EFTA02576982

From: Joscha Bach Sent: Friday, Novem er , 1 :1 To: Kevin Slavin Cc: Joi Ito; Martin Nowak; Epstein Jeffrey; Ari Gesher, takashi ikegami Subject: Re: The benefits of deception Kevin, thank you for your excellent input and inspired questions on this =opic! > The broad idea is that we'll need to build slack into systems of =biquitous computing. That there's a paradox in that in order to have a =ohesive identity, certain aspects of our lives require omission (or =.g., deception). There are two sides to this: first of all, many of our interaction =artners may have difficulties to accept (or even comprehend) the =nteractions and social norms that we maintain in other contexts (think: =iving in a small town in the midwest vs visiting Burning Man; religious =ommunity vs. less usual sexual interests; family life vs. work life; =rganizational role vs. friendships). As a result, we compartmentalize =ur social expression. In a culturally inhomogeneous society, this kind =f deception (in the sense of hiding parts of my personality depending =n the context) is a prerequisite of meaningful freedom. But on a deeper level: while we may maintain the illusion of a coherent =elf, we do not possess a single social persona. Our social identity =merges not only over our personality, but also over the individual =elationships in which it manifests. We become who we are through the =eflection of others. > The ethnographer Tricia Wang coined "The Elastic Self' after spending = lot of time with Chinese and American youth using various forms of =ocial software (...) The notion of the "Elastic Self' might capture both aspects. On one =and, we may explore different possible ways of self- actualization by =eeping social contexts isolated from each other, on the other, we =sually maintain more than one persona. > How would you build a system that provides a patient with the ability =o continue acting deceptively -- in order to maintain a cohesive =dentity -- while providing the doctor with the ability to find -- and =ct on -- perfect information? The simplest setup: the patient is a BDI (belief, desire, intention) =gent, and has accurate knowledge about himself. He also has beliefs =hat part about his beliefs, desires and intentions are acceptable to =he outside world (which here is a homogenous hospital context and does =ot need further differentiation), and which ones need to be adapted to =it the actual intentions. For perfect information, the doctor gets =irect access to both layers. The BDI model of agency (i.e. the idea of describing an agent using a =et of beliefs, desires and intentions) is a somewhat crude =implification. We may well hold different, contradictory sets of =eliefs and goals at different levels, and the one that becomes relevant =epends on the context, and the state the agent is in. Also, many =eliefs and intentions are not directly represented, but generated when =eeded, and their content and form may be highly dependent on the =nteraction. In reality, there is often no perfect information for the =octor to act upon, e.g., because some of the information will be =enerated during and depending on the actions of the doctor. We might want to skip that kind of accuracy if we just want to get an =dea of the benefits of deception for the organization. If we treat the =atients (as well as doctors, nurses, ...) as straightforward BDI =gents, the "perfect information" would be the set of beliefs and =esires that the patient functionally acts upon. If we want to include =elf- deception, we might simply add another layer on top of that: the =et of beliefs that the agent has about his own beliefs, EFTA_R1_01749385 EFTA02576982 desires and =ntentions at any given time. For functionally accurate deception of =thers, the patient needs to represent beliefs about the actual beliefs =f the doctor, and actionable desired beliefs of the doctor about the =eliefs, desires and intentions of the patient himself. (A perfectly =lairvoyant doctor would have access to all belief layers of the =atient, and the relationships between them.) Only if we want to assess =he benefits of deception to the agent himself do we need to implement a =ore detailed cognitive model. Cheers, Joscha <?xml version=.0" encoding=TF-8"?> <!DOCTYPE plist PUBLIC "-//APPle/IDTD PUST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd"> <plist version=.0"> <dict> <key>conversation-idgkey> <integer>272682</integer> <key>date-last-viewed</key> <integer>0</integer> <key>date-received</key> <integer>1383876705</integer> <key>flags</key> <integer>8590195717</integer> <key>gmail-label-ids</key> <array> <integer>6</integer> <integer>2</integer> </array> <key>remote-id</key> <string>359754</string> </dict> </plist> 2 EFTA_R1_01749386 EFTA02576983