HOUSE_OVERSIGHT_014713.jpg

Esa Origins 7 February 24 ~ 26, 2017 PROJECT An Origins Project Scientific Workshop Challenges of Artificial Intelligence: Envisioning and Addressing Adverse Outcomes ARIZONA STATE UNIVERSITY 6) DEEP SOCIETAL INFLUENCES Al, Agency, and Disempowerment (Contributions from Gireeja Ranade, Andrew Maynard, David McAllester, Stuart Russell and others) We will be benefitting from Al system that are competent at doing important tasks. People and organizations seek Al systems that bring new abilities to the table. We desire autonomous cars that drive without collisions, we medical assistants that can diagnose patients accurately and we would like to have household assistants that can infer our intentions and execute them flawlessly —and even proactively. The military wants Al systems that can help with strategy and tactics, and systems that outmaneuver human led troops, and anticipate and respond to threats either on timescales that humans cannot achieve, or over landscapes humans cannot cover. Today, there is still skepticism about performance of Al systems in a variety of domains. However, we expect that Al systems will become more central decision support, pattern recognition, autonomous decision making, and other types of problem solving. As such, we will become increasingly reliant on Al systems. This raises concerns in several areas, including personal decision support, healthcare, transportation, governance and the handling and operation of weapon systems. We shall consider example of healthcare from Gireeja Ranade. The scenario and trajectory applies to other areas as we consider the increasing role and power of Al in our lives and in society: As healthcare providers are increasingly stretched in providing consultations with patients, diagnosing conditions, and developing treatment and/or intervention plans, tech companies identify a market opportunity for Al-based digital assistants that are designed to augment healthcare providers by collecting data from consultations, cross-referencing it with existing medical records, and providing feedback to aid appropriate diagnosis and decisions on how to proceed with treatment. Given the economic and health-base potential of the technology, it receives widespread support from the federal government (predominantly through grants and initiatives supporting it’s development), together with healthcare providers and healthcare insurance companies. Initial implementations are based on modular systems that share some commonalities with digital assistants like Siri and Echo/Alexa. Under the general name “Al-consult”, they consist of a physical unit in a consulting room that constantly monitors conversations, and sends encoded information to cloud- based servers. Here, information is coded, interpreted, and parsed out to further agents that cross- reference interpreted data with identified patient and healthcare provider records. Multiple and diverse databases are interrogated at this point. The result is data packets that include key information on the patient, including medical history, life style, and current status, and on the healthcare provider, including past history of diagnoses, recommendations, successes and failures. These are forwarded to a dedicated Al engine that analyzes the packets, and returns notes, advice and recommendations to the physical unit in the consultation room. 17 HOUSE_OVERSIGHT_014713