AI alarm sounded; systems engineering inspiration; autonomous vehicle tradeoffs.
Prepare to prevent malicious AI use
According to the University of Cambridge, 26 experts on the security implications of emerging technologies have jointly authored a ground-breaking report thereby sounding the alarm about the potential malicious use of artificial intelligence (AI) by rogue states, criminals, and terrorists.
The report forecasts rapid growth in cyber-crime and the misuse of drones during the next decade as well as an unprecedented rise in the use of ‘bots’ to manipulate everything from elections to the news agenda and social media. This adds up to a clarion call for governments and corporations worldwide to address the clear and present danger inherent in the myriad applications of AI, they said.
The report – The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation – also insists on interventions to mitigate the threats posed by the malicious use of AI. Specifically, policy-makers and technical researchers need to work together now to understand and prepare for the malicious use of AI.
They acknowledge that AI has many positive applications, but it is a dual-use technology and AI researchers and engineers should be mindful of and proactive about the potential for its misuse.
Best practices can and should be learned from disciplines with a longer history of handling dual use risks, such as computer security.
The range of stakeholders engaging with preventing and mitigating the risks of malicious use of AI should be actively expanded.
The co-authors come from a wide range of organizations and disciplines, including Oxford University’s Future of Humanity Institute; Cambridge University’s Centre for the Study of Existential Risk; OpenAI, a leading non-profit AI research company; the Electronic Frontier Foundation, an international non-profit digital rights group; the Center for a New American Security, a US-based bipartisan national security think-tank; and other organizations.
The 100-page report identifies three security domains (digital, physical, and political security) as particularly relevant to the malicious use of AI. It suggests that AI will disrupt the trade-off between scale and efficiency and allow large-scale, finely targeted and highly efficient attacks.
Novel cyber-attacks are expected. These include automated hacking, speech synthesis used to impersonate targets, finely-targeted spam emails using information scraped from social media, or exploiting the vulnerabilities of AI systems themselves (e.g. through adversarial examples and data poisoning).
Additionally, the proliferation of drones and cyber-physical systems will allow attackers to deploy or repurpose such systems for harmful ends, such as crashing fleets of autonomous vehicles, turning commercial drones into face-targeting missiles or holding critical infrastructure to ransom. The rise of autonomous weapons systems on the battlefield risk the loss of meaningful human control and present tempting targets for attack.
The list goes on to include the political realm. The solution? Start rethinking cyber-security, explore different models of openness in information sharing, promote a culture of responsibility, and seek both institutional and technological solutions to tip the balance in favor of those defending against attacks.
For more, the Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation is available to download here.
Maximizing the environmental benefits of autonomous vehicles
The added weight, electricity demand and aerodynamic drag of the sensors and computers used in autonomous vehicles are significant contributors to their lifetime energy use and greenhouse gas emissions, according to University of Michigan researchers.
The good news is, when savings from the driving efficiencies associated with self-driving vehicles are factored into the equation, the net result is a reduction in lifetime energy use and associated greenhouse gas emissions of up to 9 percent compared to the conventional vehicles examined in the University of Michigan-led study.
“This study explored the tradeoffs between the increased environmental impacts from adding autonomous vehicle equipment with the expected gains in driving efficiency,” said study co-author Gregory Keoleian, director of the Center for Sustainable Systems at U-M’s School for Environment and Sustainability. “Our findings highlight the need to focus on energy efficiency when designing autonomous vehicles so that the full environmental benefits of this emerging, transformative technology can be realized. We hope this work contributes to a more sustainable mobility ecosystem.”
The study is a detailed assessment of the lifetime contributions of the sensing and computing subsystems in autonomous vehicles to energy use and associated greenhouse gas emissions. These vehicles, formally known as connected and automated vehicles or CAVs, often include multiple cameras, sonar, radar, LiDAR, a GPS navigation system, a computer and support structures.
The researchers looked at two types of CAVs: those powered by internal combustion engines and battery-powered electric vehicles. The two vehicle types were paired with sensing and computer subsystems of three sizes (small, medium and large) to create six scenarios.
Life-cycle assessment methodology was then used to estimate lifetime energy use and greenhouse gas emissions for each scenario, from cradle to grave.
One key finding is that autonomous vehicles with electric powertrains have lifetime greenhouse gas emissions that are 40 percent lower than vehicles powered by internal-combustion engines. The lower emissions result from the inefficiencies involved in producing electricity from fuel combustion, as well as a sharper fuel-consumption increase when extra mass is added to a vehicle powered by an internal-combustion engine.
In short, a battery-electric vehicle is a better platform for CAV components compared to the internal-combustion engine vehicle in terms of minimizing environmental impacts.
Also, the researchers found that the sensing and computing subsystems in connected and automated vehicles could increase a vehicle’s energy use and greenhouse gas emissions by 3 to 20 percent due to increases in power consumption, weight and aerodynamic drag. Still, the operational benefits of autonomous vehicles, which include smoother, more efficient traffic flow, are expected to outweigh those increases in most cases.
Bees: systems engineering inspiration
According to Georgia Tech researchers, while nature offers excellent design inspirations in some information technology systems, in other systems, it can bomb. Bees? Great. Ants? Hit or miss. Slime mold amoebas? Fail.
Georgia Tech systems researcher Craig Tovey has seen plenty of nature-inspired technological feats, but also foibles. Known for his work on The Honey Bee Algorithm, which tamed web traffic instabilities on servers by mimicking the behavior of bee colonies, he shared insights on this topic in a recent talk at the annual meeting of the American Association for the Advancement of Science in Austin, Texas.
The Honey Bee Algorithm, has saved significant web hosting costs. “We lucked out with the bees and web hosting,” said Tovey, who along with practical takeaways on naturally inspired technology, enjoys passing on his own awe and affection for nature’s solutions.
Interestingly, Tovey said when you study swarming bees, you discover truths that are lasting. “The algorithms that guide them evolved over millions of years, and will hopefully still be there for millions of years to come. Compare that with when you design a new microcircuit. Three years later it’s gone, forever lost; replaced by new designs.”
Whether mimicking nature is prudent in a particular engineering job depends a lot on the problem to be solved. Often, it’s just better to use something off the shelf or adapt it, he said. “When the real-life problem is static and well-defined with predictable data, then the nature-inspired methods are usually much weaker, much worse than classical optimization methods.”
The “Traveling Salesman Problem” is a typical example. A researcher tries to compute the best pathways a proverbial salesperson should travel, and in which order, to visit hundreds, thousands, or tens of thousands of proverbial cities on a map.
The goal is to travel the shortest possible total distance.
“Nature-inspired approaches will find good solutions for 100 or so cities, but not optimal ones,” said Tovey, who is also a professor and Stewart Faculty Fellow in Georgia Tech’s Stewart School of Industrial and Systems Engineering. “By contrast, the top researchers can solve 20,000 or 50,000 locations optimally with a classical algorithm, and do it really quickly.”
“People have imitated ants to find the optimal pathways through a static system, and when you compare that method with classical optimization methods, then the classical methods are about 10 billion times better.” But life is fickle, which can make it a great teacher in science and engineering. “Every living creature is very good at solving a number of different problems, otherwise it would have gone extinct,” Tovey said.
Toss unpredictability into an engineering problem, and natural algorithms that direct the movements of ants or bees can be better equipped to cope than classical solutions.
“In the Traveling Salesman Problem, the cities don’t move around. But when you’re chasing a moving target, and your data isn’t perfectly complete, then you can have great success by imitating insect swarms. You can get real-time control on data that’s quite literally on the fly,” Tovey said.
That counts for a lot in a pinch. When a hurricane looms, people check their weather apps much more frequently as the tempest encroaches. When markets tank, people sell off stocks, and data surges in and out of financial servers.
“If the patterns of user demand on the web never changed, and the requests to a server always stayed the same, all would be well without imitating honeybees,” Tovey said. “But that notion is ridiculous, as we all know. Bees have evolved to deal with flower patches that have changing characteristics. A patch that is great to visit at 10 o’clock in the morning may have its flowers closed-up at one o’clock in the afternoon, or it may be raining.”
Algorithms steering bee behavior make the insect swarms adjust to supply and demand fluxes similar to those that confront a web server. The honeybees handed Tovey and his fellow researchers valuable insights for their web hosting algorithm.
Leave a Reply