Categories
AI

AI Weekly: Is AI alien invasion imminent?

We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 – 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!


Want AI Weekly for free each Thursday in your inbox? Sign up here.

Is an AI alien invasion headed for earth? The VentureBeat editorial staff marveled at the possibility this week, thanks to the massive online traffic earned by one Data Decision Makers community article, with its impossible-to-ignore title, Prepare for arrival: Tech pioneer warns of alien invasion. 

The column, written by Louis Rosenberg, founder of Unanimous AI, was certainly buoyed not only by its SEO-friendly title, but its breathless opener: “An alien species is headed for planet Earth and we have no reason to believe it will be friendly. Some experts predict it will get here within 30 years, while others insist it will arrive far sooner. Nobody knows what it will look like, but it will share two key traits with us humans – it will be intelligent and self-aware.” 

But a fuller read reveals Rosenberg’s focus on some of today’s hottest AI debates, including the potential for AGI in our lifetimes and why organizations need to prepare with AI ethics: “…while there’s an earnest effort in the AI community to push for safe technologies, there’s also a lack of urgency. That’s because too many of us wrongly believe that a sentient AI created by humanity will somehow be a branch of the human tree, like a digital descendant that shares a very human core. This is wishful thinking … the time to prepare is now.” 

DeepMind says ‘game over’ for AGI

Coincidentally, this past week was filled with claims, counterclaims and critiques of claims around the potential to realize AGI anytime soon.

Last Friday, Nando De Freitas, a lead researcher at Google’s DeepMind AI division, tweeted that “The Game is Over!” in the decades-long quest for AGI, after DeepMind unveiled its new Gato AI, which is capable of complex tasks ranging from stacking blocks to writing poetry. 

According to De Freitas, Gato AI simply needs to be scaled up in order to create an AI that rivals human intelligence. Or, as he wrote on Twitter, “It’s all about scale now! It’s all about making these models bigger, safer, compute efficient, faster at sampling, smarter memory, more modalities, innovative data, on/offline… Solving these challenges is what will deliver AGI.”

Pushback on AGI and scaling

Plenty of experts are pushing back on De Freitas’ claims and those of others insisting that AGI or its equivalent is at hand.

Yann LeCun, the French computer scientist who is chief AI scientist at Meta, had this to say (on Facebook, of course):

“About the raging debate regarding the significance of recent progress in AI, it may be useful to (re)state a few obvious facts:

(0) there is no such thing as AGI. Reaching “Human Level AI” may be a useful goal, but even humans are specialized.

(1) the research community is making some progress towards HLAI

(2) scaling up helps. It’s necessary but not sufficient, because….

(3) we are still missing some fundamental concepts

(4) some of those new concepts are possibly “around the corner” (e.g. generalized self-supervised learning)

(5) but we don’t know how many such new concepts are needed. We just see the most obvious ones.

(6) hence, we can’t predict how long it’s going to take to reach HLAI.“

Current AGI efforts as “alt intelligence”

Meanwhile, Gary Marcus, founder of Robust.AI and author of Rebooting AI, added to the debate on his new Substack, with its first post dedicated to the discussion of current efforts to develop AGI (including Gato AI), which he calls “alt intelligence”:

“Right now, the predominant strand of work within Alt Intelligence is the idea of scaling. The notion that the bigger the system, the closer we come to true intelligence, maybe even consciousness.

There is nothing new, per se, about studying Alt Intelligence, but the hubris associated with it is. I’ve seen signs for a while, in the dismissiveness with which the current AI superstars,= and indeed vast segments of the whole field of AI, treat human cognition, ignoring and even ridiculing scholars in such fields as linguistics, cognitive psychology, anthropology and philosophy.

But this morning I woke to a new reification, a Twitter thread that expresses, out loud, the Alt Intelligence creed, from Nando de Freitas, a brilliant high-level executive at DeepMind, Alphabet’s rightly-venerated AI wing, in a declaration that AI is “all about scale now.”

Marcus closes by saying:

“Let us all encourage a field that is open-minded enough to work in multiple directions, without prematurely dismissing ideas that happen to be not yet fully developed. It may just be that the best path to artificial (general) intelligence isn’t through Alt Intelligence, after all.

As I have written, I am fine with thinking of Gato as an “Alt Intelligence” — an interesting exploration in alternative ways to build intelligence — but we need to take it in context: it doesn’t work like the brain, it doesn’t learn like a child, it doesn’t understand language, it doesn’t align with human values and it can’t be trusted with mission-critical tasks.

It may well be better than anything else we currently have, but the fact that it still doesn’t really work, even after all the immense investments that have been made in it, should give us pause.”

AI alien invasion not arriving anytime soon (whew!)

It’s nice to know that most experts don’t believe the AGI alien invasion will arrive anytime soon.

But the fierce debate around AI and its ability to develop human-level intelligence will certainly continue – on social media and off.

Let me know your thoughts!

— Sharon Goldman, senior editor and writer

Twitter: @sharongoldman

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



Repost: Original Source and Author Link

Categories
AI

AI Weekly: AI researchers release toolkit to promote AI that helps to achieve sustainability goals

Hear from CIOs, CTOs, and other C-level and senior execs on data and AI strategies at the Future of Work Summit this January 12, 2022. Learn more


While discussions about AI often center around the technology’s commercial potential, increasingly, researchers are investigating ways that AI can be harnessed to drive societal change. Among others, Facebook chief AI scientist Yann LeCun and Google Brain cofounder Andrew Ng have argued that mitigating climate change and promoting energy efficiency are preeminent challenges for AI researchers.

Along this vein, researchers at the Montreal AI Ethics Institute have proposed a framework designed to quantify the social impact of AI through techniques like compute-efficient machine learning. An IBM project delivers farm cultivation recommendations from digital farm “twins” that simulate the future soil conditions of real-world crops. Other researchers are using AI-generated images to help visualize climate change, and nonprofits like WattTime are working to reduce households’ carbon footprint by automating when electric vehicles, thermostats, and appliances are active based on where renewable energy is available.

Seeking to spur further explorations in the field, a group at the Stanford Sustainability and Artificial Intelligence Lab this week released (to coincide with NeurIPS 2021) a benchmark dataset called SustainBench for monitoring sustainable development goals (SDGs) including agriculture, health, and education using machine learning. As the coauthors told VentureBeat in an interview, the goal is threefold: (1) lower the barriers to entry for researchers to contribute to achieving SDGs; (2) provide metrics for evaluating SDG-tracking algorithms, and (3) encourage the development of methods where improved AI model performance facilitates progress towards SDGs.

“SustainBench was a natural outcome of the many research projects that [we’ve] worked on over the past half-decade. The driving force behind these research projects was always the lack of large, high-quality labeled datasets for measuring progress toward the United Nations Sustainable Development Goals (UN SDGs), which forced us to come up with creative machine learning techniques to overcome the label sparsity,” the coauthors said. “[H]aving accumulated enough experience working with datasets from diverse sustainability domains, we realized earlier this year that we were well-positioned to share our expertise on the data side of the machine learning equation … Indeed, we are not aware of any prior sustainability-focused datasets with similar size and scale of SustainBench.”

Motivation

Progress toward SDGs has historically been measured through civil registrations, population-based surveys, and government-orchestrated censuses. However, data collection is expensive, leading many countries to go decades between taking measurements on SDG indicators. It’s estimated that only half of SDG indicators have regular data from more than half of the world’s countries, limiting the ability of the international community to track progress toward the SDGs.

“For example, early on during the COVID-19 pandemic, many developing countries implemented their own cash transfer programs, similar to the direct cash payments from the IRS in the United States. However … data records on household wealth and income in developing countries are often unreliable or unavailable,” the coauthors said.

Innovations in AI have shown promise in helping to plug the data gaps, however. Data from satellite imagery, social media posts, and smartphones can be used to train models to predict things like poverty, annual land cover, deforestation, agricultural cropping patterns, crop yields, and even the location and impact of natural disasters. For example, the governments of Bangladesh, Mozambique, Nigeria, Togo, and Uganda used machine learning-based poverty and cropland maps to direct economic aid to their most vulnerable populations during the pandemic.

But progress has been hindered by challenges, including a lack of expertise and dearth of data for low-income countries. With SustainBench, the Stanford researchers — along with contributors at Caltech, UC Berkeley, and Carnegie Mellon — hope to provide a starting ground for training machine learning models that can help measure SDG indicators and have a wide range of applications for real-world tasks.

SustainBench contains a suite of 15 benchmark tasks across seven SDGs taken from the United Nations, including good health and well-being, quality education, and clean water and sanitation. Beyond this, SustainBench offers tasks for machine learning challenges that cover 119 countries, each designed to promote the development of SDG measurement methods on real-world data.

The coauthors caution that AI-based approaches should supplement, rather than replace, ground-based data collection. They point out that ground truth data are necessary for training models in the first place, and that even the best sensor data can only capture some — but not all — of the outcomes of interest. But AI, they still believe, can be helpful for measuring sustainability indicators in regions where ground truth measurements are scarce or unavailable.

“[SDG] indicators have tremendous implications for policymakers, yet ‘key data are scarce, and often scarcest in places where they are most needed,’ as several of our team members wrote in a recent Science review article. By using abundant, cheap, and frequently updated sensor data as inputs, AI can help plug these data gaps. Such input data sources include publicly available satellite images, crowdsourced street-level images, Wikipedia entries, and mobile phone records, among others,” the coauthors said.

Future work

In the short term, the coauthors say that they’re focused on raising awareness of SustainBench within the machine learning community. Future versions of SustainBench are in the planning stages, potentially with additional datasets and AI benchmarks.

“Two technical challenges stand out to us. The first challenge is to develop machine learning models that can reason about multi-modal data. Most AI models today tend to work with single data modalities (e.g., only satellite images, or only text), but sensor data often comes in many forms … The second challenge is to design models that can take advantage of the large amount of unlabeled sensor data, compared to sparse ground truth labels,” the coauthors said. “On the non-technical side, we also see a challenge in getting the broader machine learning community to focus more efforts on sustainability applications … As we alluded to earlier, we hope SustainBench makes it easier for machine learning researchers to recognize the role and challenges of machine learning for sustainability applications.”

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: Recognition of bias in AI continues to grow

Hear from CIOs, CTOs, and other C-level and senior execs on data and AI strategies at the Future of Work Summit this January 12, 2022. Learn more


This week, the Partnership on AI (PAI), a nonprofit committed to responsible AI use, released a paper addressing how technology — particularly AI — can accentuate various forms of biases. While most proposals to mitigate algorithmic discrimination require the collection of data on so-called sensitive attributes — which usually include things like race, gender, sexuality, and nationality — the coauthors of the PAI report argue that these efforts can actually cause harm to marginalized people and groups. Rather than trying to overcome historical patterns of discrimination and social inequity with more data and “clever algorithms,” they say, the value assumptions and trade-offs associated with the use of demographic data must be acknowledged.

“Harmful biases have been found in algorithmic decision-making systems in contexts such as health care, hiring, criminal justice, and education, prompting increasing social concern regarding the impact these systems are having on the wellbeing and livelihood of individuals and groups across society,” the coauthors of the report write. “Many current algorithmic fairness techniques [propose] access to data on a ‘sensitive attribute’ or ‘protected category’ (such as race, gender, or sexuality) in order to make performance comparisons and standardizations across groups. [But] these demographic-based algorithmic fairness techniques [remove] broader questions of governance and politics from the equation.”

The PAI paper’s publication comes as organizations take a broader — and more critical — view of AI technologies, in light of wrongful arrestsracist recidivismsexist recruitment, and erroneous grades perpetuated by AI. Yesterday, AI ethicist Timnit Gebru, who was controversially ejected from Google over a study examining the impacts of large language models, launched the Distributed Artificial Intelligence Research (DAIR), which aims to ask question about responsible use of AI and recruit researchers from parts of the world rarely represented in the tech industry. Last week, the United Nations’ Educational, Scientific, and Cultural Organization (UNESCO) approved a series of recommendations for AI ethics, including regular impact assessments and enforcement mechanisms to protect human rights. Meanwhile, New York University’s AI Now Institute, the Algorithmic Justice League, and Data for Black Lives are studying the impacts and applications of AI algorithms, as are Khipu, Black in AI, Data Science Africa, Masakhane, and Deep Learning Indaba.

Legislators, too, are taking a harder look at AI systems — and their potential to harm. The U.K.’s Centre for Data Ethics and Innovation (CDEI) recently recommended that public sector organizations using algorithms be mandated to publish information about how the algorithms are being applied, including the level of human oversight. The European Union has proposed regulations that would ban the use of biometric identification systems in public and prohibit AI in social credit scoring across the bloc’s 27 member states. Even China, which is engaged in several widespread, AI-powered surveillance initiatives, has tightened its oversight of the algorithms that companies use to drive their business.

Pitfalls in mitigating bias

PAI’s work cautions that efforts to mitigate bias in AI algorithms will inevitably encounter roadblocks, however, due to the nature of algorithmic decision-making. If optimizing for a goal that’s poorly defined, it’s likely that a system will reproduce historical inequity — possibly under the guise of objectivity. Attempting to ignore societal differences across demographic groups will work to reinforce systems of oppression because demographic data coded in datasets has an enormous impact on the representation of marginalized peoples. But deciding how to classify demographic data is an ongoing challenge, as demographic categories continue to shift and change over time.

“Collecting sensitive data consensually requires clear, specific, and limited use as well as strong security and protection following collection. Current consent practices are not meeting this standard,” the PAI report coauthors wrote. “Demographic data collection efforts can reinforce oppressive norms and the delegitimization of disenfranchised groups … Attempts to be neutral or objective often have the effect of reinforcing the status quo.”

At a time when relatively few major research papers consider the negative impacts of AI, leading ethicists are calling on practitioners to pinpoint biases early in the development process. For example, a program at Stanford — the Ethics and Society Review (ESR) — requires AI researchers to evaluate their grant proposals for any negative impacts. NeurIPS, one of the largest machine learning conferences in the world, mandates that coauthors who submit papers state the “potential broader impact of their work” on society. And in a whitepaper published by the U.S. National Institute of Standards and Technology (NIST), the coauthors advocate for “cultural effective challenge,” a practice that seeks to create an environment where developers can question steps in engineering to help identify problems.

Requiring AI practitioners to defend their techniques can incentivize new ways of thinking and help create change in approaches by organizations and industries, the NIST coauthors posit.

“An AI tool is often developed for one purpose, but then it gets used in other very different contexts. Many AI applications also have been insufficiently tested, or not tested at all in the context for which they are intended,” NIST scientist Reva Schwartz, a coauthor of the NIST paper, wrote. “All these factors can allow bias to go undetected … [Because] we know that bias is prevalent throughout the AI lifecycle … [not] knowing where [a] model is biased, or presuming that there is no bias, would be dangerous. Determining methods for identifying and managing it is a vital … step.”

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: UN recommendations point to need for AI ethics guidelines

Hear from CIOs, CTOs, and other C-level and senior execs on data and AI strategies at the Future of Work Summit this January 12, 2022. Learn more


The U.N.’s Educational, Scientific, and Cultural Organization (UNESCO) this week approved a series of recommendations for AI ethics, which aim to recognize that AI can “be of great service” but also raise “fundamental … concerns.” UNESCO’s 193 member countries, including Russia and China, agreed to conduct AI impact assessments and place “strong enforcement mechanisms and remedial actions” to protect human rights.

“The world needs rules for artificial intelligence to benefit humanity. The recommendation[s] on the ethics of AI is a major answer,” UNESCO chief Audrey Azoulay said in a press release. “It sets the first global normative framework while giving States the responsibility to apply it at their level. UNESCO will support its … member states in its implementation and ask them to report regularly on their progress and practices.”

UNESCO’s policy document highlights the advantages of AI while seeking to reduce the risks that it entails. Toward this end, they address issues around transparency, accountability, and privacy in addition to data governance, education, culture, labor, health care, and the economy.

“Decisions impacting millions of people should be fair, transparent, and contestable,” UNESCO assistant director-general for social and human sciences Gabriela Ramos said in a statement. “These new technologies must help us address the major challenges in our world today, such as increased inequalities and the environmental crisis, and not deepening them.”

The recommendations follow on the heels of the European Union’s proposed regulations to govern the use of AI across the bloc’s 27 member states. They impose bans on the use of biometric identification systems in public, like facial recognition — with some exceptions. And they prohibit AI in social credit scoring, the infliction of harm (such as in weapons), and subliminal behavior manipulation.

The UNESCO recommendations also explicitly ban the use of AI for social scoring and mass surveillance, and they call for stronger data protections to provide stakeholders with transparency, agency, and control over their personal data. Beyond this, they stress that AI adopters should favor data, energy, and resource-efficient methods to help fight against climate change and tackle environmental issues.

Growing calls for regulation

While the policy is nonbinding, China’s support is significant because of the country’s historical — and current — stance on the use of AI surveillance technologies. According to the New York Times, the Chinese government — which has installed hundreds of millions of cameras across the country’s mainland — has piloted the use of predictive technology to sweep a person’s transaction data, location history, and social connections to determine whether they’re violent. Chinese companies such as Dahua and Huawei have developed facial recognition technologies, including several designed to target Uighurs, an ethnic minority widely persecuted in China’s Xinjiang province.

Underlining the point, contracts from the city of Zhoukou show that officials spend as much on surveillance as they do on education — and more than twice as much as on environmental protection programs.

Given China’s expressed intent to surveil 100% of public spaces within its borders, it seems unlikely to reverse course — UNESCO policy or not. But according to Ramos, the hope is that the recommendations, particularly the emphasis on addressing climate change, have an impact on the types of AI technologies that corporations, as well as governments, pursue.

“[UNESCO’s recommendations are] the code to change the [AI sector’s] business model, more than anything,” Ramos told Politico in an interview.

The U.S. isn’t a part of UNESCO and isn’t a signatory of the new recommendations. But bans on technologies like facial recognition have picked up steam across the U.S. at the local level. Facial recognition bans had been introduced in at least 16 states including Washington, Massachusetts, and New Jersey as of July. California lawmakers recently passed a law that will require warehouses to disclose the algorithms and metrics they use to track workers. A New York City bill bans employers from using AI hiring tools unless a bias audit can show that they won’t discriminate. And in Illinois, the state’s biometric information privacy act bans companies from obtaining and storing a person’s biometrics without their consent.

Regardless of their impact, the UNESCO recommendations signal growing recognition on the part of policymakers of the need for AI ethics guidelines.  The U.S. Department of Defense earlier this month published a whitepaper — circulated among National Oceanic and Atmospheric Administration, the Department of Transportation, ethics groups at the Department of Justice, the General Services Administration, and the Internal Revenue Service — outlining “responsible … guidelines” that establish processes intended to “avoid unintended consequences” in AI systems. NATO recently released an AI strategy listing the organization’s principles for “responsible use [of] AI.” And the U.S. National Institute of Standards and Technology is working with academia and the private sector to develop AI standards.

Regulation with an emphasis on accountability and transparency could go a long way toward restoring trust in AI systems. According to a survey conducted by KPMG, across five countries — the U.S., the U.K., Germany, Canada, and Australia — over a third of the general public says that they’re unwilling to trust AI systems in general. That’s not surprising, given that biases in unfettered AI systems have yielded wrongful arrestsracist recidivism scoressexist recruitmenterroneous high school grades, offensive and exclusionary language generators, and underperforming speech recognition systems, to name a few injustices.

“It is time for the governments to reassert their role to have good quality regulations, and incentivize the good use of AI and diminish the bad use,” Ramos continued.

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: Defense Department proposes new guidelines for developing AI technologies

Hear from CIOs, CTOs, and other C-level and senior execs on data and AI strategies at the Future of Work Summit this January 12, 2022. Learn more


This week, the Defense Innovation Unit (DIU), the division of the U.S. Department of Defense (DoD) that awards emerging technology prototype contracts, published a first draft of a whitepaper outlining “responsible … guidelines” that establish processes intended to “avoid unintended consequences” in AI systems. The paper, which includes worksheets for system planning, development, and deployment, is based on DoD ethics principles adopted by the Secretary of Defense and was written in collaboration with researchers at Carnegie Mellon University’s Software Engineering Institute, according to the DIU.

“Unlike most ethics guidelines, [the guidelines] are highly prescriptive and rooted in action,” a DIU spokesperson told VentureBeat via email. “Given DIU’s relationship with private sector companies, the ethics will help shape the behavior of private companies and trickle down the thinking.”

Launched in March 2020, the DIU’s effort comes as corporate defense contracts, particularly those involving AI technologies, have come under increased scrutiny. When news emerged in 2018 that Google had contributed to Project Maven, a military AI project to develop surveillance systems, thousands of employees at the company protested.

For some AI and data analytics companies, like Oculus cofounder Palmer Luckey’s Anduril and Peter Thiel’s Palantir, military contracts have become a top source of revenue. In October, Palantir won most of an $823 million contract to provide data and big analytics software to the U.S. army. And in July, Anduril said that it received a contract worth up to $99 million to supply the U.S. military with drones aimed at countering hostile or unauthorized drones.

Machine learning, computer vision, facial recognition vendors including TrueFace, Clearview AI, TwoSense, and AI.Reverie also have contracts with various U.S. army branches. And in the case of Maven, Microsoft and Amazon among others have taken Google’s place.

AI development guidance

The DIU guidelines recommend that companies start by defining tasks, success metrics, and baselines “appropriately,” identifying stakeholders and conducting harms modeling. They also require that developers address the effects of flawed data, establish plans for system auditing, and “confirm that new data doesn’t degrade system performance,” primarily through “harms assessment[s]” and quality control steps designed to mitigate negative impacts.

The guidelines aren’t likely to satisfy critics who argue that any guidance the DoD offers is paradoxical. As MIT Tech Review points out, the DIU says nothing about the use of autonomous weapons, which some ethicists and researchers as well as regulators in countries including Belgium and Germany have opposed.

But Bryce Goodman at the DIU, who coauthored the whitepaper, told MIT Tech Review that the guidelines aren’t meant to be a cure-all. For example, they can’t offer universally reliable ways to “fix” shortcomings such as biased data or inappropriately selected algorithms, and they might not apply to systems proposed for national security use cases that have no route to responsible deployment.

Studies indeed show that bias mitigation practices like those that the whitepaper recommend aren’t a panacea when it comes to ensuring fair predictions from AI models. Bias in AI also doesn’t arise from datasets alone. Problem formulation, or the way researchers fit tasks to AI techniques, can also contribute. So can other human-led steps throughout the AI deployment pipeline, like dataset selection and prep and architectural differences between models.

Regardless, the work could change how AI is developed by the government if the DoD’s guidelines are adopted by other departments. While NATO recently released an AI strategy and the U.S. National Institute of Standards and Technology is working with academia and the private sector to develop AI standards, Goodman told MIT Tech Review that he and his colleagues have already given the whitepaper to the National Oceanic and Atmospheric Administration, the Department of Transportation, and ethics groups at the Department of Justice, the General Services Administration, and the Internal Revenue Service.

The DIU says that it’s already deploying the guidelines on a range of projects covering applications including predictive health, underwater autonomy, predictive maintenance, and supply chain analysis. “There are no other guidelines that exist, either within the DoD or, frankly, the United States government, that go into this level of detail,” Goodman told MIT Tech Review.

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: The perils of AI analytics for police body cameras

Join gaming leaders online at GamesBeat Summit Next this upcoming November 9-10. Learn more about what comes next. 


In 2015, spurred by calls for greater police accountability, the federal government provided more than $23 million to local and tribal police agencies to expand their use of body cameras. As of 2016, 47% of the country’s roughly 15,300 general-purpose law enforcement agencies had purchased body cameras, according to a report by the Bureau of Justice Statistics, the most recent study measuring nationwide usage.

Evidence on their efficacy is mixed — a recent comprehensive review of 70 studies of body camera use found that they had no consistent or statistically significant effects — but advocates assert that body cameras can deter bad behavior on the part of officers while reducing the number of citizen complaints. However, an outstanding technological challenge with body cameras is making sense of the amount of footage that they produce. As per one estimate, the average officer’s body camera will record about 32 files, 7 hours, and 20GB of video per month at 720p resolution.

A relatively new startup, Truleo, claims to solve this problem with a platform that leverages AI to analyze body cam footage as it comes in. Truleo — which has raised $2.5 million in seed funding — converts the data into “actionable insights,” CEO and cofounder Anthony Tassone claims, using natural language processing and machine learning models to categorize incidents captured by the cameras.

The Seattle Police Department is one of the company’s early customers.

“Truleo analyzes the audio stream within body camera videos — we analyze the conversation between the officer and the public,” Tassone told VentureBeat via email. “We specifically highlight the ‘risky’ language the officer uses, which most often means surfacing directed profanity or using extremely rude language. However, we can also highlight officer shouting commands, so the command staff can evaluate the effectiveness of the subject compliance.”

Potentially flawed AI

Tassone says that Truleo’s AI models were built by its data scientists and law enforcement experts looking for “de-escalation, auto-flagging of incidents, or early warning for volatile interactions” to generate searchable reports. The models can recognize if a call is about drugs, theft, a foot chase, and if there’s profanity or shouting, he claims. Truleo quantifies the classifications as metrics, such as the percentage of “negative interactions” an officer has on a monthly basis and what police language is “effective.”

“Obviously, a call that ends in an arrest is going to be negative. But what if an officer has an overwhelming amount of negative interactions but a below-average number of arrests?  Is he or she going through something in their personal lives? Perhaps something deeply personal such as a divorce or maybe the officer was shot at last week.  Maybe they need some time off to cool down or to be coached by more seasoned officers. We want to help command staff be more proactive about identifying risky behavior and improving customer service tactics — before the officer loses their job or ends up on the news.”

But some experts are concerned about the platform’s potential for misuse, especially in the surveillance domain. “[Body cam] footage doesn’t just contain the attitude of the officer; it also contains all comments by the person they were interacting with, even when no crime was involved, and potentially conversations nearby,” University of Washington AI researcher Os Keyes told VentureBeat via email. “This is precisely the kind of thing that people were worried about when they warned about the implications of body cameras: police officers as moving surveillance cameras.”

Truleo

Above: Truleo’s analytics dashboard.

Image Credit: Truleo

Keyes also pointed out that natural language processing and sentiment analysis are far from perfect sciences. Aside from prototypes, AI systems struggle to recognize examples of sarcasm — particularly systems trained on text data alone. Natural language processing models can also exhibit prejudices along race, ethnic, and gender lines, for example associating “Black-aligned English” with higher toxicity or negative emotions like anger, fear, and sadness.

Speech recognition systems like the kind used by Truleo, too, can be discriminatory. In a study 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. More recently, the Algorithmic Justice League’s Voice Erasure project found that that speech recognition systems from Apple, Amazon, Google, IBM, and Microsoft collectively achieve word error rates of 35% for African American voices versus 19% for white voices.

“If it works, it’s dangerous. If it doesn’t work — which is far more likely — the very mechanism through which it is being developed and deployed is itself a reason to mistrust it, and the people using it,” Keyes said.

According to Tassone, Truleo consulted with officials on police accountability boards to define what interactions should be identified by its models to generate reports. To preserve privacy, the platform converts footage into an MP3 audio file during the upstream process “in memory” and deletes the stream after analysis in AWS GovCloud, writing nothing to disk.

“Truleo’s position is that this data 100% belongs to the police department,” Tassone added. “We aim to accurately transcribe about 90% of the audio file correctly … More importantly, we classify the event inside the audio correctly over 99% of the time … When customers look at their transcripts, if anything is incorrect, they can make those changes in our editor and submit them back to Truleo, which automatically trains new models with these error corrections.”

When contacted for comment, Axon, one of the world’s largest producers of police body cameras, declined to comment on Truleo’s product but said: “Axon is always exploring technologies that have [the] potential for protecting lives and improving efficiency for our public safety customers. We gear towards developing responsible and ethical solutions that are reliable, secure, and privacy-preserving.”

In a recent piece for Security Info Watch, Anthony Treviño, the former assistant chief of police for San Antonio, Texas and a Truleo advisor, argued that AI-powered body cam analytics platforms could be used as a teaching tool for law enforcement. “For example, if an agency learns through body camera audio analytics that a certain officer has a strong ability to de-escalate or control deadly force during volatile situations, the agency can use that individual as a resource to improve training across the entire force,” he wrote.

Given AI’s flaws and studies showing that body cams don’t reduce police misconduct on their own, however, Treviño’s argument would appear to lack merit. “Interestingly, although their website includes a lot of statistics about time and cost savings, it doesn’t actually comment on whether it changes the outcomes in any way,” AI researcher at the Queen Mary University of London, Mike Cook, told VentureBeat via email. “Truleo claims they provide ‘human accuracy at scale’ — but if we already doubt the existing accuracy provided by the humans involved, what good is it to replicate it at scale? What good is a 50% reduction in litigation time if it leads to the same amount of unjust, racist, or wrongful police actions? A faster-working unfair system is still unfair.”

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: The intractable challenge of bias in AI

Last week, Twitter shared research showing that the platform’s algorithms amplify tweets from right-of-center politicians and news outlets at the expense of left-leaning sources. Rumman Chowdhury, the head of Twitter’s machine learning, ethics, transparency, and accountability team, said in an interview with Protocol that while some of the behavior could be user-driven, the reason for the bias isn’t entirely clear.

“We can see that it is happening. We are not entirely sure why it is happening,” Chowdhury said. “When algorithms get put out into the world, what happens when people interact with it — we can’t model for that. We can’t model for how individuals or groups of people will use Twitter, what will happen in the world in a way that will impact how people use Twitter.”

Twitter’s forthcoming root-cause analysis will likely turn up some of the origins of its recommendation algorithms’ rightward tilt. But Chowdhury’s frank disclosure highlights the unknowns about biases in AI models and how they occur — and whether it’s possible to mitigate them.

The challenge of biased models

The past several years have established that bias mitigation techniques aren’t a panacea when it comes to ensuring fair predictions from AI models. Applying algorithmic solutions to social problems can magnify biases against marginalized peoples, and undersampling populations always results in worse predictive accuracy. For example, even leading language models like OpenAI’s GPT-3 exhibit toxic and discriminatory behavior, usually traceable back to the dataset creation process. When trained on biased datasets, models acquire and exacerbate biases, like flagging text by Black authors as more toxic than text by white authors.

Bias in AI doesn’t arise from datasets alone. Problem formulation, or the way researchers fit tasks to AI techniques, can also contribute. So can other human-led steps throughout the AI deployment pipeline.

A recent study from Cornell and Brown University investigated the problems around model selection, or the process by which engineers choose machine learning models to deploy after training and validation. The paper notes that while researchers may report average performance across a small number of models, they often publish results using a specific set of variables that can obscure a model’s true performance. This presents a challenge because other model properties can change during training. Seemingly minute differences in accuracy between groups can multiply out to large groups, impacting fairness with regard to specific demographics.

The study’s coauthors underline a case study in which test subjects were asked to choose a “fair” skin cancer detection model based on metrics they identified. Overwhelmingly, the subjects selected a model with the highest accuracy — even though it exhibited the largest gender disparity. This is problematic on its face because the accuracy metric doesn’t provide a breakdown of false negatives (missing a cancer diagnosis) and false positives (mistakenly diagnosing cancer when it’s not actually present), the researchers assert. Including these metrics could have biased the subjects to make different choices concerning which model was “best.”

Architectural differences between algorithms can also contribute to biased outcomes. In a paper accepted to the 2020 NeurIPS conference, Google and Stanford researchers explored the bias exhibited by certain kinds of computer vision algorithms — convolutional neural networks (CNNs) — trained on the open source ImageNet dataset. Their work indicates that CNNs’ bias toward textures may come not from differences in their internal workings but from differences in the data that they see: CNNs tend to classify objects according to material (e.g. “checkered”) and humans to shape (e.g. “circle”).

Given the various factors involved, it’s not surprising that 65% of execs can’t explain how their company’s models make decisions.

While challenges in identifying and eliminating bias in AI are likely to remain, particularly as research uncovers flaws in bias mitigation techniques, there are preventative steps that can be taken. For instance, a study from a team at Columbia University found that diversity in data science teams is key in reducing algorithmic bias. The team found that, while individually, everyone is more or less equally biased, across race, gender, and ethnicity, males are more likely to make the same prediction errors. This indicates that the more homogenous the team is, the more likely it is that a given prediction error will appear twice.

“Questions about algorithmic bias are often framed as theoretical computer science problems. However, productionized algorithms are developed by humans, working inside organizations, who are subject to training, persuasion, culture, incentives, and implementation frictions,” the researchers wrote in their paper.

In light of other studies suggesting that the AI industry is built on geographic and social inequalities; that dataset prep for AI research is highly inconsistent; and that few major AI researchers discuss the potential negative impacts of their work in published papers, a thoughtful approach to AI deployment is becoming increasingly critical. A failure to implement models responsibly could — and has — led to uneven health outcomes, unjust criminal sentencing, muzzled speech, housing and lending discrimination, and even disenfranchisement. Harms are only likely to become more common if flawed algorithms proliferate.

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: AI model training costs on the rise, highlighting need for new solutions

This week, Microsoft and Nvidia announced that they trained what they claim is one of the largest and most capable AI language models to date: Megatron-Turing Natural Language Generation (MT-NLP). MT-NLP contains 530 billion parameters — the parts of the model learned from historical data — and achieves leading accuracy in a broad set of tasks, including reading comprehension and natural language inferences.

But building it didn’t come cheap. Training took place across 560 Nvidia DGX A100 servers, each containing 8 Nvidia A100 80GB GPUs. Experts peg the cost in the millions of dollars.

Like other large AI systems, MT-NLP raises questions about the accessibility of cutting-edge research approaches in machine learning. AI training costs dropped 100-fold between 2017 and 2019, but the totals still exceed the compute budgets of most startups, governments, nonprofits, and colleges. The inequity favors corporations and world superpowers with extraordinary access to resources at the expense of smaller players, cementing incumbent advantages.

For example, in early October, researchers at Alibaba detailed M6-10T, a language model containing 10 trillion parameters (roughly 57 times the size of OpenAI’s GPT-3) trained across 512 Nvidia V100 GPUs for 10 days. The cheapest V100 plan available through Google Cloud Platform costs $2.28 per hour, which would equate to over $300,000 ($2.28 per hour multiplied by 24 hours over 10 days) — further than most research teams can stretch.

Google subsidiary DeepMind is estimated to have spent $35 million training a system to learn the Chinese board game Go. And when the company’s researchers designed a model to play StarCraft II, they purposefully didn’t try multiple ways of architecting a key component because the training cost would have been too high. Similarly, OpenAI didn’t fix a mistake when it implemented GPT-3 because the cost of training made retraining the model infeasible.

Paths forward

It’s important to keep in mind that training costs can be inflated by factors other than an algorithm’s technical aspects. As Yoav Shoham, Stanford University professor emeritus and cofounder of AI startup AI21 Labs, recently told Synced, personal and organizational considerations often contribute to a model’s final price tag.

“[A] researcher might be impatient to wait three weeks to do a thorough analysis and their organization may not be able or wish to pay for it,” he said. “So for the same task, one could spend $100,000 or $1 million.”

Still, the increasing cost of training — and storing — algorithms like Huawei’s PanGu-Alpha, Naver’s HyperCLOVA, and the Beijing Academy of Artificial Intelligence’s Wu Dao 2.0 is giving rise to a cottage industry of startups aiming to “optimize”  models without degrading accuracy. This week, former Intel exec Naveen Rao launched a new company, Mosaic ML, to offer tools, services, and training methods that improve AI system accuracy while lowering costs and saving time. Mosaic ML — which has raised $37 million in venture capital — competes with Codeplay Software, OctoML, Neural Magic, Deci, CoCoPie, and NeuReality in a market that’s expected to grow exponentially in the coming years.

In a sliver of good news, the cost of basic machine learning operations has been falling over the past few years. A 2020 OpenAI survey found that since 2012, the amount of compute needed to train a model to the same performance on classifying images in a popular benchmark — ImageNet — has been decreasing by a factor of two every 16 months.

Approaches like network pruning prior to training could lead to further gains. Research has shown that parameters pruned after training, a process that decreases the model size, could have been pruned before training without any effect on the network’s ability to learn. Called the “lottery ticket hypothesis,” the idea is that the initial values parameters in a model receive are crucial for determining whether they’re important. Parameters kept after pruning receive “lucky” initial values; the network can train successfully with only those parameters present.

Network pruning is far from a solved science, however. New ways of pruning that work before or in early training will have to be developed, as most current methods apply only retroactively. And when parameters are pruned, the resulting structures aren’t always a fit for the training hardware (e.g., GPUs), meaning that pruning 90% of parameters won’t necessarily reduce the cost of training a model by 90%.

Whether through pruning, novel AI accelerator hardware, or techniques like meta-learning and neural architecture search, the need for alternatives to unattainably large models is quickly becoming clear. A University of Massachusetts Amherst study showed that using 2019-era approaches, training an image recognition model with a 5% error rate would cost $100 billion and produce as much carbon emissions as New York City does in a month. As IEEE Spectrum’s editorial team wrote in a recent piece, “we must either adapt how we do deep learning or face a future of much slower progress.”

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: Companies look to ‘scale up’ their use of AI

All the sessions from Transform 2021 are available on-demand now. Watch now.


Companies are looking to move from small-scale, proof-of-concept AI deployments to operating AI at a massive scale. That’s according to Bratin Saha, VP and general manager at Amazon AI, who spoke with VentureBeat in a recent phone interview about general trends in the AI industry.

The pandemic supercharged the adoption of AI, in part because it caused companies to digitally transform their lines of business. A 2021 survey commissioned by IBM found that almost one-third of business surveyed are now using AI, with 43% reporting that they accelerated existing rollouts.

“We’re seeing [companies] saying, ‘How do I make [AI] a systematic engineering discipline? How do we standardize this? How do we introduce the right tools and procedures to make this pervasive?,” Saha said. “[Companies] want to go beyond deploying a handful of [AI] models to deploying thousands of models — in fact, [AWS has] one customer that wants to deploy a million models.”

Scaling up AI

The risks of failing to scale AI are substantial. Accenture estimates that it could put 75% of organizations out of business, particularly if they execute the shift from experimentation to execution too slowly. In a 2019 report, the research firm found that 84% of C-level executives believe they won’t achieve their business strategy without scaling AI, yet only 16% had made inroads creating an organization powered by “robust AI capabilities.”

“You can use machine learning to do some really cool demos … and they’re really compelling. But those demos … are expensive,” Saha said. “The high-profile …  demos can capture the imagination, but they they’re not repeatable — they come at high costs and they’re not really providing [business value].”

One of the steps enterprises must take in scaling AI is ensuring they have “good-quality” data, Saha said. Another key ingredient for success is creating a standardized set of tools, which includes software and hardware infrastructure for building and training models.

“The cloud becomes a very important factor [in this,] because it makes it easy for [companies] to standardize … [using the] same set of rules and tools and processes,” Saha added.

Indeed, the cloud is increasingly factoring into enterprise AI scaling efforts. This is due to its potential to improve training and inferencing performance, while also lowering costs in some cases. Even companies that have private datacenters often opt to avoid ramping up the hardware, networking, and data storage required to host big data and AI applications.

Hybrid cloud adoption

The cloud isn’t the end-all be-all when it comes to ramping up deployments of AI, however. Hybrid cloud approaches have come into vogue as companies look to complement their homegrown infrastructure with highly scalable public clouds. For example, a hybrid AI app might tap an on-premises database while running app code both in the on-premises private cloud and scaling to the public cloud when demand increases.

It’s clear that challenges remain in scaling AI. An MIT Technology Review and Databricks report found that just 13% of organizations are delivering on their data strategy, owing to issues around managing the end-to-end lifecycle. Other surveys cite a lack of executive buy-in as a top reason for AI deployment failures. And yet others attribute it to a lack of institutional knowledge about machine learning modeling and data science, data engineering, and business use cases.

But Saha — who has a vested interest in the success of AI services as they relate to Amazon and Amazon Web Services, it should be noted — is optimistic about the future. He points out that companies are using AI for use cases from personalizing their products to forecasting demand supply chain. They’re also using computer vision and a lot of natural language processing technologies, including chatbots and intelligent document processing. “What [I] see coming down the road is the industrialization of machine learning…” he said. “[It’s] leading to explosive growth in machine learning.”

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

Repost: Original Source and Author Link

Categories
AI

AI Weekly: AI adoption is driving cloud growth

All the sessions from Transform 2021 are available on-demand now. Watch now.


The adoption of cloud technologies continues to accelerate. According to the newest report from Canalys, in Q2 2021, companies spent $5 billion more on cloud infrastructure services compared to the previous quarter. While a number of factors are responsible, including an increased focus on business resiliency planning, the uptick illustrates the effect AI’s embracement has had — and continues to have — on enterprise IT budgets.

In a recent survey, 80% of U.S. enterprises said they accelerated their AI adoption over the past two years. A majority consider AI to be important in their digital transformation efforts and intend to set aside between $500,000 to $5 million per year for deployment efforts. Organizations were projected to invest more than $50 billion in AI systems globally in 2020, according to IDC, up from $37.5 billion in 2019. And by 2024, investment is expected to reach $110 billion.

The cloud is playing a role in this due to its potential to improve AI training and inferencing performance, lowering costs and in some cases providing enhanced protection against attacks. Most companies lack the infrastructure and expertise to implement AI applications themselves. As TierPoint highlights, outside of corporate datacenters, only public cloud infrastructure can support massive data storage as well as the scalable computing capability needed to crunch large amounts of data and AI algorithms. Even companies that have private datacenters often opt to avoid ramping up the hardware, networking, and data storage required to host big data and AI applications. According to Accenture global lead of applied intelligence Sanjeev Vohra, who spoke during VentureBeat’s Transform 2021 conference, the cloud and data have come together to give companies a higher level of compute, power, and flexibility.

Cloud vendor boost

Meanwhile, cloud vendors are further stoking the demand for AI by offering a number of tools and services that make it easier to develop, test, enhance, and operate AI systems without big upfront investments. These include hardware optimized for machine learning, APIs that automate speech recognition and text analysis, productivity-boosting automated machine learning modeling systems, and AI development workflow platforms. In a 2019 whitepaper, Deloitte analysts gave the example of Walgreens, which sought to use Microsoft’s Azure AI platform to develop new health care delivery models. One of the world’s largest shipbuilders is using Amazon Web Services to develop and manage autonomous cargo vessels, the analysts also noted. And the American Cancer Society uses Google’s machine learning cloud services for automated tissue image analysis.

“The symbiosis between cloud and AI is accelerating the adoption of both,” the analysts wrote. “Indeed, Gartner predicts that through 2023, AI will be one of the top workloads that drive IT infrastructure decisions. Technology market research firm Tractica forecasts that AI will account for as much as 50% of total public cloud services revenue by 2025: AI adoption means that, ‘essentially, another public cloud services market will be added on top of the current market.’”

With the global public cloud computing market set to exceed $362 billion in 2022 and the average cloud budget reaching $2.2 million today, it appears clear that investments in the cloud aren’t about to slow down anytime soon. As long as AI’s trajectory remains bright — and it should — the cloud industry will have an enormous boom from which to benefit.

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

Repost: Original Source and Author Link