Skip to main content

Asimo, the best and last... of the modern day mechanical turks.

To those of you uninitiated to the astonishing revolution of advances that have attended the space of machine learning / AI in the last 15 years who gawk in wonder at the "amazing" tricks that Honda's Asimo robot can achieve be not fooled.

Asimo is the culmination of robotics done precisely the WRONG way.

What do I mean by WRONG? How can I state that given the seemingly impressive things Azimo can do.

Asimo can walk and turn in stride.

Asimo can pick itself up after falling.

Asimo can hold and manipulate objects.

Asimo can identify objects.

Asimo can open bottles.

Asimo can climb up and down stairs.

Asimo can identify obstacles and redirect walking path out of their way.

Asimo can truly run.

So why do I say Asimo is doing things the wrong way??

Because Asimo, unlike an animal capable of similar functionality is using tens of thousands of times the power to get it done.

Why is that important?

Power is how computational ability and requirements are metered out by reality. The more energy it takes you to do something to lazier you are doing it compared to some other agent that requires less power and does it with greater fidelity.

Case in point our biological cousins, a raccoon can do everything I described above. (Yes, including the opening of bottles)

Let me be blunt, Asimo is a parlor trick, a mechanical turk. A simulacrum of intelligence, Asimo doesn't learn at all like you or I or even a roach does, Asimo is a bunch of highly refined programs filled with conditions and sub conditions and should any external stimulus not be handled by those conditions Asimo fails and fails spectacularly.

True AI involves dynamic real time learning of the world environment it is embedded in and constant and continual refinement to that world. This is done not by using a programmed method where you try to identify all the possible cases of success and failure (a futile act just in the short thought of it) but instead in being able to experience *everything* and then determine salience gradually to the needs of the system.

This is what Marc Raibert realized in the 1980's when he was working at MIT's leg lab and it was this realization that led him to radically change how he thought about creating ambulating robots and what a change it has been.

Since that time his method has been adopted by computer scientists to solve all types of problems in machine learning that before were "solved" the same way Asimo does, brute force and programming.

No, no no....the ant has a fraction of a gram of brain and yet is able to forage, climb, descend and ambulate in real time far faster than the best robots we've yet created. How could an ant which requires orders of magnitude less computational ability do MORE than our best robots that require *pentiums* of computing power just to walk and then only if in a prescribed line with no rocks on the floor??

Well that is true for Azimo but not for the results of Raibert's vision. He founded Boston Dynamics in the 90's in order to build robots (with US government and thus citizen money) that could act and react biologically...applying the statistical learning approach he pioneered successfully in his days at the Leg Lab.

The results of that work are Big Dog, Little Dog, Cheetah and PetMan and Atlas.

These robots make ample use of sensors and memory to drastically reduce the computational load (and thus power requirements)  for performing ambulation but more importantly they allow the real time discovery of optimal solutions for performing the same...who did not scream in shock when they saw the robotic Big Dog slip but refuse to fall on ice in the youtube video that went viral?

Who did not gaze in astonishment as Atlas did pushups or balanced one one leg while being pushed by a human being in order to make it fall??

That's what you call robotic success, not a mechanical turk that only has a regime of success but a fully dynamic and agile system that can learn from its mistakes in real time and write its own programs for success into its ever growing memory of the world.

The fire that Raibert has lit has spread through out the field of machine learning and artificial intelligence. The neural network methods that saw traction in the early 90's are now being used in hybrid systems that combine intelligence and memory and sensing along with simple statistical algorithms to enable systems to learn about their environment by being taught (just as animals do). The statistical approach has been applied to solve more efficiently a host of problems that formerly were computationally far more difficult...from language translation to vision and classification to voice recognition. Those of you who have read my posts on my theory of cognition and intelligence understand that this statistical approach in my view is different between animals mostly in scale and some what in connection, how modules in brains are connected. The salience theory of consciousness I proposed last year from the culmination of my thoughts on the matter puts forward my set of hypothesis for how full cognitive function with consciousness can be emerged from the advanced versions of such systems. Raibert's work was always the "right" way to me from as early as 1989 when I read of the successes at the leg lab and of his views regarding biological ambulation. It was obvious to me he was right then...and now over 20 years later that vision is getting closer and closer to emulating biology at least in ambulation.

So next time you see Asimo doing it's parlor tricks at the next Car show...chuckle to yourself as you realize....the best is yet to come!


Marc Raibert:

Big Dog:




Wildcat (Cheetah):


The Turk:


Popular posts from this blog

Highly targeted Cpg vaccine immunotherapy for a range of cancer


This will surely go down as a seminal advance in cancer therapy. It reads like magic:

So this new approach looks for the specific proteins that are associated with a given tumors resistance to attack by the body's T cells, it then adjusts those T cells to be hyper sensitive to the specific oncogenic proteins targeted. These cells become essentially The Terminator​ T cells in the specific tumor AND have the multiplied effect of traveling along the immune pathway of spreading that the cancer many have metastasized. This is huge squared because it means you can essentially use targeting one tumor to identify and eliminate distal tumors that you many not even realize exist.

This allows the therapy for treating cancer to, for the first time; end the "wack a mole" problem that has frustrated traditional shot gun methods of treatment involving radiation and chemotherapy ...which by their nature unfortunately damage parts of the body that are not cancer laden but …

Engineers versus Programmers

I have found as more non formally trained people enter the coding space, the quality of code that results varies in an interesting way.

The formalities of learning to code in a structured course at University involve often strong focus on "correctness" and efficiency in the form of big O representations for the algorithms created.

Much less focus tends to be placed on what I'll call practical programming, which is the type of code that engineers (note I didn't use "programmers" on purpose) must learn to write.

Programmers are what Universities create, students that can take a defined development environment and within in write an algorithm for computing some sequence or traversing a tree or encoding and decoding a string. Efficiency and invariant rules are guiding development missions. Execution time for creating the solution is often a week or more depending on the professor and their style of teaching code and giving out problems. This type of coding is devo…

AgilEntity Architecture: Action Oriented Workflow

Permissions, fine grained versus management headache
The usual method for determining which users can perform a given function on a given object in a managed system, employs providing those Users with specific access rights via the use of permissions. Often these permissions are also able to be granted to collections called Groups, to which Users are added. The combination of Permissions and Groups provides the ability to provide as atomic a dissemination of rights across the User space as possible. However, this granularity comes at the price of reduced efficiency for managing the created permissions and more importantly the Groups that collect Users designated to perform sets of actions. Essentially the Groups serve as access control lists in many systems, which for the variable and often changing environment of business applications means a need to constantly update the ACL’s (groups) in order to add or remove individuals based on their ability to perform certain actions. Also, the…