Satoshi Morinaga(PhD in Engineering/Executive Research Fellow/NEC Data Science Research Laboratories)

September 28, 2018

Creating a "discussing artificial intelligence"

The research I am working on now concerns next-generation artificial intelligence technology that enables artificial intelligences to discuss and negotiate between each other. In our world today, we see artificial intelligence in a variety of applications such as image recognition and weather forecasts. Their active fields will further expand in the future, such as into autonomous driving and drones.
However, the proliferation of such automatically controlled devices will inevitably cause situations where AIs meet and their interests clash. Think of self-driving cars for example. What if two of these cars were coming towards an intersection? If the other's movement cannot be completely predicted, the correct response is to drop your own speed to see what the other will do. However, both AIs are thinking the same way, so what happens is that both slowly decelerate as they get closer and will finally stop right at the intersection. Since this is extremely inefficient, it will be necessary for the AIs to communicate with one another to discuss and coordinate their actions.
What I am researching now is the technology to streamline this, by enabling the cars to discuss between themselves who is to stop and who is to go first in such situations. When such autonomous negotiation technology comes into its own, self-driving cars will become more common and new mobile services building on autonomous driving may also emerge. This may perhaps solve problems such as traffic congestions and contribute to making the society richer.

Accelerating the societal utilization of artificial intelligence

Applications where coordination between artificial intelligences is necessary are not limited to autonomous driving. The same applies to drones. For flying drones to avoid crashing into each other, they need to negotiate to coordinate their flight routes. When this becomes reality, it may be commonplace for the last mile of the delivery route to be taken on by drones in the near future.
Coordination of AIs has more applications in fields other than mobile devices. Examples can be the sale of things and working styles, where coordination of AIs can play a key role.
The automatic coordination between AIs is an essential element for a wide-ranging utilization of artificial intelligence technology in society. I am currently devoted to the research of this technology so that people can enjoy the blessings of benefits that artificial intelligence can bring to the world and to change our lives to a richer, more convenient one.

Creating a common language for artificial intelligence

Development of advanced algorithm is not all that is necessary to complete automatic coordination technology. Our other critical mission is to formulate the language to be used among the AIs. Automatic coordination technology has no meaning if AIs do not have common language for discussion.
This formulation of a common language to be used among various AIs cannot be done myself alone. We need to work on the research together with an extensive network of enterprises and research institutes. For that purpose, I am currently facilitating a cross-corporate project called Council on Competitiveness-Nippon (COCN) as a leader in this area of research. Also, for the research in this area, a Collaborative Laboratory was established by RIKEN and AIST, both of which are centers of artificial intelligence research representing Japan, where I take on the role of a sub-leader.
Autonomous negotiation and behavioral coordination by artificial intelligence is a totally new, unprecedented field of research. There are no previous research findings that we can reference. Having said that, I believe it most important for a researcher to have the attitude to pioneer and build up a new field of research based on their firm foundation in basic scientific concepts in mathematics and physics.

Relevant Laboratories