ring has reportedly partnered with over 2,000 police and fire departments across the U.S. dating back to 2015, when Ring let the Los Angeles Police Department test how front-door footage might reduce property crimes.

Advocacy groups like Fight for the Future and the Electronic Frontier Foundation have accused Ring of using its cameras and Neighbors app (which delivers safety alerts) to build a private surveillance network via these partnerships. The Electronic Frontier Foundation in particular has singled Ring out for marketing strategies that foster fear and promote a sale-spurring “vicious cycle,” and for “[facilitating] reporting of so-called ‘suspicious’ behavior that really amounts to racial profiling.”

“We don’t have a large number of police departments that are using our facial recognition technology, and as I said, we’ve never received any complaints of misuse. Let’s see if somehow they abuse the technology — they haven’t done that,” Jassy told PBS Frontline in a 2020 interview. “And to assume they’re going to do that and therefore you shouldn’t allow them to have access to the most sophisticated technology out there doesn’t feel like the right balance to me.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers — and be sure to subscribe to the AI Weekly newsletter and bookmark The Machine.

Thanks for reading,

Kyle Wiggers

AI Staff Writer

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: Announcing our ‘AI and the future of health care’ special issue

January 30, 2021 Big Data

Artificial intelligence and health care both deal heavily with issues of complexity, efficacy, and societal impact. All of that is multiplied when the two intersect. As health care providers and vendors work to use AI and data to improve patient care, health outcomes, medical research, and more, they face what are now standard AI challenges. Data is difficult and messy. Machine learning models struggle with bias and accuracy. And ethical challenges abound. But there’s a heightened need to solve these problems when they’re couched within the daily life-and-death context of health care.

Then, in the midst of the AI’s growth in health care, the pandemic hit, challenging old ways of doing things and pushing systems to their breaking points. In our upcoming special issue, “AI and the future of health care,” we examine how providers and vendors are tackling the challenges of this extraordinary time.

The biggest hurdle has to do with data. Health care produces massive amounts of data, from electronic health records (EHR) to imaging to information on hospital bed capacity. There’s enormous promise in using that data to create AI models that can improve care and even help cure diseases, but there are barriers to that progress. Privacy concerns top the list, but worldwide health care data also needs standardization. There are still too many errors in this data, and the medical community must address persisting biases before they become even more entrenched.

When humans rely on AI to help them make clinical decisions like injury or disease diagnoses, they also have to be aware of their own biases. Because bias exists in the data AI models are built upon, practitioners have to be careful not to fall into the trap of automation bias, relying too much on model output to make decisions. It’s a delicate balance with profound impacts on human health and life.

The pandemic has also challenged the practical day-to-day functions of health care systems. As COVID-19 cases threaten to overwhelm hospitals and patients and doctors risk infection during in-person visits, providers are figuring out how to deliver patient care remotely. With more doctors shifting to telemedicine, chatbots and other tools are helping relieve some of the burden and allowing patients to access care from the safety of their own homes.

For particularly vulnerable populations, like senior citizens, remote care may be necessary, especially if they’re in locked-down residential facilities or can’t easily get to their doctor. The technologies involved in monitoring such patients include wearables that track vitals and even special wireless tech that offers no-touch, personalized biometric tracking.

These are sea changes in health care, and because of the pandemic, they’re coming faster than anyone expected. But a certain optimism persists — a sense that despite unprecedented challenges to the medical field, careful and responsible use of AI can enable permanent, positive changes in the health care system. The astonishing speed with which researchers developed a working COVID-19 vaccine offers ample evidence of the way necessity spurs medical innovation. The best of the technologies, tools, and techniques that health care providers are employing now could soon become standard and lead to more democratized, less expensive, and overall better health care.

You can get this special issue delivered straight to your inbox next week by signing up here.

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

up-to-date information on the subjects of interest to you
our newsletters
gated thought-leader content and discounted access to our prized events, such as Transform
networking features, and more

Become a member

Let’s block ads! (Why?)

Big Data – VentureBeat

Tag Archives: Weekly

AI Weekly: Continual learning offers a path toward more humanlike AI

AI Weekly: Here’s how enterprises say they’re deploying AI responsibly

AI Weekly: With RPA on the rise, security challenges remain

AI Weekly: Facebook, Google, and the tension between profits and fairness

AI Weekly: These researchers are improving AI’s ability to understand different accents

VentureBeat

AI Weekly: Biden calls for $37 billion to address chip shortage

VentureBeat

AI Weekly: The challenges of creating open source AI training datasets

AI Weekly: Techno-utopianism in the workplace and the threat of excessive automation

VentureBeat

AI Weekly: What Andy Jassy’s ascension to CEO means for Amazon’s AI initiatives

VentureBeat

AI Weekly: Announcing our ‘AI and the future of health care’ special issue

VentureBeat

Recent Posts

Categories

Archives