Coherent Blended Volition
When discussing machine ethics and the extrapolation of human values for use in AGIs, CAV appears to the author as being easily misinterpreted, with a confusion emerging between aggregation and averaging. This misinterpretation is then clarified through CBV, which suggests instead a “conceptual blend” – a term borrowed from Fauconnier and Tunner’s works exploring creativity (2002) - between different values and perspectives. This would allow for an harmonious and elegant incorporation of divergent views and goal sets. The major detail here is that each and every single one of the persons whose views are being blended would have to agree that these blend includes enough of their own contributions.
According to original definition of this conceptual blending, new concepts are formed by combining the most important details of already existing ideas. This process, however, is done carefully, in order to obtain a useful and high-value new concept. As such, the result is similar in a way to the concepts that gave birth to it, but at the same time possessing its own novel integrity. When considering the problem of aggregating volitions for use in AGI, this blending of different people’s goals and views is proposed as a novel, fine-tuned solution with which no human would disagree.
CBV vs CEV
If a thorough collective interactive process to arrive at a form of Coherent Blended Volition is to be designed, it can probably encompass many of the same kind of extrapolations that are part of Yudkowsky's Coherent Extrapolated Volition. The main differente between the two methods rely on who's responsible for guiding them. In CEV, this process is to be carried by a highly developed software framework. On the other hand, approach inherent to Coherent Blended Volition, this is to be done by a collective human activity. This perspective relies on the assumption that the creation and definition of collective human values is probably better carried out through human workand colaboration than through an artificial, machine based process. At the same time, the author links this blending process to already existent tools like the Internet, which already support deep sharing and collective engagement with and for humanity.
Another major problem with the original CEV definition is that it imagines that humanity’s values, when extrapolated, will become coherent and converge. There is, however, no strong basis to believe in this propositions. Humanity may share many biologically imprinted ancestral goals but, as thoroughly discussed we can be very divergent in our opinions, goals and ethics. As such, a blend of these features along with an approval from each opinion holder - through CBV - is seen as a viable bypass to this situation with a more useful and secure output.
Further Reading & References
- Fauconnier, Gilles and Mark Turner. (2002). The way we think: Conceptual blending and the mind’s hidden complexities. New York: Basic Books.
- Nine Ways to Bias Open-Source AGI Toward Friendliness by Goertzel and Pitt