How do you feel about artificial intelligence? Excited? Apprehensive?
Perhaps you wish there was “grown-up supervision” on hand. Relax, there is: Founded in 1979, the Association for the Advancement of Artificial Intelligence (AAAI) has kept a fairly low profile outside of academic arenas, emerging when called upon to add reason behind closed doors in Washington.
But the organization now has 4,000 members worldwide and its goal is to: promote research in and guide the responsible use of AI; enhance public understanding of the field; raise standards in training AI innovators; and provide guidance to those funding major AI initiatives.
To find out more, we spoke with Dr. Yolanda Gil, who just took over as AAAI’s 24th president, via phone, at her office in USC’s Information Sciences Institute (ISI). Dr. Gil joined ISI in 1992 and is currently its Director of Knowledge Technologies and Associate Division Director; she’s also a Research Professor in Computer Science and in Spatial Sciences with a focus on intelligent interfaces for knowledge capture and discovery. Here are edited and condensed excerpts from our conversation.
Dr. Gil, could you tell us why you accepted the role of AAAI president? What was it about the organization and its mission that really drives you?
These are exciting times as AI increasingly permeates our lives. We see it in systems from chatbots to self-driving cars to scientific discovery and many other applications doing useful tasks. I believe AAAI is the leading forum to coordinate many areas of AI, and that we also have a strong responsibility to design AI systems that have—and encourage—ethical and responsible behaviors. My career has always included a focus on service to the AI and computer science communities so this was a natural step for me. I’m really excited about it.
Can you talk about three major objectives you have for AAAI moving forward?
The first thing to tell you is that I really see this as a listening experience, at least initially, so I can be responsive to what the community is looking for. Having said that, one big area is to enhance and strengthen AAAI links with industry. Our annual conference has a lot of participants from industry but I’d like to see more presence from industry research labs. Traditionally it’s been a very academic conference but today, many professors spend time in industry. We need to give that sector a lot more presence. That’s a major focus. I am also looking to include underserved communities in our membership to diversify it strongly; launch K-12 initiatives to grow the pipeline; and ensure we include professionals in other areas.
As young students learn about AI but go into other fields, it’ll spread understanding, rather than fear?
Exactly. Many K-12 students will go on to be doctors, entrepreneurs, engineers, or whatever they choose. But through exposure to AAAI, they’ll know more about the potential for AI in their chosen field.
Your predecessor launched a focus on AI and ethics.
Yes, and we’ll be continuing that through 2019…particularly within the second conference on “AI and Ethics in Society” during our annual conference. We need to look at employing ethics within AI at every level: how systems need to be designed with different mechanisms to respond ethically to events; understand when an AI system could do harm; and so on. I’m very excited about our initiative—it’s in partnership with ACM—and, as a community, we need to take a leadership role and do more research in this area: properly, clearly and creatively rather than letting circumstances shape AI.
Pivoting to your own research, for a moment, we first encountered your work back in 2015, at DARPA. How has that project progressed since then?
At the time [at DARPA] we were just starting that project on using intelligent systems for scientific discovery, assisting scientists with intelligent systems that analyze data, test hypotheses, and make new discoveries. At the beginning, we were focused on capturing scientific processes as semantic workflows. We have been working on several science domains. Now we have several related projects and are starting to see some results.
These results have not yet been published, but can you share some insights and specific areas of focus for your intelligent systems?
Yes, for example, in the scientific field of proteomics, the biochemical study of proteins within an organism, we have now captured a lot of workflows about this kind of analysis. And, interestingly, what we realized is that, when the studies are published, they just used one model [to identify the proteins] but they didn’t explore others, which means many proteins are missed.
And your AI system is smart enough to comprehend this omission and correct it?
Precisely. Our systems are now intelligent enough to be diligent and keep trying other methods, other algorithms, to detect hundreds of proteins that have been left out otherwise. The system itself is working on making new discoveries.
That’s amazing. Your AI systems will be making new scientific breakthroughs. A popular lab partner, one assumes.
[Laughs] We hope so. We are now working on a mechanism to measure “interestingness,” so that when the system finds something new it can check if it is significant breakthrough. This is a very challenging; it requires that the system has knowledge on what’s the state of the art in the field, the latest thinking on those proteins.
Otherwise it might just get excited, but the proteomics researchers might say: “yeah, but that’s merely a prosaic protein.”
That is right. And a good lab partner would not bother a scientist with minutia or unimportant findings. So we have to design an “interesting lab partner.”
Can your AI system ingest and analyze multiforms of data input?
Yes. We’re now automatically generating machine learning workflows. We give it data and a metric—i.e. the desired goal—and the system will start looking at what type of data it is. If it’s audio, it will look for a way to process it. If it’s visual data, it will find a means to understand and catalogue that. Then it will apply algorithms to maximize the metric (eg the accuracy of the solution). It’s very systematic.
Aside from inside the proteomics lab, aren’t your systems tackling problems out in the field with your new $13 million DARPA award?
Yes, that award is for a 4-year project called MINT for Model INTegration, part of DARPA’s World Modelers program. We are building workflows to integrate complex models of the world that cover hydrology, food production, climate, social and economics. We’re asking our intelligent systems to help us understand, in particular, prospective food shortages, poverty and food insecurity for at-risk communities around the world.
Right now, we’re collaborating with Kimetrica, the data company which provides large-scale investment appraisals and government statistics reports, to use our automated machine-learning workflows, and it turns out that they are better than the ones they did by hand. This is still in initial stages, but we’re getting good results. It’s exciting to have their domain expertise, combined with our AI research finding solutions to important world problems.
In the current issue of AI Magazine, Dr. Lynne Parker, co-leader of the National Artificial Intelligence Research and Development Strategic Plan writes: “it is incumbent upon us as technologists to focus on the positive, ethical development and use of AI, ensuring that everyone can benefit from the practical application of AI across society, regardless of which nation leads in the strategic development of the technology.” What was the significance of this document?
AI has always been very international—and distributed—with cultural richness embedded from all regions, many of which take different approaches to AI. For example, in Europe there’s an incredible tradition of logic-based approaches, in Asia it comes from advanced computational mathematics, whereas in Africa and Australia there’s a strong focus on applied research. The document you refer to is very significant because it means that the US recognizes the importance of having a national plan and making strategic investments.
An increased focus to compete with China and elsewhere?
To be clear, the US government has always made AI an important area for investment, but it hasn’t traditionally been a large investment, whereas the EU and China have been prioritizing government-led research investments in AI for a while. That report is a recognition that the US needs to continue to play a leadership role. It was a great document, very thorough and its recommendations are very right.
One thing it pointed out is that machine learning is important but human-computer collaboration is something that we cannot neglect. It also emphasized basic research in AI because, when you invest for 10, 20, or 30 years into the hard problems we have in AI—like speech recognition—you can see amazing results, but only when investments are sustained over a significant period of time.
It’s very frustrating for major speech-recognition AI researchers when the popular view is that there’s a magic Google Assistant lab that “figured it out over a late night pizza-fueled session in 2018.”
Right! [Laughs] Speech-recognition research took decades, and some directions didn’t work, and had to be abandoned. But because there was a significant investment people stuck with the problem and kept being creative and pushing forwards.
What about the current administration, considering that national AI plan was a document drawn up under President Obama?
Actually, the current administration just released a statement that they are going to make significant investments in AI and have started a special study to come up with a roadmap for AI research. I will be co-chairing that effort.
Well that’s a cheering thought. When, as co-chair of this new AI strategic plan for the US, are you delivering the first paper?
We’re going to be taking the National Artificial Intelligence Research and Development Strategic Plan and thinking very hard about basic research, human-computer collaboration, understanding the benefits of society as we invest—particularly around health and scientific discoveries and so on—to create an extensive roadmap by spring 2019. We also want to emphasize how innovation in AI can boost competitiveness and benefit both the industry and government sectors.
To learn more, the AAAI Fall Symposium is scheduled for Oct. 18-20 in Arlington, Virginia. The registration deadline is Sept. 21.
This article originally appeared on PCMag.com.
Originally appeared at
The Daily Caller