Check out all the on-demand sessions from the Intelligent Security Summit here.
When considering transformational ways to use computer vision on the edge in devices like robots, drones, cameras, and other devices, Booz Allen Hamilton VP Josh Sullivan advises caution, urging people to take security seriously on what’s become a whole new attack vector.
“For me, deploying an AI model in your IT environment is an entirely new attack vector. I’ve seen a model working correctly that can identify tanks and other military equipment be fooled into seeing a school bus because someone sent poisoned data into the model,” he said.
Failure to keep models secure can lead to adversarial machine learning attacks to make malicious code appear as benign or a range of other bad outcomes. Sullivan was joined in conversation at VentureBeat’s Transform 2020 conference by Nvidia VP of federal initiatives Anthony Robbins, Intel IoT VP Stacey Shulman, and Joint AI Center acting director Nand Mulchandani. Advances in encrypted models or use of federated learning may be required so that models are not as valuable if left in a war zone or discovered by a combatant. Booz Allen Hamilton signed a $800 million contract with the Pentagon’s Joint AI Center last month.
“I think for the conversations we’re having like about facial recognition and law enforcement and privacy are crucial ones for us to have as a nation, but I also don’t want to ignore the equally profound conversation around adversarial machine learning and the ability to hack AI models with poisoned data in attempts to influence their predictions,” Sullivan said. “We’re going to keep relying more and more on these predictions and you’re going to create an outsized impact that becomes possible by hijacking just a few bits of training data to change the entire outcome. I don’t want to bolt the front door and leave the entire back door popped open.”
Intelligent Security Summit On-Demand
Learn the critical role of AI & ML in cybersecurity and industry specific case studies. Watch on-demand sessions today.
The Joint AI Center was created by the Department of Defense in 2018 to lead Pentagon AI efforts. The initiative was led by Air Force Lt. Gen. Jack Shanahan until he retired and was replaced by Mulchandani, a civilian who said he sold four startups and spent 20 years in Silicon Valley before coming to the Pentagon one year ago. Mulchandani told VentureBeat in an interview last month how JAIC is evolving to adopt a culture and workflow more in line with how Silicon Valley treats software sprints and investment.
In a press conference at the Pentagon last week, Mulchandani talked about a “global war for AI talent” and said JAIC’s first lethal application of AI, what he describes as JAIC’s “flagship product” for joint warfighting operations, will be a “tactical edge” device.
Returning to the subject of tactical edge AI today, Mulchandani said advances in batteries, chip sets, algorithms, and other areas will be necessary to make progress in deploying AI at the edge. “We are really relying on the private industry and breakthroughs and the technology platforms that private industry will enable that we can consume but also customize and deploy for our own use case, so it’s been a fascinating move,” he said. “That entire end to end process is an incredibly complicated one and is going to be. I think if somebody is thinking about starting a company or are partners on this webcast — if they’re investing, which they are — this is really where I think the game is going to get played out from an exciting standpoint in the next gen of technology here.”
Each participant in the panel conversation has active contracts with the Department of Defense. Interest in government contracts may increase for some businesses as COVID-19 has introduced more volatility to the economy.
In addition to Mulchandani’s appointment to acting director last month, other recent high-profile moves between the Pentagon and Silicon Valley include current White House CTO Michael Kratsios. He was chief of staff to Palantir founder and investor Peter Thiel at one time, but on Monday news emerged that he has been appointed Acting Under Secretary of Defense for Research and Engineering, according to his LinkedIn page. And going the other way, Josh Marcuse was executive director of the Defense Innovation Board in the office of the Secretary of Defense, but in March became head of strategy and innovation in Google’s public sector division vying for government contracts.
Connections between Silicon Valley and the Pentagon appear to be increasing. Interaction has grown in part through the activity of groups like the Joint AI Center and the Defense Innovation Unit, a group created in 2015 with offices in Palo Alto.
For example, Google employees spoke out against the company’s involvement in Project Maven, an initial JAIC project. A few months after the Maven contract came to light, Google adopted AI principles, but in a conversation with former JAIC director Air Force Lt. Gen. Jack Shanahan last fall, Google VP of global affairs Kent Walker reaffirmed the company’s commitment to competing for government contracts.
The Defense Innovation Board is led by former Google CEO Eric Schmidt, while the National Security Council on AI, a group advising Congress on AI policy and investments, includes leaders from companies like AWS, Google Cloud, and Microsoft. Each of the groups advocates for more partnership between industry, academia, and government. Analysis by Tech Inquiry earlier this month found extensive government contracts between tech companies and the military.
Nvidia’s Anthony Robbins called public-private partnerships a “giant team sport” and identified opportunities for startups in cybersecurity, robotics, and other services. “Running and retraining models at the edge is going to define the next decade,” he said.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.