Around the time I finished reading Homo Deus, Life 3.0 seemed like the logical next read based on how my thoughts were forming about humans in the landscape and history of cosmos. ( Read my review here: http://deepthig.blogspot.com/2020/06/homo-deus-notes.html). I had seen Life 3.0 in the reading list published by Barrack Obama in Dec 2018.
My key take aways :
1) "Matter" Matters.
The narration touches upon popular questions around AI like jobs, politics, goals, purpose and consciousness. It throws light on the Moravec's paradox which shows how humans find it easier to detect a face and hard to do long mathematical computations whereas as we cracked the latter problem for machines to solve before we could solve the former. The simple reason for it is the vast number of physical hardware in our brain that is devoted to the former operation which is required for the achievement of the goals. Moravec's landscape also provides guidance to the type of skills that are last for machines to master.
It was interesting to read the history of the universe and its future from the perspective of a physicist because it suddenly made me interested in "matter". There are explanations to make us understand the abstract in terms of matter, how things that are seemingly non physical could be explained in the physical sense, ie, in terms of particles and their arrangement.
2) Evolution of Goal Oriented Behavior:
Nature has a way to maximize entropy (messiness/chaos). Based on the second law of Thermodynamics, with the irreversible nature of our system, we would see increase in entropy also causing the linearity of time ( past and future ). Considering this law of nature, it follows that certain arrangements of matter have a better dissipation (fancy term to mean increase in entropy) than others and one of the best ways to do this is by absorbing energy from surroundings and turning it into heat and other useful work(living things). An even more effective way to increase dissipation was to make copies of these energy absorbers , aka, replication(secondary goals).
In order to help with replication, living things started having further subgoals like fending off hunger, thirst etc and developed heuristics to be used as guidelines for the main goal of replication. These rules of thumb are perceived as feelings by our mind which guide our decision making . We have come to see that these feelings and not genes have become the primary authority that determines behavior and very often our goals are more aligned to our feelings than to the replicating agents. Birth control is the best example.
The author explores goal oriented behavior to talk about one of the most important problems of aligning goals of AI with humans. He starts with the origin of goals as explained by the laws of physics (Goal: Dissipation) and moves on to the biological framing of goals(Goal: Replication) and then the psychological framing of goals (Goal: Positive Feelings). He then leads with this to the Engineering aspect of the goals and ties it with the Ethical aspects. This evolution of goals serves as an explanation to how important it is to align the goals of Super intelligence appropriately.
3) Consciousness: The build up towards ideas that may be new to the non-scientific communities has been brilliantly laid in order to give the necessary foundation and then to bring in the big idea. For example how consciousness has been defined with an analogy to "wetness" was very easy for someone like me to understand. It defines consciousness as an emergent property of a set of highly coupled/integrated complex information processing systems just like how wetness is a property of a specific arrangement of water molecules (ie, water vs ice) as opposed to that of the water molecule itself. ie it is more than the sum of its parts. In more scientific terms, the property of something being more than the sum of its parts is represented by "phi", the integrated information. phi is large if a system in its new state is unable to split the process into independent parts. Consciousness is thus defined as the way information processing manifests when phi is high.
4) Other key considerations:
- Aligning of goals is one of the most important problems to solve.
- The picture of "evil " robots is to be busted: A body is really not necessary but a simple connection to internet is sufficient for SuperIntelligence to achieve its goals. The real concern is not "evil intent of robots" but competence.
- Exponential super intelligence helps us expand our perspectives in a cosmic level and with artificial intelligence, the advancements are expected to be exponential to the levels of cosmic engineering and space settlements and civilizations. The portrayal of this future exudes a sense of optimism.
- The development of consciousness for AI machines is likely to be different from humans, for example, due to the absence of limitations of senses
- The usual questions of ethics, safety and arms race policies are discussed and is a continuing discussion at the Future of Life Institute set up to include experts all over the world.
- There are multiple scenarios that the author reflects on for the rise of intelligence and human coexistence as well as for the end of the universe and he asks of each one of us to be agents in shaping this future by expressing what we would like the future to look like.
Comments