Cloud Hosting

Vendors of artificial intelligence technology should not be shielded by intellectual property claims and will have to disclose elements of their designs and be able to explain how their offering works in order to establish accountability, according to a leading official from the Cybersecurity and Infrastructure Security Agency.

I dont know how you can have a black-box algorithm thats proprietary and then be able to deploy it and be able to go off and explain whats going on, said Martin Stanley, a senior technical advisor who leads the development of CISAs artificial intelligence strategy. I think those things are going to have to be made available through some kind of scrutiny and certification around them so that those integrating them into other systems are going to be able to account for whats happening.

Stanley was among the speakers on a recent Nextgov and Defense One panel where government officials, including a member of the National Security Commission on Artificial Intelligence, shared some of the ways they are trying to balance reaping the benefits of artificial intelligence with risks the technology poses.

Experts often discuss the rewards of programming machines to do tasks humans would otherwise have to labor onfor both offensive and defensive cybersecurity maneuversbut the algorithms behind such systems and the data used to train them into taking such actions are also vulnerable to attack. And the question of accountability applies to users and developers of the technology.

Artificial intelligence systems are code that humans write, but they exercise their abilities and become stronger and more efficient using data that is fed to them. If the data is manipulated, or poisoned, the outcomes can be disastrous.

Changes to the data could be things that humans wouldnt necessarily recognize, but that computers do.

Weve seen ... trivial alterations that can throw off some of those results, just by changing a few pixels in an image in a way that a person might not even be able to tell, said Josephine Wolff, a Tufts University cybersecurity professor who was also on the panel.

And while its true that behind every AI algorithm is a human coder, the designs are becoming so complex, that youre looking at automated decision-making where the people who have designed the system are not actually fully in control of what the decisions will be, Wolff says.

This makes for a threat vector where vulnerabilities are harder to detect until its too late.

With AI, theres much more potential for vulnerabilities to stay covert than with other threat vectors, Wolff said. As models become increasingly complex it can take longer to realize that something is wrong before theres a dramatic outcome.

For this reason, Stanley said an overarching factor CISA uses to help determine what use cases AI gets applied to within the agency, is to assess the extent to which they offer high benefits and low regrets.

We pick ones that are understandable and have low complexity, he said.

Among other things federal personnel need to be mindful of is who has access to the training data.

You can imagine you get an award done, and everyone knows how hard that is from the beginning, and then the first thing that the vendor says is OK, send us all your data, hows that going to work so we can train the algorithm? he said. Those are the kinds of concerns that we have to be able to address.

Were going to have to continuously demonstrate that we are using the data for the purpose that it was intended, he said, adding, Theres some basic science that speaks to how you interact with algorithms and what kind of access you can have to the training data. Those kinds of things really need to be understood by the people who are deploying them.

A crucial but very difficult element to establish is liability. Wolff said ideally, liability wouldbe connected to a potential certification program where an entity audits artificial intelligence systems for factors like transparency and explainability.

Thats important, she said, for answering the question of how can we incentivize companies developing these algorithms to feel really heavily the weight of getting them right and be sure to do their own due diligence knowing that there are serious penalties for failing to secure them effectively.

But this is hard, even in the world of software development more broadly.

Making the connection is still very unresolved. Were still in the very early stages of determining what would a certification process look like, who would be in charge of issuing it, what kind of legal protection or immunity might you get if you went through it, she said. Software developers and companies have been working for a very long time, especially in the U.S., under the assumption that they cant be held legally liable for vulnerabilities in their code, and when we start talking about liability in the machine learning and AI context, we have to recognize that thats part of what were grappling with, an industry that for a very long time has had very strong protections from any liability.

View from the Commission

Responding to this, Katharina McFarland, a member of the National Security Commission on Artificial Intelligence, referenced the Pentagons Cybersecurity Maturity Model Certification program.

The point of the CMMC is to establish liability for Defense contractors, Defense Acquisitions Chief Information Security Officer Katie Arrington has said. But McFarland highlighted difficulties facing CMMC that program officials themselves have acknowledged.

Im sure youve heard of the [CMMC], theres a lot of thought going on, the question is the policing of it, she said. When you consider the proliferation of the code thats out there, and the global nature of it, you really will have a challenge trying to take a full thread and to pull it through a knothole to try to figure out where that responsibility is. Our borders are very porous and machines that we buy from another nation may not be built with the same biases that we have.

McFarland, a former head of Defense acquisitions, stressed that AI is more often than not viewed with fear and said she wanted to see more of a balance in procurement considerations for the technology.

I found that we had a perverse incentive built into our system and that was that we took, sometimes, I think extraordinary measures to try to creep into the one percent area for failure, she said, In other words, we would want to 110% test a system and in doing so, we might miss the venue of where its applicability in a theater to protect soldiers, sailors, airmen and Marines is needed.

She highlighted upfront a need for testing a verification but said it shouldnt be done at the expense of adoption. To that end, she asks that industry help by sharing the testing tools they use.

I would encourage industry to think about this from the standpoint of what tools would we needbecause theyre using themin the department, in the federal space, in the community, to give us transparency and verification, she said, so that we have a high confidence in the utility, in the data that were using and the AI algorithms that were building.

See the rest here:
Artificial Intelligence Systems Will Need to Have Certification, CISA Official Says - Nextgov

In 2018, Congress established the National Security Commission on Artificial Intelligence (NSCAI)a temporary, independent body tasked with reviewing the national security implications of artificial intelligence (AI). But two years later, the commissions activities remain little known to the public. Critics have charged that the commission has conducted activities of interest to the public outside of the public eye, only acknowledging that meetings occurred after the fact and offering few details on evolving commission decision-making. As one commentator remarked, Companies or members of the public interested in learning how the Commission is studying AI are left only with the knowledge that appointed people met to discuss these very topics, did so, and are not yet releasing any information about their recommendations.

That perceived lack of transparency may soon change. In June, the U.S. District Court for the District of Columbia handed down its decision in Electronic Privacy Information Center v. National Security Commission on Artificial Intelligence, holding that Congress compelled the NSCAI to comply with the Federal Advisory Committee Act (FACA). Under FACA, the commission must hold open meetings and proactively provide records and other materials to the public. This decision follows a ruling from December 2019, holding that the NSCAI must also provide historical documents upon request under the Freedom of Information Act (FOIA). As a result of these decisions, the public is likely to gain increased access to and insight into the once-opaque operations of the commission.

Lawmakers established the NSCAI in the John S. McCain National Defense Authorization Act (NDAA) for fiscal 2019 1051, which tasked the commission with consider[ing] the methods and means necessary to advance the development of artificial intelligence, machine learning, and associated technologies to comprehensively address the national security and defense needs of the United States. The commissions purview includes an array of issues related to the implications and uses of artificial intelligence and machine learning for national security and defense, including U.S. competitiveness and leadership, research and development, ethics, and data standards.

The NSCAI is currently chaired by Eric Schmidt, the former executive chairman of Googles parent company, Alphabet. The commissions 15 membersappointed by a combination of Congress, the secretary of defense and the secretary of commercereceive classified and unclassified briefings, meet in working groups and engage with industry. They report their findings and recommendations to the president and Congress, including in an annual report.

The Electronic Privacy Information Center (EPIC), a research center focused on privacy and civil liberties issues in the digital age, submitted a request to the NSCAI in September 2019, seeking access to upcoming meetings and records prepared by the commission under FACA and FOIA. In the six-month period prior to the request, the NSCAI held more than a dozen meetings and received over 100 briefings, according to EPIC. At the time it filed the lawsuit, EPIC noted that the commissions first major report was also one month overdue for release. When the commission did not comply with the requests under FOIA and FACA, EPIC brought suit under the two laws.

EPICs complaint alleged that the NSCAI had conducted its operations opaquely in its short lifespan. Since its establishment, the commission has operated almost entirely in secret with meetings behind closed doors[,] and has failed to publish or disclose any notices, agendas, minutes, or materials. If Congress had intended the NSCAI to comply with FOIA and FACA, such activity would not satisfy the statutes requirements. Given the potential implications of federal artificial intelligence decisions for privacy, cybersecurity, human rights, and algorithmic bias, EPIC argued that [p]ublic access to the records and meetings of the AI Commission is vital to ensure government transparency and democratic accountability. The complaint also noted the potential ramifications of commission activities for the government, private sector, and public, as well as the importance of artificial intelligence safeguards in the national security context due to limited public oversight. According to EPIC, increasing public participation would permit greater input into the development of national AI policy by those whose privacy and data security could potentially be affected.

The U.S. District Court for the District of Columbia addressed EPICs FOIA claim in a December 2019 decision. FOIA requires agencies to disclose their records to a party upon request, barring exemptions (including for information classified to protect national security). EPIC alleged that the NSCAI failed to uphold its obligations under FOIA to process FOIA requests in a timely fashion; to process EPICs FOIA requests in an expedited manner, in accordance with EPICs claims of urgency; and to make available for public inspection and copying its records, reports, transcripts, minutes, appendixes, working papers, drafts, studies, agenda, or other documents. The commission, which at the time did not have a FOIA processing mechanism in place or other pending FOIA requests, argued that it was not an agency subject to FOIA.

The courts inquiry centered on whether the NSCAI is an agency under FOIA. Comparing the language establishing the NSCAI with FOIAs definition of agency, the court held that the NSCAI is subject to FOIA. In his decision, District Judge Trevor McFadden noted that Congress could have hardly been clearer. As a result, since that time, the commission has had to produce historical documents in response to FOIA requests.

FACA, by contrast, applies forward-looking requirements specifically to federal advisory committees. These mandates include requiring committees to open meetings to the public and announce them in the Federal Register, and to make reports, transcripts and other commission materials publicly available. The measures aim to inform the public about and invite public engagement with the committees that provide expertise to the executive branch. EPIC alleged that the NSCAI violated FACA by failing to hold open meetings and provide notice of them, and by failing to make records available to the public. EPIC sought mandamus relief pursuant to the alleged FACA violations.

In its June decision, the district court ruled that FACA applies to the NSCAI. The commission had filed a motion to dismiss the FACA claims, arguing that it could not be subject to both FOIA and FACA. Since the court had previously held the NSCAI to be an agency for purposes of FOIA, the commission reasoned that it could not simultaneously be an advisory committee under FACA. District Judge McFadden disagreed. Invoking the Roman God Januss two facesone forward-looking and the other backward-facinghe wrote, [L]ike Janus, the Commission does indeed have two faces, and ... Congress obligated it to comply with FACA as well as FOIA. The court could not identify a conflict between the requirements of the two statutes, despite differences in their obligations and exceptions. Rather, it noted that if such conflicts arise, it will be incumbent on the parties and the Court to resolve any difficulties. The court dismissed additional claims under the Administrative Procedure Act (APA) for lack of subject matter jurisdiction, as it determined that the commission is not an agency under the APA definition.

The courts decision turned on whether the NSCAI is an advisory committee subject to FACA. The court determined that the statutory text of the 2019 NDAA establishing the NSCAI fit[s] the [FACA] definition of advisory committee like a glove. Furthermore, turning to the full text of the 2019 NDAA, the court noted that the law contains at least two instances in which it explicitly exempts a government body from FACA. The court read the 2019 NDAA as silent when FACA applies and explicit when FACA does not apply. Given Congresss silence on the applicability of FACA to the NSCAI in the 2019 NDAAand again in the 2020 NDAAthe court reasoned that Congress intended the NSCAI to be subject to FACA.

In determining the NSCAI to be subject to FACA, in addition to FOIA, the court has compelled the commission to adopt a more transparent operating posture going forward. Since the December 2019 decision on FOIA, the NSCAI has produced a number of historical records in response to FOIA requests. The recent ruling on FACA grounds requires the NSCAI to hold open meetings, post notice of meetings in advance and make documents publicly available. As a result, the commissions process of compiling findings and developing recommendations for government action related to artificial intelligence and machine learning will likely become more accessible to the public.

The two court decisions come in time to have a noticeable impact on the remaining term of the temporary commission. While the NSCAI was previously due to disband later in 2020, the NDAA for fiscal 2020 1735 extended the commissions lifespan by one year, to October 1, 2021. Citing federal budgetary timelines and the pace of AI development, the commission released its first set of recommendations in March 2020 and expressed its intent to publish additional recommendations on a quarterly basis thereafter. The commission is due to submit its final report to Congress by March 1, 2021. As the NSCAI prepares to enter its final year of operations and develop its closing recommendations, the public will have a clearer window into the commissions work.

View original post here:
Increasing Transparency at the National Security Commission on Artificial Intelligence - Lawfare

While artificial intelligence systems continue to make huge strides forward, they're still not particularly good at dealing with chaos or unpredictability. Now researchers think they have found a way to fix this, by teaching AI about physics.

To be more specific, teaching them about the Hamiltonian function, which gives the AI information about the entirety of a dynamic system: all the energy contained within it, both kinetic and potential.

Neural networks, designed to loosely mimic the human brain as a complex, carefully weighted type of AI, then have a 'bigger picture' view of what's happening, and that could open up possibilities for getting AI to tackle harder and harder problems.

"The Hamiltonian is really the special sauce that gives neural networks the ability to learn order and chaos," says physicist John Lindner, from North Carolina State University.

"With the Hamiltonian, the neural network understands underlying dynamics in a way that a conventional network cannot. This is a first step toward physics-savvy neural networks that could help us solve hard problems."

The researchers compare the introduction of the Hamiltonian function to a swinging pendulum it's giving AI information about how fast the pendulum is swinging and its path of travel, rather than just showing AI a snapshot of the pendulum at one point in time.

If neural networks understand the Hamiltonian flow so where the pendulum is, in this analogy, where it might be going, and the energy it has then they are better able to manage the introduction of chaos into order, the new study found.

Not only that, but they can also be built to be more efficient: better able to forecast dynamic, unpredictable outcomes without huge numbers of extra neural nodes. It helps AI to quickly get a more complete understanding of how the world actually works.

A representation of the Hamiltonian flow, with rainbow colours coding a fourth dimension. (North Carolina State University)

To test their newly improved AI neural network, the researchers put it up against a commonly used benchmark called the Hnon-Heiles model, initially created to model the movement of a star around a sun.

The Hamiltonian neural network successfully passed the test, correctly predicting the dynamics of the system in states of order and of chaos.

This improved AI could be used in all kinds of areas, from diagnosing medical conditions to piloting autonomous drones.

We've already seen AI simulate space, diagnose medical problems, upgrade movies and develop new drugs, and the technology is, relatively speaking, just getting started there's lots more on the way. These new findings should help with that.

"If chaos is a nonlinear 'super power', enabling deterministic dynamics to be practically unpredictable, then the Hamiltonian is a neural network 'secret sauce', a special ingredient that enables learning and forecasting order and chaos," write the researchers in their published paper.

The research has been published in Physical Review E.

See the rest here:
Artificial Intelligence Can't Deal With Chaos, But Teaching It Physics Could Help - ScienceAlert

"With the combination of technologies in the new AUTOSORT, we are able to see a different kind of chemical (makeup) of the same material," explains Atienza. For example, for different PET bottles and trays, "It's all PET for NIR, but combined with the other technologies, we are able to see the difference.

"We are able to identify and sort, for example, different kinds of PET, such as trays, bottles, etc., which is very important for recyclers, because the quality of the end product can change when you have a lot of trays on the PET line. We definitely see more and more PET trays on the market," which he points to as one of the drivers for the development of this technology. Key to taking material detection to the next level is DEEP LAISER, which is an optional add-on to the system.

According to TOMRA, the DEEP LAISER is one of the first fully integrated systems of its kind, and stands out for its compactness and flexible range of uses, with object recognition that enables a deeper sorting sharpness. This includes the use of Artificial Intelligence (AI) via Deep Learningwhich TOMRA first introduced in 2019.

In basic terms, FLYING BEAM's NIR/VIS (Visual Spectrum) technology makes an overall determination of what kind of plastic and colour is on the belt, and SHARP EYE looks more closely at the chemical composition, to give a deeper sort, differentiating between different types of PET, for example. When you add on the DEEP LAISER component, it detects material that other optical sorters cannot detect.

"It is a laser combined with an artificially intelligent camera, using our IOR software," explains Atienza. "With DEEP LAISER we are sorting material on the conveyor that the FLYING BEAM and SHARP EYE cannot detect, such as black plastic or glass." He says it detects and sorts these materials out as an impurity, and that due to the IOR, the system sees, and is able to learn about, a very wide range of different types of material, detecting chemical composition, shapes and colour.

"With our AUTOSORT technology we can see types of materials that many other technologies cannot see. For example, our machine can differentiate LDPE, (low density polyethylene) from linear LLDPE. It's mostly the same material, because it is all LDPE, but they have differences, and we are able to see it, even at 6m/second."

Another example is LDPE and HDPE film, he says. "With films, you have low-density and high-density polyethylene, and we can see that difference, because our sensor has very high detection sensitivity."

In addition to its vastly increased sorting capacity overall, the new AUTOSORT ValveBlock geometry system is also 25% more powerful, with a reduction in air consumption of about 15%, compared to the previous version. This contributes to lower operating costs, including less electrical consumption, and notably this new generation AUTOSORT also auto calibrates every millisecond for optimal operation through changing conditions.

"This is very important," says Atienza, especially because NIR is very sensitive to temperature. "The auto calibration every single millisecond allows us to operate efficiently through a wide range of temperatures during a day," he says.

Go here to see the original:
Powerful optical sorting technologies, artificial intelligence and robotics reduce contamination for recycling operations - Recycling Product News