How AI Can Outsmart Criminals and Improve Society

If you're a fan of heist movies similar The Italian Job, y'all're familiar with the scene where the bad guys get together around a map, and the kingpin says something like: "The guards do a security patrol at 05:xxx hours. Synchronize your watches. We'll enter here at 05:47 hours. Fingers will disable the alarm earlier Maxi and her squad enter the building via the roof."

Cheers to artificial intelligence, it'southward a plot signal Hollywood can't utilise anymore, every bit many security companies at present use AI-configured game theory to randomize patrols. The best known is ARMOR, developed by Dr. Milind Tambe, a professor at USC'south Viterbi Schoolhouse of Engineering, and in utilize at locations like LAX.

PCMag met Professor Tambe in his AI lab at USC's Salvatori Computer Science Center to learn more about this, as well as his focus on using AI to address social club's ills as co-founder of the USC Center for Artificial Intelligence in Society (CAIS). Hither are edited and condensed excerpts from our chat.

Dr. Tambe, tell u.s. how you got interested in AI as a field of inquiry.
[MT] When I was growing up in Bharat, AI belonged to the future, a globe portrayed in Star Wars and Star Trek, and in the science fiction I read, mostly Asimov. In fact, the 60s original series of Star Trek started playing in India in the 1980s, and I constitute it fascinating. That was the genesis of how I learned about the field. And so I ended upward coming to the United states of america and got my PhD in AI at Carnegie Mellon.

You're at present using AI to solve a lot of society'due south problems, including security issues, with the ARMOR software, which has won many awards.
We have been working with AI for security since 2007. Information technology started here at LAX following 9/11. But terrorist attacks happen everywhere. For example, the train attacks in Mumbai in 2006 brought the issue very close to domicile for me, as my mother travels on that line. Dr. Milind Tambe, a professor at USC's Viterbi School of Engineering And so I wanted to contribute to counter-terrorism, to commencement to improve security. Our ARMOR software has since been modified to exist used by the US Federal Air Marshals (as IRIS) and, as PROTECT for the US Coast Guard for maritime security.

Who funded the original research?
ARMOR was adult in partnership with the University of Southern California through grants from the Department of Homeland Security, Department of Defense, and the United states Army Research Office. At that place are so many federal agencies, with relatively limited resource, and so many areas of the U.s.a. to protect.

Many of these agencies came to prominence during the Common cold War.
Indeed. Only nosotros're dealing with very different adversaries now. What is needed today are new, avant-garde, bogus intelligence operations, such every bit ours, to randomize and provide highly evolved risk assessment tools.

In 2022, your book—Security and Game Theory: Algorithms, Deployed Systems, Lessons Learned—was published. Then, in 2022, you co-founded a commercial spin-off of the ARMOR software, into a visitor now chosen Avata Intelligence.
Avata, which means "open" in Finnish, develops products that bridge the gap between data and decisions. The core applied science was adult here at USC, and Avata has taken that work into diverse areas such as public condom and infrastructure, defense force, healthcare, global awareness, and social impact. I retain my Main of Research part at that place, but spend most of my time here with my PhD students, iterating new concepts in AI to further this field.

What'due south the tech platform behind the original ARMOR software?
We built the algorithms ourselves, which randomize the security patrols and provide chance assessment scores, none of information technology is plug-and-play, we can have it up and running inside xxx days at a new organization.

Right, and the lawmaking layers are probably non available on GitHub either.
[Laughs] No. But information technology is published research work. I told our program managers at the TSA and Department of Homeland Security, 'If you want this work validated, what better way to get information technology peer-reviewed at international conferences by the top people in AI?' The security comes from the obfuscation, the randomizing. We don't hide anything in hole-and-corner, because that invites break-ins. The force is that, due to the algorithms, you don't know what's going to happen tomorrow, because the AI hasn't decided information technology yet.

Are you using machine learning?
We're starting to do that now because there's a growing body of historical information. Nosotros presume that the antagonist is very strategic, always thinking of [how to] inflict great damage on the states. When we go to domains where the loss isn't as consequential as human life, and the adversaries are non equally strategic, say wildlife poaching, and so nosotros can be more than ambitious, and nosotros have much more information on poaching. We can develop models based on exploiting known weaknesses at that place.

Talking of poaching, you're moving your AI to focus not just on security, but also on social issues.
In Oct 2022, I attended a joint event hosted past the USC Suzanne Dworak-Peck School of Social Piece of work and the Viterbi School of Engineering. The thought was to introduce social work faculty to new concepts in technology, and engineering faculty—who were interested in social bug—to emerging needs in that field. It was there that I got talking to Eric Rice, associate professor [at the] School of Social Work. Eric's team had merely finished collecting HIV risk and prevention activities amongst several panels of social networks of homeless youth in Los Angeles. We so collaborated to create an algorithm to place the most efficient peers within a social network of homeless youth.

This and then became a formal collaboration?
Aye, we and then set up up the USC Center for Artificial Intelligence in Guild (CAIS). Our mission at that place is to describe inspiration from the Grand Challenges of Social Piece of work, the Grand Challenges of Technology, and the Un Sustainable Development Goals, and the United Nations Millennium Development Goals, which provide of import new directions for AI and social scientific discipline research. Based on these goals, our initial projects focus on ending homelessness, fighting substance abuse, preventing suicide, improving access to health care, social responses to global climatic change, reducing gang violence, and protecting wildlife.

And you've taken the algorithms and machine learning behind the HIV work into other preventable diseases.
We are now heavily involved in using AI for public health, to aid allocate the right resources, well-nigh recently in at-risk groups for tuberculosis in India.

What's next for you?
This is basic research in early stages, but we're starting to build a qualitative understanding of how gangs office, together with a quantitative response to their culture, peculiarly with recruitment and radicalization, using game theory to model predictive algorithms and provide a useful response.

Not just keeping LA citizens prophylactic, but besides getting inside the gang civilisation to protect vulnerable youth from recruiting?
Exactly. It'due south heady piece of work. Much of AI doesn't leave the lab, in that information technology doesn't get validated in the real earth. And so there are hundreds of academic papers which accept never been proven in society. That'due south why nosotros believe the piece of work we're doing is so important. In all of our enquiry and development, using AI for "good domains," whether it'southward security, social piece of work, public health or poaching, we accept empirical information to bear witness that the AI supports professionals in the field and helps them do their jobs more effectively.

If any PCMag readers are in Sweden this summer, Professor Tambe will exist presenting his work at the 27th International Joint Conference on Artificial Intelligence in Stockholm in July.