Human Level AI

Question 1

Pathways to Human Level AI

Answer 1

Path 1: Use Genetic Algorithms

Path 2: Emulate a whole brain

Question 2

Two Assumptions underlying Whole Brain

Emulation

Answer 2

The non-organicism assumption

Scale Separation

Question 3

Scale Separation

Answer 3

Assumption underlying Whole Brain
Emulation

○ Total understanding of the brain is
not needed, just understanding the
component parts and their functional
interactions.
○ A simple example: Compilers do not
understand software and merely
perform syntactic operations that
transform human source code into
machine executable code.
○ Thus a Whole Brain Emulation
pipeline might (without any
understanding) mechanically convert
a physical system (a brain) into a
software system (a simulation)

Question 4

The non-organicism assumption

Answer 4

Assumption underlying Whole Brain
Emulation

○ There exists a cut-off point beyond
which we don’t actually need to
emulate in order to get the
functionality we want
○ An example: Modeling the
movement of planets, doesn’t
require a weather model of Jupiter
(or even further, of the Amazon
rainforest )

Question 5

Seed AI

Answer 5

machine learning without human intervention, in which an AI improves itself by recursively rewriting its own
source code. A process known as recursive self
improvement.

Turing (1950) said: “Instead of trying to produce a
programme to simulate the adult mind, why not
rather try to produce one which simulates the
child’s? If this were then subjected to an
appropriate course of education one would obtain
the adult brain.”

Question 6

Genetic Algorithms

Answer 6

Evolution gave rise to intelligent systems. Could we
emulate this process on a computer and get similar
results?

Arguments for using genetic algorithms to get to
AGI:
● Computational power is just rising and rising
and we can expect it to keep doing so
● Blind, undirected evolution gave rise to
intelligence one. That’s proof it could work.
● Surely then, a well-engineered evolutionary
process can get there much faster?

Arguments Against using Genetic Algorithms to get
to AGI
Not because it did it once that it will do it again
● There is a massive level of
selection/survivorship bias in the above
argument. The fact that we exist doesn’t
make our existence likely…It could, for all
we know, be the only case in the whole
universe! We could be extremely, extremely
lucky to be here. (I.e: Evolution is a very
narrow path)
● Natural intelligence and artificial intelligence
might not be the same thing. A plane can fly
not because it is working on the same
principle as a bird. They really are quite
different things.
● Computational power might not actually rise
high enough – could be an almost
unreachable number

Question 7

Emulate a whole brain

Answer 7

would also have other applications than general AI
● Step 1: Take a particular person’s brain
● Step 2: Scan it in detail
● Step 3: Construct a software model of it
that’s faithful to the original
● Step 4: Run on appropriate hardware where
it will behave in essentially the same way as
the original brain
● Step 5: ???
● Step 6: Profit
● Side note: Could potentially allow digital
backup of our consciousness / ability to
repair brain damage at will / immortality

Question 8

The Church Turing Thesis

Answer 8

● The Church Turing Thesis is simply the idea
that a Turing Machine can compute the
same functions as any other Turing
Machines.
● A machine is called Turing Complete if it
can emulate a Turing Machine. Thus all
Turing Machines are Turing Complete.

The Physical Church-Turing Thesis
● The idea that every function which is
physically computable…can be computed by
a Turing machine (Turing machines are
universal)
● Note that the function might be infeasible,
but at the very least, if a Turing Machine
had infinite time and infinite memory, it
could do it.

Question 9

Simulation of planet sized population

Answer 9

Proba of having the abilities
Proba of the the interest
Number of times simulation would run

If simulations can be possible, its likely we are ourselves in a simulation

Simulating evolution could mean simulating evolution to intelligence

Question 10

Levels of detail of brain emulation

Answer 10

• Brain emulation: software that models the states and functional dynamics of a brain at a relatively
  fine-grained level of detail
• Mind emulation: brain emulator that is detailed and correct enough to produce the
  phenomenological effects of a mind (e.g. self-awareness)
• Person emulation: mind emulation that emulates one particular mind