Thursday, June 28, 2018

Humanoid Robots May Be Imminent

Star Trek: The Next Generation, set in the 2360s had a crew member, Data, that was an android or humanoid robot.  While a Data is not likely any time soon, a reasonable facsimile may be available far sooner than is generally imagined.  Current technological progress is not inconsistent with a product introduction around 2030.  However, while completely functional, it will in no way be artificial general intelligence as imagined in ST:NG or by Singularitarians.

The problem of creating a humanoid robot is comprised of two pieces -.  one, the robotic body, two the artificial intelligence.  Both technological pieces are currently exploding. What is necessary, primarily, is to bring together the various technologies that are currently being developed into one robot.

First, about simulated artificial general intelligence (AGW)
In 2005, Ray Kurzweil published, The Singularity is Near.  I had not yet started blogging, but I argued vociferously within my intellectual circles that, because Moore's Law would not continue to 2040, the Singularity was not only not near, it likely was far, far in the future.  I got nothing but push back at the time, but now most observers recognize that Moore's Law is dead and that progress henceforth will be much slower.  There are, of course, those who still believe that we are in a pause and that Moore's Law will reignite.  Many of them still pursue the goal of AGI (Artificial General Intelligence) which is a cornerstone of the Singularitarian vision.  While that is both anthropomorphic and overblown, if not totally improbable, simulated artificial intelligence is right around the corner, it will likely be given an I/O modality that will be a very credible humanoid.  I am not talking about in the mid 2040s; I am talking about something closer to the late 2020s.

Right now, there are quite a few natural language internet search engines.  Apple's Siri, Microsoft's Cortana, Amazon's Alexa or Android's Google Voice are the mostly widely used.  With these systems, the user asks a question of the 'assistant' verbally and they receive an answer, not from a third party website, but from the virtual voice assistant.  Sometimes, as Siri does quite often, it will be with a citation, such as 'According to the World Health Organization, France has the best Health Care system in the World'.  

These natural language interfaces are becoming more and more capable, accessing your calendar, Google Maps, Pandora, Waze, etc.  "Out of the box", Siri will not play chess with you, but that is a simple interface problem that techies can and do solve.  The same is probably true for IBM's Doctor Watson or other medical databases.  In fact, Siri can piggy back on IBM's Watson software.  In other words, making Siri or any other virtual voice assistant appear to be as smart as the total Internet is just a simple problem in programming.  In fact, likely, within five to ten years, natural language assistants will be able to do just about anything you want them to do.  They will likely pass the Turing Test and seem to be an insanely knowledgeable and capable human.

There is no doubt that such an advanced Siri or Alexa is not true Artificial General Intelligence.  When it interacts with the user, it is acting as a natural language I/O device.  When it plays chess with the user, it is accessing programs that determine the next best chess move through algorithms, not through anything resembling true thought.  Siri or Alexa is not conscious.  It has no reflective ability that allows it to contemplate its own existence.  One may argue that the Internet, collectively, is intelligent, but it is not so in the same way as we mean when a human is intelligent.  It is most definitely an artificial intelligence and the 'general' in AGI is simulated.

The 'Personal Assistant' Robot
This 'personal assistant' can, today, easily be put in a robot.  In fact, right now, it is being put in telephones.  The PAR is farther in the future than the simulated AGI into which the current personal assistants are morphing.  However, they are not that far in the future and the first ones will probably appear in about a decade.  Already, Boston Dynamics has plans to offer its 'spotmini' to the general population.



********
Unbiased analysis and commentary at the highest level of intellectual sophistication 
Subscribe to The Polymath.
********

By this method, the PAR it will accurately simulate AGI and appear to be a conscious entity more knowledgeable,  and expert than any human.  However, as we know with current personal assistants, chess programs, etc. it is not AGI.  It is SGI or simulated general intelligence.  While the quest for AGI will likely continue in the lab, SGI will be a cheaper and less troublesome product.  SGI is the future and it is here in rudimentary form, already.  And the competition is stiff and improvement is therefore rapid.

Scene analysis is critical to the effectiveness of a PAR.  Fortunately, due to multiple applications, the technology is advancing rapidly.  Clearly, Boston Dynamics has installed relatively good software.  This acrobatic display by their Atlas requires astonishing scene analysis.  It may be that soon their robots will exceed human capabilities.  Obviously, the average human could not duplicate this display of scene analysis and body control.




Not surprisingly, but equally important, Boston Dynamics Atlas robot can run over uneven terrain with amazing ability.  So, we see that the technology needed to allow robots to move about freely in a human or nature environment is already here.  The question is, how low can they get the cost.  A million USD robot will have very little consumer application but a 25,000 USD robot probably will assuming that it comes with sufficient manual dexterity and SGI.

Boston Dynamics is not a leader in robotic hands and fingers.  This research and development has been driven primarily by prosthetics and it is already very advanced.



********
I am recruiting intellectually sophisticated bloggers to create a weekly pdf magazine of analysis and commentary.  Participate in the innovative  crowdfunding of The Polymath
  25,200% return is our goal.  And, there is the added benefit of discombobulating the propaganda machines of the Liberal and Conservative news media   Learn all about it
********

The construction of human-like bodies and faces, driven, in part, by competition between sex robot manufacturers, is progressing rapidly.  The companies that are creating the most humanlike faces are not at state of the art voice interaction, but, while not at the point of fooling anyone are getting quite good.  It is important to note, especially those presented by Hanson Robotics, that many videos of human-like robots are scripted and presented as real AI that is well beyond current technology.

However, as Siri and Alexa demonstrate, the current AI is quite sufficient to function as a useful personal assistant.  We are seeing that these executive programs are advancing rapidly.

It is clear that bringing together the most advanced current technology, in just a few years a first shot at a humanoid robot would likely be successful.  The business potential may be enormous.

Of course, the likelihood of market acceptance will be based, among other things, upon price vs. benefits.  One of the primary benefits is to organize and execute the details of one's life.  As Google demonstrates here, much of what can be done already. However, picking up dry cleaning, grocery shopping, etc. will require a physical presence.

The other significant benefit is as a domestic, that cleans, cooks, gardens, does home and auto maintenance, etc.  It is, in fact, the primary benefit of a humanoid robot, over a virtual robot, such as Alexa.

Robots have no life of their own, so the only down time is for maintenance.  They may need 10% for charging and maintenance.  At $10 per hour equivalent, the net present value of a robot's work will be around 500K USD.

While the present value may be 500K USD, the cost will need to be less in order to be a successful product.  A basic robot may cost the same as a luxury car, say 100K USD.  There will also be some operating costs.  A total monthly cost may be around 1,000 USD.

With, say a 2% initial market penetration, the industry may start St about 100 billion USD but has the potential to grow much, much larger.

The basic models will likely look like Atlas or Asimo.  Many people find the more human robots 'creepy' and they actually prefer the Asimo style models. There is controversy right now over Harmony, the sex doll, by Real Doll.  Real Doll and and it's competitors are naturally the most aggressive group in pursuing human like-robots.  On one hand, most people will likely have a negative reaction to this degree of human simulation, but on the other hand, some people may be willing to spend substantial funds to create very life like and custom designed robots, a la Stepford Wives.

Most humanoid robots are currently not for sale, the exception being Harmony, which costs 15,000 USD.  She is, however, far from complete, with no movement below the neck.  However, if the initial price point is 100,000USD there is plenty of room for improvement.

There are many activities that will be appropriate for PARs, but there will also be plenty of roles where people will prefer humans.  Home domestics, hotel house cleaning, etc. will likely be robots, however, fashion consultants, personal trainers, wait staff, doctors, police, etc. likely will remain human because people will prefer the 'high touch' of real people.