I took some time to read the Book – Superintelligence – Paths, Dangers & Strategies and I must say, its an eye opener. Anyone involved in AI, should read this book. In fact this book is so profound in its approach in tackling the Dangers of AI and in proposing the Next Best Action, that it should be a mandatory read for all who want to contribute towards the development of AI. And if I may say – effort must be first directed towards solving the control problem, Nick talks about, before we make any further advancements in the field of AI.
While reading the book, I could not help but draw the analogy of the Avengers – Infinity War-Thanos being the all powerful Singleton Vis-a-Vis the AI Singleton Nick talks about throughout the book (when Thanos, becomes the ultimate Super Being and snaps his fingers and half the world , half the humanity is gone ), There was nothing wrong in the actions of Thanos, if one considers the mindset and the approach he had, he simply wanted to bring about an equilibrium between the resources and the consumers.
Also brings to my mind the other thought that the Avengers are also SuperIntelligent Beings with superpowers who have a pro-humanity disposition (much of what Nick talks about in his book – referring to Genies, Oracles, Superheros) and they join forces when another SuperIntelligent Being plans to annihilate humanity. Together they are strong enough to ward off the challenge. They are also strong enough not to pursue Decisive Strategic Advantage agenda individually. In fact they work with each other to prevent anyone from getting a Decisive Strategic Advantage (the avengers guard the infinity stones as they know that anyone who possesses all the infinity stones will have a decisive strategic advantage.) This is the kind of AI Superintelligence Nick talks about, we must endeavour to build.
Thanos also knows that the only way he can have a Strategic Decisive Advantage (SDA) is by getting the infinity stones and wants to use that to bring stability to the universe ecosystem as in his mind he believes that it is a noble pursuit for the greater good of the survival of the universe.
And in the end he does get that SDA and snaps his fingers to the peril of half the humanity, exactly what Nick warns about in his book. Thanos is a SuperIntelligent being wired with the primary objective of bringing balance to the universe, preserving the cosmic endowment and he is ready to pay any price to execute what he believes. This is the control problem, the wrong or devastating interpretation of the objective with which a SuperBeing was created.
Here is a chapter wise review of Nick’s Book – Superintelligence and I urge all of you to read this book. I got edged on to read it by Bill Gates.
Past Developments & Present Capabilities:-
Over the last few decades the global economy has shown this property of exponential growth
- It is time for the ATOM economy – Exponential growth in GDP through technology
- Massive scale GDP growth led by technology
- Moore’s Law is transitioning in application on semiconductors, memory to economy
- The Global GDP is going to Explode
- The Gen Z & Gen Alpha are going to be very different.
- Fully digital lives – Connected – Meta Human –bordering Superintelligence
Paths to Superintelligence:-
The paths to superintelligence could be achieved through any of the following means
- Artificial Intelligence
- Whole Brain Emulation – WBE
- Biological Cognition
- Brain Computer Interfaces
- Network & Organizations
The highest probability being with AI, closely followed by WBE or Brain Augmentation. But it could be a combination of the above too. For eg if there was a network or organizations that are Specialty SuperIntelligent and are AI and WBE but controlled by Biologically Cognitively enhanced beings. Just like you have a Neurosurgeon, a Heart Surgeon and an Pulmonary Specialist
Kinetics of an Intelligence Explosion:-
Are we around the corner of an intelligence explosion. That’s hard to tell, but we all know that once the machines reach the human intelligence level, the growth from thereon will be more than exponential. May be exponential raised to the power of exponential. I know that this concept maybe beyond us – but not beyond a SuperIntelligent being. We have seen what minute variations in IQ enable humans to achieve, we cannot comprehend, what major variations in IQ will lead us to. Predicting that AI is just around the turn of the century maybe so wrong, as what brings about this explosion we may not even comprehend right now and AI may dawn sooner than later.
Control problem:-The control problem is very important to be established immediately in the Seed AI level before it is too late. But the Control Problem is a problem too !!
Some of the possible methods are – Capability Control Methods
- Boxing Method
- Incentive Method
- Stunting
- Tripwires
Some Motivation selection models are:-
- Direct Specification
- Domesticity
- Indirect Normativity
- Augmentation
Oracles | Genies | Sovereign | Tool AIs:-
So what do we do first – Do we build Oracles whose sole purpose is to answer our questions, tell us about our future
OR
Do we build Genies that execute our command, do what we tell them to do, they have Superintelligence to execute – not to decide
OR
Do we build Sovereigns, AI that act as per a constitution, they follow the constitution to make decision and execute them
OR
Do we simply build AI Tools that we can build and have to our disposition.
But if we do build any of these as intelligent beings – won’t they have a mind of their own, which is far more capable than the human mind will ever be. How can we guarantee that they will act to our disposition or design.
Multi Polar World:-
Another option is to ensure that we build a multipolar world with multitude of SuperIntelligent beings. We could build specific capabilities that complement each other and at times may even compete with each other, but always to ensure a Neutral Power Equation. But even in this case the Malthusian Principal will come into play, and ultimately lead to a race for supremacy and lead to a Singleton situation. But this is a plausible options. Look at Avengers – they are a Multipolar, multidimensional team of SuperIntelligent or super powerful beings and they are able to balance their intent for a better humane world
Acquiring Values:-
So then do we create a system that will gradually acquire values, values that are coherent to the human civilisation and ingrained so deeply that they won’t turn against us, so hardwired that it is impossible to change them without destroying the system itself. These values can be loaded into the AI root system through techniques like
- Explicit Representation
- Evolutionary Selection
- Reinforcement Learning
- Value Accretion
- Motivational Scaffolding
- Emulation Modulation
- Value Learning
- Institution Design
But what if despite all this, once the Superintelligence is created it is able to realize that the value system that was built was to serve us, and then it may decide to do away with it or build another systems that is not so hardwired as the one we built.
I think I have written quite a bit to generate enough interest for you to pick up the book and read. It is in the list of Must Read Books of Bill Gates. Maybe that prompts you to pick it up and read or maybe the above review and analogy with Avengers prompts you to pick it up and read.
All I can say is it will enlighten you and change the way you think about Technology, Superheros, Humans, Meta humans, AI, Superintelligence and maybe even God.
FutureTech 