xltd.com weblog

I came across a 1994 paper by Minsky where he discusses the notion of Negative Expertise. This concept runs contrary to "rule-based expert systems" that were the rage in the 1980s. Instead of an "if this, then do that" pattern, humans are more likely driven by "don't do that" patterns. An intelligent entity, human or otherwise, is driven by goals. A primary goal is to survive. An examination of Asimov's Three Laws of Robotics, follows a mixture of these two types of rules.

1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey orders given it by human beings except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

Asimov even later added a Tolkienesk "law to rule all other laws". This 0th law states "A robot may not harm humanity, or, by inaction, allow humanity to come to harm". With these laws in hand, we are prepared to set up an Orwellian order where ""All intelligent life is valued equally, but some intelligent life is more equal than others". Robots must preserve humanity, but in the process, they may wipe themselves out? Not such a good deal for robots. As the mad dash is on to adopt robot ethics, other bloggers have made the connection to Asimov's laws. Questions are raised as to what constitutes rape of a robot or what constraints are required to prevent Terminator-like autonomous war machines from killing humans.

But back to negative expertise. The humans where I live are guided by rules from an early age such as

• don't run out into a busy street
• don't stick a fork into an electric socket
• don't drink from the bottles under the kitchen sink

Minsky relates pleasant emotions to a double negative on these types of rules. Watching something where this type of conduct it exhibited is regarded as humor. The Three Stooges built a career out of it. The Jackass franchise is a more recent example. Beauty can result from a suspension of critical analysis in order to behold something perceived. Decisiveness results for cessation of consideration of options. So many things we consider desirable result form not doing something.

In robotics, if the goal was to build an autonomous entity, then modeling the absence of behavior is indeed a daunting task. The laws of robotics pose such a challenge.

• Don't hurt humans
• Don't just stand there if a human is being hurt
• Hurt a human if humanity is at stake
• Destroy yourself if it can save a human life

Yikes! What is a robot to do? Its cognitive system cas easily be frozen to inaction and that is also a violation of a law! Damned if they do and damed if they don't. Ah, but Minsky comes to the rescue with the notion of dividing the mind into parts that can monitor one another. Given part B is monitoring part A where B's entire existence is observation of A with no other stimuli or input. One of B's rules would be a basic General Problem Solver, meaning that it would monitor A's progress toward a goal and force A to review its goal if it was not make progress or caught in an apparent repetitous loop. "Apparent" is the operative word here since factory robots must be repetitious. Cooperating mind parts reminded me of the Tine dog-like creatures from Vernon Vinge's 1993 Hugo winner A Fire Upon the Deep. These Tines exist as small packs of individuals, but each individual consciousness is generated by the "marriage" or enlistment of several Tines. A factory floor robot by itself exhibits no intelligence, but a "marriage" of many robots of many shapes a sizes (including virtual machines), may constitute a consciousness.

Combining robots in mix and match fashion to allow a consciousness to emerge raises all kinds of ethical conundrums. Machine consciousness (assuming there will be such a thing) may be a frequently changing notion, like a shifting sand dune. Back to Minsky's part B observing part A proposition, if A does something B considers good, then B will reinforce A's learning system. Conversely, if A appears to be making things worse, B may suppress it in favor of another agency. Furthermore, mind part A may not comprehend whys' it should perform some act or avoid doing some act. This is time to shift to another agency. This agency may result from the enlistment of other robots in the area or even in a metaverse. Perhaps control or behavior change will result from a rule by committee or intervention.

What constitutes consciousness in such a context? If one chooses to perform sexual acts with a future generation iRobot Roomba, would it be rape if the devise did not consent? Can you ask its controlling mind part to detach for your temporary enjoyment? Can you request other robot parts to intervene on your behalf to for an altered consciousness that will acede to you wishes?

Finally Minsky proposes that the B mind part monitoring the A mind part will exhibit something that I find similar to the Buddhist Middle Way path of moderation. Namely, if A is occupied with too much detail, then B forces A to take a higher level view. Conversely, if A is not being specific enough, then B makes A focus on more details.

Connections
Asimov has been attributed as describing Minsky as one of two people he has ever met who were flat out smarter than himself, the other being Carl Sagen, the author of a great book and film Contact. This book has been embraced by the intelligent design movement. For me, this connects back to a fabulous Asimov short story The Last Question that I saw as a multi media presentation back in the 1970s at the Triton College Planetarium in River Grove, Illinois. This experience I have now labeled as a Deep Engagement, based on the MIT Open Courseware seminar. This course is is based on this premise: "Innovation in expression, as realized in media, tangible objects, performance and more, generates new questions and new potentials for human engagement...once we experience deep engagement, it is a state we long for, remember, and want to repeat." The course readings explore underlying principles and innovative methods for development of higher-quality "deep engagement" products. The first reading was Minsky's book The Emotion Machine and its predecessor Society of Mind. That led me to the Minsky acedemic site and thus the refernced paper.

Labels: expertise, intelligence, robots