Presently we are exposed to the low level of Artificial Intelligence where a software some times embedded in a specific hardware such as a robot can do a specific job efficiently. Just as the Computer was initially used for doing repetitive tasks and work as a tool of the human, the Artificial Narrow Intelligence (ANI) can work as a tool to relieve humans of some repetitive tasks.
The ANI machines are as good as they are programmed. The programming of an ANI device includes the “Training” that it receives where the developer feeds standard inputs and commands specific types of responses from the algorithm. The response of such machines can be intentionally or otherwise biased because of the training process.
The next generation machines may have self learning capabilities based on the feedback received by the users in the process of its service. In such a process the continued learning may be a form of automated input generation. The quality of response may still be dependent on the quality of the inputs.
In such scenario it is possible to introduce some ethical guidelines to ensure that the algorithm will behave in a controlled manner.
At the third level, the “Deep Learning” or “Artificial Neuro networks” are algorithms that can recognize patterns in data and make decisions or predictions. In such a process there could be several layers of analysis as the learning process makes a step by step progress to reach its final decision which may be stored in the memory for responding to the standard use case.
As long as the AI is dependent on the training, its functions may be under human control.
However the latest generation of AI is designed to think of breaking out of the standard instruction led or prior data input lead decisions and adopt a human like approach of an alternative way of doing a thing.
It is in this context that it is felt that AI will become capable of self learning both from its own earlier activities (Meta Learning) or from exploratory expeditions. With its ability to gather large amount of data and process them, this ability to learn from learnings can create outcomes which are outside the imagination and expectation of the developer.
Such AI which is capable of applying its learning from one field in another will exhibit an ability which may be called “Artificial General Intelligence”. (AGI)
Legally speaking, at this stage the applicability of laws like Information Technology Act 2000 which attribute the activity of a software to its creator starts failing. We need new laws to take care of such scenarios.
While creating an Artificial Intelligence Act that tries to address the creation, maintenance and destruction of self learning AI algorithms is within the capabilities of the humans, there would a possibility of the Super Intelligent AI which will learn through its General Intelligence that it can break away from the limitations of human intelligence and the human controls and re-set itself to be beyond the capability of being human controlled is a possibility.
Those who are aware of “hypnotism” are aware that the hypnotist can implant a “key suggestive trigger” which can instantly bring out the hypnotized out of the trance or stop doing a specific task. Some stage demonstrators of hypnotism use the term such as “Sleep” to suddenly make the hypnotized drop into a deep sleep.
Similarly, we have seen in movies that Robots may have a “Kill Switch” which can disable it. But super intelligent AI can alter it’s program to kill the kill switch. Even the oft quoted rules such as “Dont harm humans” etc can be over ridden by the super intelligent robot and the robot can turn rogue at its own will. This may today be a move script but well within the realms of possibility in the next 50 years.
While Naavi is today advocating the AI regulation to ensure that human race is not destroyed by the future generation of AI led robots, Naavi feels that the human race should be ready to allow itself to be replaced by the new generation of robots as if they are our own progeny.
Today we are aware that our children are not necessarily an exact replica of their parents. The parents may be good guys but the children may not be so. Partly this may be due to the genes and partly it may be driven by the environment…like an orphaned child growing up in a hostile atmosphere.
Similarly we may have to develop a philosophical attitude that our robotic children who may be humans with implanted chips (Cyborgs) or Totally artificial robots are actually part of our own progeny and accept them as part of the new human race.
This perhaps is the “Techno Philosophy” that we need to develop as Plan-B if our Plan-A to reign-in the uncontrolled development of AI that could destroy the human race fails.