Testing Our Assumptions about AI: robots / cyborgs
March 26, 2003

Overview of Lecture: Different approaches to making artificial life / hybridizing human life: recognition of the importance of the "body' (see Mar. 24 lecture):

1. make human into cyborg

2. make machine human

3. make machine animal-like

******

1. Cyborg Experiments

What is a cyborg?: a hybrid of human and machine

Warwick's research: see http://www.wired.com:80/wired/archive/8.02/warwick.html
And www.rdg.ac.uk/Kevin Warwick/htm/faq.html

--> Warwick is experimenting on himself to become a cyborg: “I was born human, but it was an accident of fate.”
In 2002 implanted the second silicon chip in his arm "linking the nervous system in my left arm to a radio transmitter receiver to send signals from my nervous system to a computer and vice versa. ”

Why is he going this? “genetic changes offer short term, slight modifications. However the step to Cyborgs offers humans a natural, technological upgrade in the technological world we have instigated. Yes, I feel it will be the next evolutionary step. Indeed we will need to do it if we are to compete with intelligent machines.

2. Traditional AI researchers - focus on reasoning, learning and symbolic processing to make a machine with human intelligence + a body - the top-down approach:

robots develop a "model of the world"

(Minsky, MIT
http://web.media.mit.edu/~minsky/

Tilden, MIT
http://www.ai.mit.edu/people/brooks/projects.shtml

and now Roy, MIT (Ripley)
http://www.media.mit.edu/cogmac/robot.html

--> Moravec, Carnegie Mellon)
http://www.frc.ri.cmu.edu/~hpm/hpm.pubs.html

(and in fiction, Asimov)

3. Renegade AI researchers: get machines to act according to reflex:

robots do not have models of the world; they're not smart, they just "go" (just like reflex-driven insects)

(Mark Tilden)
http://www.beam-online.com/Robots/Galleria_other/tilden.html

*****

3. Rationalization for Third Approach: it's wrong to try to replicate the human brain combining complex seeing and thinking systems

-->move down the evolutionary ladder and model robots on brainless creatures which rely on instinct rather than reasoning.

-- >How do they work?

Subsumption Architecture:

robot's actions broken down into simple primitive instincts:

move forward, back off, etc.

-- >robot has no central brain to "blend" these simple behaviours

-->each behaviour acts as an individual "intelligence" that competes for control of the robot.

-->the winner is determined by what the robot's sensors detect at any particular moment, at which point all other behaviours are temporarily subsumed.

Tilden's robots:

-->Not made to be human-like but a "cockroach's cousin" with SURVIVAL LAWS:

Rule 1: Robots will protect their existence at any cost.

Rule 2: Robots must be able to find a power source and continually nourish themselves from it.

Rule 3: Robots must continually seek better power sources.

(compare these with Asimov's 3 Laws of Robotics (in kit)

-- > Tildon's robots called "junkbots"

made out of old walkmans, laser-printer cartridges, computer chips from birthday cards, calculators, and answering machines; etc.:

-- >can cost less than $100 to make a robot.

Future uses: herds of animal-like robots sweep floors, mow lawns or simply keep humans company.

2. Traditional AI researchers' approach

2.1. Recent work in Cognitive-Machines Group at MIT – “Ripley”
www.medi.mit.edu/cogmac/robot.html

“ our goal is to enable Ripley to perform collaborative manipulation tasks mediated by naturally spoken dialogue. Key issues include representation and learning of spatial language, object reference, and physical actions/verbs.

“… the architecture involves a physical robot (Ripley) and a virtual world that reflects Ripley’s mental model. Ripley is an interactive robot with vision, speech and grasping capabilities. The world model is constructed from a physics simulator that models dynamics of physical worlds. This framework provide a foundation for our ongoing experiments in developing new models of natural language processing in which words are grounded in terms of sensory-motor representations…”

2.2. Hans Moravec’s ongoing work at Carnegie-Mellon

www.ri.cmu.edu/~hpm/project.archive/robot.papers/2000/Cerebrum.html

Robot Evolution (according to Moravec)

First Phase: Programmable Manipulator - Industry Robot (phase we're in now)

Second Phase: Sensors and Effectors - he says achievable by 2010

Third Phase: Movement ( +beginnings of consciousness)

Fourth Phase: "Human-Like"---> symbiosis of human and robot. (by 2050)

What's involved in the "Second Phase"? : Moravec's Mass-Produced General Purpose Robot

Requirements: (see pictures of what this general pupose robot might look like:

http://www.frc.ri.cmu.edu/~hpm/book98/fig.ch4/p108.html

http://www.frc.ri.cmu.edu/~hpm/book98/fig.ch4/p124.html

http://www.frc.ri.cmu.edu/~hpm/book98/fig.ch4/p105.html

1. Minimum Level of functionality

--> locomotion: e.g., Hitachi Five Legs

2. Minimum Level of Manipulation

--> Salisbury Hand

3. Minimum Level of Navigation

--> sensory system - sonar

4. Minimum Level of Recognition

--> able to recognize and localize specific objects in the robot's surroundings

5. Software Applications (like PC based programs) supplied by vendors.

Third Phase: Moravec's Projections in 40 years

-- > reasoning ASTONISHINGLY better than humans

-- > perceptual and motor abilities the same as ours

Their role will be as our tools.

Fourth Phase:

Robots will evolve into Autonomous Beings: - once robots have an internal model of the world around them this beginning of awareness will evolve into consciousness comparable with that of humans

--> it will be a time when intelligent machines, however benevolent, threaten our existence because they are alternative inhabitants of our ecological niche.

According to Moravec, robots have the following advantages:

-->their production and upkeep cost less, so more can be put to work with the resources at hand.

-->they can be optimized for their jobs and programmed to work tirelessly.

-->we took millions of years to evolve; machines making similar strides in a few years.

According to Moravec it is inevitable that this will happen.

But what about human evolution? won't we continue to get better? nope

Will humans be able to share in this "magical world" [Moravec's term] to come?

Is the solution to:

1) replace a human brain into a specially designed robot body? (no)

or 2) get our mind out of our brain? (YES--in the future, we will undergo "mind transfer")

(If you think that your identity = your body, you'll be out of luck.)

Since for Moravec, a person is the process going on in his head, if you can move that to a permanent location, we will have merged with computers and ensured our immortality!