I shall not be collecting statistics from this, merely trying to get a feel for the spread of views. Most of the questions take the form of a statement associated with a set of possible answers expressing agreement or disagreement. Select your answer by putting an X between one pair of brackets, or deleting the answers you disagree with, or any other way that suits you. If you don't like any of the answers to a question, feel free to send me your own answer. If you don't like the wording of some of the questions feel free to send me your preferred formulation. If you think I've missed out some important questions feel free to let me know. I am sorry some of the questions are bit a long-winded: I was trying to prevent some types of ambiguity. Later I'll produce a summary of responses. Tell me if you want your contribution to be excluded, or included anonymously. There are no prizes, not even for guessing what my answers are. The list of questions below may be modified in the light of feedback. If you are able to provide answers to these questions, please copy the text below, paste it into an editor (preferably keeping line breaks) select your answers and post them by email (not paper, please): To: Aaron Sloman (A.Sloman@cs.bham.ac.uk), Subject: Answers to questions about AI/Biology A plain text (.txt) format is preferred to others (e.g. doc pdf html).Each question makes a statement, with which you are invited to agree
Please indicate what you agree/disagree with or have no opinion on.
(1) AI overlaps with Biology, since it includes attempts to model
animal intelligence:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(2) AI has nothing say about animal intelligence because animal brains
are so different from computers:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(3) Machines that have very different shapes, sensors, and effectors
from humans can have cognitive systems that share many features of
humans, despite the physical differences:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(4) It is very clear what the main unsolved problems in AI are, and what
needs to be done next in order to make progress towards solving
them:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
[Feel free to send me your list of major unsolved problems in AI]
COMPOUND QUESTION:
AI techniques in the following fields have made so much progress
that the problems are nearly solved -- all we need is more computing
power and another three or four decades (negative answers here do
not imply impossibility, but do imply disagreement about the time
scale, so if you think something is impossible say so):
In the next three or four decades:
(5) Replicating human visual capabilities will be achieved
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(6) Replicating visual capabilities of good guide dogs for the
blind will be achieved:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(7) Replicating visual capabilities of some other animals will be
achieved:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
If yes or Maybe please give examples of the animals
[ ]
(E.g. What about bumblebees?)
(8) Replicating development of young humans in the first three
years, including vision and other forms of perception,
locomotion, manipulation of objects, social interactions
will be achieved:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(9) Replicating development of some other vertebrate, e.g.
orangutan, squirrel, raven, will be achieved
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(10) Replicating human abilities to perceive, understand, and move
around purposefully in, a wide range of terrains, natural and
artificial, and to describe routes taken, explaining choices
made where there were alternatives, and noticing new short-cuts
that become available (eg because a gate that is normally locked
has been left open):
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(11) Replicating some non-human animal abilities to perceive
understand and move around in a wide range of terrains,
including noticing short cuts on the way back.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(12) Replicating a human child's ability to look at picture of a
meccano model of a toy crane like this:
http://www.cs.bham.ac.uk/~axs/fig/crane-pic-small.jpg
and construct an instance like this from a box of parts
http://www.cs.bham.ac.uk/~axs/fig/crane-built.jpg
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(13) Replicating a human's ability to point out discrepancies
between the picture and the built model?
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(14) Replicating a human's ability to explain how the crane works,
and why the various parts are needed.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(15) Replicating a human child's ability to dress and undress toy
dolls of various kinds, including selecting appropriate items
of clothing of various sorts jumbled in a drawer (e.g. under
clothes, shirts, dresses, trousers, belts, hats, boys clothes,
girls clothes, unisex clothes, etc.)
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(16) Replicating human abilities to perceive manipulable structures,
and abilities to perceive *processes* in which such structures
are changed, rearranged, assembled, disassembled combined with
others.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(17) Replicating some abilities of some other animal to perceive
manipulable structures, processes in which such structures are
changed, rearranged, assembled, disassembled (e.g. birds
building nests, carnivores consuming prey)
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(18) Replicating human abilities to work out actions that suffice to
manipulate 3-D structures, e.g. to assemble or repair a machine;
and to carry out the task, to explain what was done, and why
other things were not done, to answer questions about what would
have happened if something different had been done, and
sometimes to propose better ways of achieving the same result as
a result of performing the action.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(19) Replicating human abilities to tell when a child is likely to do
something dangerous or something annoying to others, and taking
action to distract or divert the child.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(20) Replicating human abilities to learn to reason about topology,
geometry, sets, numbers, counting operations, orderings,
properties of different kinds of stuff (e.g. mud, sand, water,
skin, writing paper, tissue paper, cotton wool, cloth, leather,
wood, plastic, metal, cooked meat, blamange, jelly, ...):
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(21) Replicating the human ability spontaneously to notice that some
generalisation discovered empirically is actually a necessary
truth (e.g. counting the same set of objects in two different
orders produces the same result, any three non-collinear points
define a unique plane through the points, containment is
transitive, two meshed gearwheels made of rigid impenetrable
stuff can only rotate in opposite directions, ...):
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(22) Replicating human abilities to learn languages in various ways,
including learning sign languages if deprived of hearing (this
includes having a physical body with parts that can be moved in
ways required for sign language -- hands, face).
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(23) Acquiring new tastes, preferences, ambitions, values, ideals,
likes, dislikes, attitudes, in the same way as humans
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
Being able to enjoy making/doing or experiencing music, poetry,
stories, jokes, acting, playing competitive games (physical and
intellectual).
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(24) Being able to help another individual acquire some competence or
skill by providing appropriate help, challenges and advice
(scaffolding) while interacting with the learner e.g. by playing
games, setting problems, asking questions, making deliberate
mistakes, etc.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(25) Experiencing philosophical puzzlement e.g. about metaphysical,
epistemological, ethical, aesthetic problems, and forming
opinions about some of the answers:
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[]
(26) Being able to form an independent opinion as to whether the
question "Do other individuals have qualia?" is a sensible
one, and to produce arguments for or against various
controversial theses regarding what machine intelligence can
do.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[] Undecided[]
[An earlier version of this question was ill-formed as a result of
an editing error.]
====
[Back to self-contained questions.]
(27) A highly capable human personal assistant uses knowledge and
competences acquired through a normal human developmental
trajectory, learning things about space, time, motion, causation,
the effects of human actions, the variety of sources of information
available to humans, the forms of reasoning humans have, the
varieties of motivation humans have, the ways humans communicate,
travel, organise their lives, the buildings they live in, the ways
weather affects them, the kinds of sustenance they need, etc. An
intelligent artificial personal assistant will need to have a
similar broad and deep background, which cannot be acquired by
statistical learning based on observation of human experts.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[] Undecided[]
(28) Biological evolution discovered, long before we did, that
making use of virtual machines was a requirement for solving many
complex design problems, including problems of self-monitoring, self
control and development. We need to understand natural architectures
that make use of layered virtual machines and the benefits and
problems they lead to:
http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#wpe08
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[] Undecided[]
(29) All the above problems will be solved well enough for the
solutions to be deployed in commercial products within
5 years []
10 years []
30 years []
50 years []
100 years[]
300 years[]
(30) Making significant progress towards the long term goals of AI
requires a breed of researchers with broad and deep knowledge of a
variety of disciplines, including mathematics, computing,
philosophy, psychology, neuroscience, biology, linguistics,
education, ....
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[] Undecided[]
(31) It is extremely difficult to find researchers who have the
required broad and deep knowledge, and our educational system is not
equipped to produce them. As a result AI progress will continue to
be patchy and localised, with many factions competing for funds
instead of collaborating and researchers constantly promising to
deliver results 'soon'.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[] Undecided[]
(32) John McCarthy wrote
"Evolution solved a different problem than that of starting a
baby with no a priori assumptions."
.......
"Instead of building babies as Cartesian philosophers taking
nothing but their sensations for granted, evolution produced
babies with innate prejudices that correspond to facts about the
world and babies' positions in it. Learning starts from these
prejudices. What is the world like, and what are these
instinctive prejudices?"
(The Well-designed Child, AIJ December 2008)
http://www-formal.stanford.edu/jmc/child.html
As implied by McCarthy, AI systems designed to start as a blank
slate attached to a powerful learning mechanism (proposed by Turing
in his Mind 1950 article, and many others), will fail. To design
intelligent human-like systems we need to understand the environment
human minds evolved to interact with. (Ulric Neisser wrote something
like this in 1976)
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[] Undecided[]
(33) Some people feel that problems of consciousness and free-will
will either prove obstacles to progress in AI, or will have to be
solved before we can design systems with human-like intelligence.
Such views are mistaken. It's the other way round: in order to solve
old problems about consciousness and free-will we have to make
progress in AI demonstrating what is right and what is wrong in
proposed theories.
Obviously yes[] Maybe[] Dunno[] unlikely[] Obviously not[] Undecided[]
[end]
Thank you for your help.
For more ad-hoc discussions, notes reflections, see
Maintained by
Aaron Sloman
School of Computer Science
The University of Birmingham