plan. This is a many-month process, and on day 1
you don’t see how you could possibly measure improvements, but eventually something clicks, and
you can really see the path forward.
Editorial Team: Many companies are now interested in commercializing AI and creating their own
narratives around AI. What are your thoughts about
what you see?
Gil: I have been doing AI research for more than 30
years, and I have always seen companies presenting
AI in a way that promotes their commercial interests.
This is natural and is nothing new. In the 1980s, all
the banks wanted investors and customers to know
that they were using expert systems and they had
AI capabilities in-house. As an AI researcher, I like
to take the opportunity to reflect on whether I agree
with a particular company’s approach to AI. For example, do I like the ethics they use when they release
AI capabilities? As AI researchers, we have tremendous responsibility to work for (or with) companies
that we are in agreement with and whose narratives
for AI we find compelling.
Editorial Team: As the incoming president of AAAI,
what influence do you hope to have?
Gil: I view a position like this in an organization
like AAAI as a position of service. One is there to
serve the community, to really tap into what the
community wants and needs. I may have some ideas
but I am very much interested in understanding
how the community wants to move forward. One
aspect that the AI community cares about deeply is
the connections between industry and academia. We
have the Innovative Applications in AI conference
during AAAI, and that is very exciting but we need
additional efforts in this area. For example, AAAI is
one of the founding board members of the Partnership for AI, which brings together companies with
academic and scientific organizations to understand
important issues concerning the practical uses of AI.
Another topic that our community really cares about
is education. We have the Educational Advances in
AI conference during the AAAI conference, which
brings together professors and teachers in a wide
range of institutions for higher education. AAAI just
started a new initiative on K- 12 education, which is
incredibly important. Not only do we need to build
a pipeline of researchers in AI, but we have to recognize that a K- 12 student today will be a consumer of
AI tomorrow, and a voter (possibly even a member of
Congress!). We want all students to understand the
big questions in AI and provide them with the ability
to think critically about how AI systems work and
how they may affect their lives. Diversity and inclusiveness in AI is incredibly important and a very effective way to address this by making AI accessible to
all students in early education.
Editorial Team: What types of AI research should
we do less of? Or more of?
Gil: I think AI researchers are extremely creative
and diversified in their topics of research, and this is
very healthy for our field. So, let me talk about an
area where I think we should do more: Our AI systems
have no notion of their own limitations. They don’t
know what their capabilities are or what exists beyond
their capability. In the human world, if you walk into
a pharmacy and ask someone for a flu shot, if that
person is a cashier or a receptionist, they will tell you
why they can’t give you one and they will direct you
on who to go to instead. Our AI systems are rarely
designed to do this. When asked a question, our AI
systems will give you an answer, but they don’t know
the context, and they often don’t know the risks in-
volved. I wrote a paper recently about thoughtful AI2
that talks about how an AI system could have more
awareness of how it fits into the world so that it could
say “I don’t know a lot about that so I cannot be help-
ful, but here’s how you can find help.” I wish there
was a lot more research on this topic.
Editorial Team: What excites you most about your
Gil: One of the things we are able to do much better today than 5 years ago is to capture information
about a scientist’s data analysis process and represent it as a semantic workflow. This year we have
been studying how data analyses are performed in hydrology and agriculture modeling to manage water
resources and food production. As we continue to
study different scientific disciplines, we see more
commonalities. Semantic workflows will allow us to
teach machines to help us make scientific discoveries. And, I don’t mean giving a machine data and
asking it to come up with correlations. I mean teaching a machine to actually design an entire approach
to test hypotheses by finding appropriate data, performing data analysis, and making decisions about
what results may be significant. This is how a scientist would approach a problem. Today, AI systems are
not full-fledged scientists, but they are beginning
to be able to design approaches to data analysis. I believe we are on the brink of seeing AI systems deeply
transform how we approach scientific discoveries.
Editorial Team: Thank you for sharing your perspectives with us today, Dr. Gil.
Gil: You’re welcome, I really enjoyed your thoughtful questions.
2. The paper is available at doi.org/10.3233/DS-170011.
Yolanda Gil is director of knowledge technologies and
research professor at the Information Sciences Institute of
the University of Southern California, USA, and president
Biplav Srivastava is a distinguished data scientist and master inventor at IBM’s Chief Analytics Office.
Ching-Hua Chen is a research staff member at the IBM T.J.
Watson Research Center in Yorktown Heights, New York.
Oshani Seneviratne is the director of health data research
at the Institute for Data Exploration and Applications at the
Rensselaer Polytechnic Institute.