Interim Report from the Panel Chairs: AAAI Presidential Panel on Long-Term AI Futures

Abstract

The AAAI 2008-09 Presidential Panel on Long-Term AI Futures was organized by the president of the Association for the Advancement of Artificial Intelligence (AAAI) to bring together a group of thoughtful computer scientists to explore and reflect about societal aspects of advances in machine intelligence (computational procedures for automated sensing, learning, reasoning, and decision making). The panelists are leading AI researchers, well known for their significant contributions to AI theory and practice. Although the final report of the panel has not yet been issued, we provide background and high-level summarization of several findings in this interim report.

Appendices

Itamar Arel on Horwitz’s "AAAI Presidential Panel on Long-Term AI Futures"<!--RH>Itamar Arel on Horwitz’s "AAAI Presidential Panel on Long-Term AI Futures"<!--/RH>

The term Artificial Intelligence (AI), which has been coined over 50 years ago, was followed by era of optimism and, in retrospect, great naiveness regarding the field’s prospect. Despite the many impressive achievement in building complex computerized systems, such as robots that can drive a car and programs that play chess at grandmaster level, the holy grail of building a machine with human-level intelligence remains an unfulfilled dream. In many ways, the report of the 2009 AAAI Presidential Panel on Long-Term AI is a solid reflection of both the disappointment and frustration from the lack of a much-anticipated AI breakthrough. In the absence of the latter, one is inevitably confined to the current, somewhat narrow, interpretation of AI systems. However, it is important to understand that while a conceptual AI breakthrough has yet to materialize, it may very well occur in the not so distant future. When such a breakthrough does take place it will undoubtedly have profound impact on our lives in ways which are difficult to imagine at this time.

The scientific fronts that many feel offer the most promise when it comes to revolutionizing the field of AI are computational neuroscience and neuropsychology. Great advances have been made in these areas over the past couple of decades with many anticipating continuing emergence of novel insight, which will contribute to our understanding of the mammalian brain. Recent progress in developing experimentally-grounded cognitive models, such as accurate descriptions of cortical circuitry, not only deepens our knowledge of how the brain works but also inspires researchers to propose new ideas pertaining to intelligent systems design. Hence, biological inspiration, rather than explicitly reverse engineering biological circuits, seems a promising approach for moving forward in mimicking cognitive functionality using machines. An example of a machine learning niche that emerged as a result of recent neuroscientific findings is deep machine learning, which employs hierarchical architectures for multi-modal perception in a manner resembling that of the neocortex.

As one would assert from reading this book, there are different interpretations of the singularity, many of which pertain to machines reaching a critical intelligence level beyond which predicting the future of humanity becomes impractical. Regardless of whether human-level intelligent machines will be the result of a singular AI breakthrough or not, now is the time to consider its various implications, particularly in guaranteeing that such a technology is not exploited maliciously. Rather than discarding the possibility of human-level machine intelligence, we may begin by asking key questions that assume its plausibility. Questions such as: how much time will humanity have before it will be impossible to control the evolutionary trajectory of this new life form on earth? Would it be possible to prevent the technology from reaching adversarial entities, and if so how can we effectively enforce policies that achieve this important goal? Human-level intelligence does not necessarily imply human-like intelligence; to that end, how would these new creatures behave, particularly as they interact with humans? Would it be possible at all to prevent a catastrophe for humanity as a result, for example, of a grand existential conflict of interests? The questions above cannot be comprehensively answered at this time, suggesting that open-mindedness to multiple futuristic scenarios is the logical position to take. Simply dismissing such scenarios may prove one day to be a historical mistake of epic proportions.

Vernor Vinge on Horvitz’s “AAAI Presidential Panel on Long-Term AI Futures”<!--RH>Vernor Vinge on Horvitz’s “AAAI Presidential Panel on Long-Term AI Futures”<!--/RH>

Points in the Panel report I especially agree with:

• Explicit AI research is just part of a very large human undertaking. (The demand for progress in computation, communication, and automation is coming from almost all directions).

• Improving software and hardware for collaborations is important. (In my opinion, it is the most important measure—along with the ongoing conversation about these issues—for assuring a good outcome).

The Singularity is very different from concerns such as climate change. For one thing, talk about the Singularity is only in part about avoiding disaster. No one need cry in the wilderness for help on this: as time passes, more and more people will be involved in the endeavor, mostly for happy reasons.