Many researchers appear to agree with my assessment that the human
blueprint is an unsuitable foundation for development - but some of
them think that building on the human brain is a good idea.
The human brain has been a big factor in the progress of civilisation
to date. However, it seems to me that it is an extremely poor quality
component - and will probably be among the first to be made redundant
by engineering and intelligent design.
In my view, the human brain is just as unmaintainable as most other
structures cobbled together by random mutations and natural selection
are. The effort required to duplicate its functionality in another
substrate would be enormous. Other ways of constructing intelligent
machines will arrive well before it becomes feasible - at which point
few will be interested in pursuing organic brain emulation further -
in much the same way that few are interested in building aeroplanes
with flapping wings.
First of all, let us hear from some of the proponents of the idea of
basing the architecture of intelligent machines on the same pattern
as is used by the human brain.
Here is IBM's Dharmendra Modha:
[Dharmendra Modha footage]
Here Ray Kurzweil:
[Ray Kurzweil footage]
The position I am going to criticise is that whole brain emulation is
a probable route to the construction of intelligent machines. The idea
is that we will effectively be able to scan an individual human's
brain - and then boot it up in cyberspace - before other routes to
intelligent machines become viable.
This is an extreme position. The general idea of using the brain for
clues which might help with constructing intelligent machines is a much
more moderate and defensible position - and I don't see much of a
problem with it.
However, here Anders Sandberg on the topic of how
whole brain emulation might work:
[Anders Sandberg footage]
Why consider such an idea? According to the proponents, the alternatives
have less predictable timescales. Here is Robin Hanson, explaining the
It is true that we do not know exactly how long it will take to
engineer machine intelligence. However, it seems pretty clear to me
that the task will be much simpler than booting up an
emulation of a human brain.
Here is Robin Hanson explaining the whole brain emulation approach:
[Robin Hanson footage - the basic idea]
One of the advantages of whole brain emulation is that the resulting
product has already been trained - and does not need to be taken
through a long and complicated process of education.
Having said that, such training is only time-consuming for real-world
robot controllers. If your training data is on the internet, or
in a virtual world somewhere, training can often be done at high speed -
thus reducing training times dramatically.
Another of the advantages of whole brain emulation is that we
may not need to understand the macro-structure of the brain in order
to reproduce its functionality - since that structure can be copied
from a mature human brain using a scanning process. Here is Robin
Hanson explaining that point:
[Robin Hanson footage - like a CPU emulation]
Alas, I do not think that the analogy between moving a brain to a
new computing substrate and porting a computer program to a different
architecture through emulation in a virtual machine is a particularly
good intuition pump in this case.
In the case of porting a computer program to a new architecture by
using an emulator, most of the complexity lies in what is being ported
across - and not in the emulator - else you wouldn't bother with
building an emulator in the first place.
If you are building a human brain emulator, the situation is pretty
different - as the macrostructure of the brain is mostly produced by a
process involving self-organisation of the microstructure elements, in
conjunction with the environment. In the brain, most of the complexity
comes from the microstructure, and from the environment, rather than
directly from the organisation of the macrostructure. Also, in a brain,
there is not really a natural split between the program and its
execution environment - brain hardware and software are muddled
A more general argument can be made - concerning the extent to which
engineered structures are usually heavily inspired by biology. This
also weighs against whole brain emulation.
The most commonly-given example is that of flight. Engineers did not
learn how to fly by scanning and copying birds. Nature may have
provided a proof of the concept, and inspiration - but it didn't
provide the details the engineeres actually used. A bird is not much
like a propellor-driven aircraft, a jet aircraft or a helicopter.
The argument applies across many domains. Water filters are not
scanned kidneys. The hoover dam is not a scan of a beaver dam. Solar
panels are not much like leaves. Humans do not tunnel much like moles
do. Submarines do not closely resemble fish. From this perspective, it
would be very strange if machine intelligence was much like human
So far, machines have mostly complemented humans - compensating for
their weaknesses. The first powerful machine intelligences will
probably arise from domains where they complement human
intelligence - under the influence of economic pressures to avoid
competing directly with humans.
Nor do other economic forces favour whole brain emulation.
The brain-scanning process gives you a fleshy-robot controller - which
you then have to find a body for - either in a real or a virtual
world. If you look at the organisations best placed to fund and
develop intelligent machines - search engine companies, hedge fund
managers, the NSA, and so on, then typically, they don't want a
human-sized intelligent agent, but rather something enormous with
Whole brain emulation is a technology which doesn't work -
and so which has no applications. That is not a situation likely to
promote funding or research.
The brain's design is awful. Tell a telephone engineer that he must
connect callers directly to each other with point-to-point cable
connections, and he will laugh at you and tell you that such a system
would never scale - and that you should read up on how voice over IP
Humans like to enslave their machines. If machines are people, how is
that going to work? It isn't. Who is going to buy a toilet-cleaning
robot that says that its ambition is to emigrate to Australia and
become a TV star? No one is.
Humans seem unlikely to stomach the idea of machines being people
initially. They will not allow them equal rights and opportunities
under law - and they will not allow them to vote. Rather there will
probably be an apartheid situation. The machines and robots will be
second-class citizens - and the humans will eventually come to act
like parasites upon them. Such a situation is likely to prolong the
viability of humans - thus allowing for a smoother transition where
there is a reduced chance of important things being lost.
Nor are human minds particularly friendly or safe to be around. They
get angry and aggressive too easily. The last thing humans will want
to have around are intelligent machines that have pychotic breaks.
So: aside from all the other problems, whole brain emulation is not a
safe route to intelligent machines.
Frankly, whole brain emulation seems like such a ridiculous
joke to me that I have difficulty understanding why it has produced as
much interest as it has.
A public relations exercise
My impression is that the concept of whole brain emulation is
primarily a public relations exercise. IBM's involvement seems to be
an attempt to reproduce the kind of publicity they obtained with
Deep Blue. Ray Kurzweil is a kind of cheerleader for high
technology, who promotes its benefits to the general public.
I can certainly imagine how the public might prefer to hear that
they will be able to upload their personalities into the matrix,
when the time comes - rather than having intelligent machines take
their jobs and leaving them redundant.
However, it seems to me that the credibility gap is very large here -
such a scenario is simply unbelievable - and I wonder whether people
are going to be taken in by it for very much longer.