Superintelligence by Nick Bostrom – Book Summary

Super Intelligence, Paths , Dangers and Strategies is a book written by Nick Bostrom. This is a 2014 book. Super Intelligence can be described as an Intelligence that outperforms the best current human minds. The author starts the book by writing about past developments and present capabilities in terms of machine intelligence. If the world economy continues to grow at the current rate as it has over the last 50 years, the world will be 4.5 times richer by 2050 and about 34 times richer by 2100 than it is today. Human Level Machine Intelligence (HMLI) is one that can carry out most human professions at-least as well as typical humans. The author shares some survey results and the respondents suggest that there is a 90% chance we will get there by 2100.  The google search engine is probably the best AI system that has been built. Ths U.S Department of State operates a face recognition system with over 75 million photographs for visa processing. There is a need for pre-installed and auto executing safety functionality as opposed to reliance on human supervision.

Paths to Super Intelligence:

There are multiple paths to super intelligence and it is not just one route. The ability to learn is key for individuals. The same holds true for systems. The first path through super intelligence is through AI (Artificial Intelligence). The systems are designed in such a way that they are able to learn. Alan Turing wrote about a child machine in the 1950’s. ‘Instead of trying to produce a programme to stimulate an adult’s mind, why not produce one that stimulates the child? Appropriate course of education would obtain an adult brain’. The existence of birds widened our imagination and led to the invention of flying machines. Turing’s idea of design is that machines acquire most of their intelligence by learning and it is not required to pre-program everything. Recursive Self Improvement is the key in AI. 

The second path to superior intelligence the author writes about is by selective breeding. It is possible without technology, but will take a longer time. This also has political and other consequences. Unless the selections are very strong, the author writes it will require many generations to produce substantial results. With human genetics, you produce very smart people with high intelligence. A world with a large population of such individuals will constitute a collective intelligence. The author writes that there could also be a cultural shift towards parenting norms that present the use of selection. The author writes that with further research and advances in genetic technology, it may be possible to synthesize genomes to specification. The author also writes on eugenics (the study of how to arrange reproduction within a human population to increase the occurrence of heritable characteristics regarded as desirable). The author writes that countries like China and Singapore may actively promote genetic selection and genetic engineering to promote its population. (Lee Kuan Yew wrote on similar lines in his book From Third World to First). The author also writes on In vitro fertilization (IVF) , the process of joining a woman’s egg and a man’s sperm in a laboratory dish.  (P:S: These are just notes and views from the author and I do not share these views.)

The author also writes on some wild ideas where you are able to read someone’s brain or communicate from one brain to another. Placing a chip outside the brain is far easier than placing one inside. Placing a chip inside the brain is also bound to have other health consequences.  Keeping machines outside the body makes it also easier to upgrade. Brain simulation could potentially help in the treatment of Parkinson’s disease. The Parkinson’s implant supplies a stimulating electric current to the subthalamic nucleus. 

The third path to Super intelligence is through networks and organizations. Networks collect a lot of data and it will help enhance the intellectual capacity of individuals which may lead to super intelligence.The author writes that soon, people might begin uploading continuous life recordings from microphone and video cameras embedded in their smartphones or eyeglass frames.

Forms of Super Intelligence:

The intelligence of machines will always be superior to biological intelligence. Machines have some fundamental advantages.  Speed Superintelligence is a system that can do all that a human intellect can, however much faster (e.g a calculator). Collective Super intelligence is a system achieving superior performance by aggregating large numbers of smaller intelligence. This refers to problems that can be broken down into parts and solved individually and we group them all together. This is what happens in teams, workplace, cities , countries and more. The third form of super intelligence is Quality Super intelligence (system as fast as a human mind, but better in terms of quality).

Resistance and Strategic Advantage:

The author writes about recalcitrance (stubbornly resistant to authority, control, or orders) and the strategic advantage one would have by super intelligence. The author is of the opinion that it depends on how fast super intelligence spreads. It could be a slow, medium or a fast process.  Nick Bostrom also shares some historic examples. E.g China managed to maintain a monopoly in silk production for over 2000 years. Porcelain used in Tableware, Bathroom fixtures dates back to AD 600, however it was mastered by the Europeans only in the 18th century. Human species took tens of thousands of years to spread across the globe, the agricultural revolution took thousands of years, the industrial revolution hundreds of years and the information revolution spread in decades. The author also shares an analogy of two cyclists climbing up the hill, one reaches the top and starts descending and the other is still climbing. The one descending is bound to have an advantage. The super intelligence could also come from a small group of people who succeed in putting everything discovered together. With super intelligence, the author writes that there could also be State resistance (Government) and it takes time for the government to understand things. The author also discusses the collaboration between countries on super intelligence and working together (e.g International Space Station where the United States and Russia worked together).

Control: There are multiple ways we could control the AI. One is by offering less hardware compute limiting the AI to perform its functions. Another approach is boxed where it is tasked to perform only a certain set of tasks and functions. Motivation – An AI is trained to operate in the Principal’s interest. Tripwire is another approach, similar to a circuit breaker. The AI is being monitored and if it does not perform in the right interests, it is automatically disabled.

Future of Work: The future of work and how we are compensated is a question we need to ask. For example, humans commuted with the help of horses and carts. The demand for horses was a lot. However with the invention of combustion engines and transportation, the need for horses reduced. This also contributed to the reduced population of horses. In the US, there were 26 million horses in 1915 and that number reduced to 2 million by the early 1950s. What does this mean to us? The human population has increased thousandfold over the past 9000 years. How will the tools and AI being created today reduce the demand of human work? If a machine can replace humans and if the cost is less, we would prefer a machine over a human to do our work. (in some cases this might differ). In communities like jews and muslims, they have a certain way of making the food and this is a good example of imposing restrictions on how work is being done.  The author also asks if the AI plans be put to human review before being deployed? 

The author also shares information about cat burning which was popular in the 16th century in Paris. Slavery was widely practiced until about 150 years ago. Though these are wrong, people still did it. Nick compares this to AI and how people may use it. 

My Take on the book:

This was not an easy book for me to read. Maybe because I did not understand everything it was written or I found it a bit dry. It took over 6 months, on and off to complete reading this. I would not blame this on the author, but it has to do with me. I also do not agree with everything the author writes.  Nevertheless, I could relate to the first few chapters of the book and post this, I could not stay connected. Thanks Nick for writing.