The biological brain is vastly more complex than any machine intelligence. Deep learning and its offshoots are impressive in their computational ability to solve a variety of classification and control tasks. Transfer and Zero shot learning are developments in ML that enable problem specific algorithms to solve and recognise new
problems and classes that previously werent trained.
Problem solvers that share learning solutions for different problems may be one step towards modularity and plasticity - the two abilitys that enable the biological brain to adapt to new functions using existing wetware.
Borrowing from nature and extending existing technology to perform better in a variety of problem domains depends heavily on what problems they are designed to solve.
A goal directed machine composed of various modules all solving their discrete problems in a hierachical structure has long been posited as a means to acheive a state machine that exhibits biologicaly plausable goal directed behaviours.
Such a machine would have as its constituents primary modules of vision, language, and speech, mobility - as required.
In fact such a machine could behave with a biological intelligence and be comparable to an insect level intelligence and yet be in command of a robotic exoskeleton that with a minimal intelligence could navigate through a real environment using limited visual cues and employing a motor control function that responds like a zombie to external stimula.
Machine vision neednt capture all the exentricites of human vision in order to solve huiman level problems such as navigation and object avoidance.
In fact there are lots of examples in nature where by the vision compnent is merely a byproduct of photons exiting cells and creating certain variations in chemical activity's.
It is entirely plausable that we neednt capture human intelligence in order to complete human tasks. What are the raw compnents for a machine to solve a human taks within constraints of affordance. A machine with high processing power need'nt be set to task on data of an overly complex nature. We expect too much from these primal archectures to solve classifications of an overly complex nature when this computational power could better be applied to a more simple task of distinguishing light from dark.
With this one simple function a machine could employ a motor control function based on a prelearnt reinforcement stragety that uses variations in light to guide itself through and environment.
With yet another module that recognises objects from edges it could locate these objects.
With an ability to find the invarent shape in moving objects it could very easily track a moving object through its environment.
These are all basic constituents of an heirachical AI system capable of performing human-like tasks that don't rely on biologically plausable intelligence.
So much AI these days doesnt consider this more simple option and instead apply's itself to tasks of increasing complexity in order to prove its power in solving problems that are not in the domain of any intelligence apart from the complex nature of the data we train them on. This isnt machine intelligence rather it is a function of complexity of data. The real world is simple. Life is simple. It quite often uses the minimum function to interpret only mwhat it needs to survive in its world.
We have created and endowed our world with complexity and data that to a machine intelligence is superfulus to requirements.
We are always considering how to build a machine that can learn more dimensions and acheive a greater grasp of complexity and fidelity in continous real-time data. When in actuality there are far more intelligent biological examples that do not require either in order to function the photosensitive cells the many marine species being but one example.
Mimicking human fidelity in machines isnt necessary neither is surpassing it with our own peculiar choice in complexity that we have ourselves created from our over evolved data strutures.
Life on earth machine or otherwise didnt evolve from being continually exposed to someones facebook. We often estol virtues on our AI as being helpful and useful in our world of emmersive connectivity. But in fact most life evolved in isolation from anything remotely complex and its environment was a foreign world in which it tested its neighbouring spaces tentatively there was no emmersion aside from the fact that it belonged to that environment just as much as any rock or leave existing in perfect isolation harmonious or otherwise.
We have extended our notions of God and Society to encompass all life and that nothing can exist without us knowing of its existence. Control is absolute and we continually search for more ways to surround our world with an overabundance of meaning.
20 years ago AI could have thrived in this world but unfortunately complexity won out, now we are entering a new world of complexity and fidelity as we have enhanced ourselves instead of creating machines that we fear would outdo us if given the chance.
Ofcourse with every revolution there is a counter-revolution. Robots do indeed exist humanlike and capable of exhibiting preprogrammed behaviour that we would normally only give to humans. But they dont possess a mind and have limited AI. Their coordinated abilitys are just as agile as any machine all they await is a modular based intelligence that can breath life into the machine.
Ofcourse this doesnt mean it wont happen it just means when it does it will be from a
level of fidelity and complexity that approaches God and not Life.
We have collectively decided that AI could be devestating to mankind due to its computational power being far greater than man's. Therefore we have stimied its progress. Allowing it to either be God like or to do exactly one thing very well therefore complementing our already exagerated and self-obsessed intelligence.
Eventually the simple life of a robot intelligence given free reign and having a will and intelligence of its own will break through of these human centered control. It could be self driving cars, robotic construction workers in space or on other planets.
But when it does happen we all better watch out.
I see three stages in creating a robotic revolution.
- Exoskeleton control of humanoid robots. Achieved.
- Quantum Entanglement as an instantanous high bandwidth communication over long distances. Formative.
- The colonisation of other planets with the aid of VR teleoperated humanoid robots. Formative.
At least when it does happen it will be on another planet. jUice.
Brain Synthesis...
