[Paleopsych] NS: Introducing the glooper computer

[Premise Checker]

Introducing the glooper computer
http://www.newscientist.com/article.ns?id=mg18524921.000&print=true
      * 26 March 2005
      * Duncan Graham-Rowe

    MOST of us find a shot of caffeine or a brisk walk does the trick. But
    when Andrew Adamatzky feels his brain needs a little extra
    stimulation, he gets a robot to dabble its metal fingers in it.

    Adamatzky is a computer scientist at the University of the West of
    England in Bristol, UK, and his prototype brain is a dish of chemicals
    sitting on a lab bench. Its "thoughts" are waves of ions that form
    spontaneously and diffuse through the mix. And occasionally, when
    things get too sluggish, the brain instructs a robotic hand to dip its
    fingers into the dish and wiggle them about, literally stirring the
    creative juices.

    Designed to do nothing more than mimic the kind of feedback that
    occurs between our own fingers and brains, this experiment is part of
    an ambitious programme to develop chemical-based processors that run
    on ions rather than electrons, and which sit in dishes rather than on
    circuit boards. Adamatzky calls it gooware: hardware you can store in
    a bottle.

    Now, after more than a decade of development, Adamatzky has worked out
    how to make liquid logic gates, building arrays that he believes could
    lead to powerful processors that are infinitely reconfigurable and
    self-healing. Even computing giant IBM has begun to think along
    similar lines they suspect that much the same technology could power a
    new breed of processor chips.

    But that's not to say that chemical computers will replace
    conventional silicon anytime soon. Besides, for now Adamatzky is
    focusing his attention on another goal constructing gooware powerful
    enough to deserve the description "liquid brain". And to help prove
    the concept's potential, Adamatzky is building the perfect host for
    his liquid brain - a jelly robot. Equipped with artificial eyes and
    synthetic hormones, it might one day sense its surroundings and even
    feel emotions. Welcome to the world of blobotics.

    Chemical computing owes its power to an intriguing but complex piece
    of chemistry called the Belousov-Zhabotinsky or BZ reaction. It
    consists of a repeating cycle of three separate sets of reactions,
    each with its own characteristic mix of ions and molecules (see
    "Making waves"). Once you combine the ingredients, any local
    fluctuation in concentration (or a catalyst) will start the first set
    of reactions. The products of this trigger the second, which starts
    the third, which starts the first again, and so on. The reactants also
    change colour through the sequence, typically from red to blue and
    back again. And because this reaction is self-propagating - reactions
    in one place diffuse outwards and prompt neighbouring regions to start
    reacting - it creates alternating waves of red and blue that diffuse
    outwards from the point where the reaction starts.

Waves in a maze

    Researchers have already found ways to exploit light-catalysed BZ
    reactions to solve problems, such as working out the shortest path
    through a maze. Solving a maze using a conventional computer is
    complex as the program has to examine all possible routes to work out
    which is the shortest. Instead, a team of US researchers used the fact
    that diffusing reactions like a BZ wave always travel by the shortest
    path. They built a physical representation of the maze with plastic
    walls and added the reactants, and when they triggered BZ waves at a
    point, they found that by recording time-lapse images they could work
    out the shortest route from any other point in the maze back to the
    spot where the reaction started.

    On the face of it, BZ-based processing seems to offer significant
    advantages over silicon-based systems. Firstly, the BZ reaction is a
    form of parallel processor: every point on a wave front is like a
    separate calculation - working out how long it takes to get to that
    point in the maze, for instance. Computation occurs as the wave
    spreads or interacts with the walls of the container, and the results
    can be read out in parallel by simply recording the pattern of waves
    created. In theory, a BZ reaction could solve a class of difficult
    problems that have large numbers of possible solutions, such as the
    so-called travelling salesman problem - computing the shortest loop
    route between several cities. These tasks, known as NP-complete
    problems, are extremely time-consuming for conventional computers.

    Unfortunately, chemical computers also have one major drawback: you
    need to translate your problem into a physical representation, such as
    a maze, add the reactants and let the waves diffuse through. Simply
    designing and constructing a maze to solve a complex problem could
    take months. To most researchers it seemed like a dead end.

    But in the mid-1990s Adamatzky began to suspect that the BZ reaction
    might have potential for computing. In 1996 he met Ben De Lacy
    Costello, a chemist at the University of the West of England, and the
    pair set out to try something rather ambitious: build their own
    chemical-based processor. By 1999 they had teamed up with Nicolas
    Rambidi, a physicist at Moscow State University in Russia, and proved
    the concept by making a robot controlled by little more than a dish of
    chemicals (see Diagram).

    Spurred on, they created a menagerie of strange mechanical-chemical
    hybrids. One used robotic fingers and a BZ-based "brain" to mimic
    interactions between human hands and brains - the BZ reaction
    controlled the fingers, while the fingers themselves were tipped with
    catalysts that could stimulate the BZ reaction. Another was a bot with
    two BZ-based brains that could navigate through a furniture-filled
    room. One brain guided the robot towards its destination while the
    other steered round obstacles en route.

    Although the chemical processors worked well enough for relatively
    simple tasks like these, Adamatzky quickly realised that for more
    complex processing he would have to find a way to build the chemical
    equivalent 2333333333of a programmable computer. And for that he
    needed logic gates.

    Logic gates perform the operations on which all conventional
    processors depend. A NOT gate, for example, turns a digital 0 into a
    digital 1, and vice versa. An OR gate outputs a 1 as long as at least
    one of its two input numbers is a 1. And if you can build OR and NOT
    gates, then provided you can link them together, it should be possible
    to construct all other kinds of logic circuit.

    Other researchers had already attempted this. They built circuits with
    physical channels to carry the BZ waves - the equivalent of wires -
    and junctions between channels where the waves could interact the
    equivalent of logic gates. But to Adamatzky this represented a return
    to the problems of the original BZ processors. Miniaturisation would
    be difficult, and routing one wire over another would be well nigh
    impossible. "You would simply get a poor imitation of conventional
    computer architecture," he says.

When balls collide

    Then Adamatzky stumbled across some theoretical work by two American
    physicists, Tommaso Toffoli and Edward Fredkin from Boston University
    in Massachusetts. They suggested that you could create a simple form
    of processor using little more than billiard balls. They set up a
    scheme in which each ball represents either a digital 1 or 0.
    Computation occurs when the balls collide, and the exact logical
    operation performed depends on how the balls collide and the direction
    in which they rebound. In other words, collisions could create the
    equivalent of logic gates. Adamatzky began to wonder whether he could
    collide BZ waves to create a chemical processor.

    Conventional BZ waves certainly wouldn't do the trick. Instead of
    moving in a straight line, they radiate out, making it difficult to
    see how individual waves interact. So Adamatzky started to ask
    colleagues and experts in BZ systems whether anyone knew how to create
    the chemical equivalent of a billiard ball.

    In 2002 his efforts finally paid off. He discovered that a team of
    Spanish and American researchers had created a light-sensitive BZ
    mixture containing chemical inhibitors that suppressed the formation
    of the usual BZ wave patterns. With just the right amount of
    stimulation - provided by light - the mixture generated wavefront
    fragments that travelled through the reactor dish in straight lines,
    without spreading out.

    Last year Adamatzky tried it himself: he introduced a BZ mixture into
    a thin layer of gel loaded with silver halide ions. The viscous gel
    slows the diffusion reaction and the halide ions act as chemical
    inhibitors. Instead of forming circular waves, it spontaneously
    generated wave fragments less than a millimetre long that didn't grow
    or shrink, and which travelled in straight lines. He nicknamed them BZ
    bullets (see Diagram).

    Experiments showed that these bullets seem to behave more like
    quasi-particles than waves, sometimes even bouncing off one another
    like billiard balls, so he realised they really could be used to
    create the logic gates. In experiments he found that when two bullets
    collide at a certain angle, they create a single "output" bullet
    travelling in a specific direction. With only one input, there was no
    output in that direction, creating an AND gate - one that outputs 0
    unless both inputs are 1. Adamatzky also created other gates called
    NOT and XOR. Now he plans to begin combining different gates together
    to create more complex logic circuits.

    Although his work is still at an early stage, Adamatzky is confident
    he can control and organise BZ bullets to form circuits, and he
    already has a good idea of how to construct them. He can create BZ
    bullets by illuminating a light-catalysed BZ reaction, and is now
    working on how to steer the bullets and send the output from the gates
    to specific points such as sensors. He hopes to use fixed impurities
    in the gel layer to act like mirrors, bouncing the bullets in specific
    directions.

    Since there is no need for wires, you can route multiple signals
    through the same volume by controlling their timing. And you can read
    out the results of a calculation using a high-resolution digital
    camera or sensors mounted around the edge of the dish. "Potentially we
    can pack a very complicated circuit in a very small volume," says
    Adamatzky. And rather than using simple binary logic, it might be
    possible to employ a more complex, multi-valued logic, based on the
    relative sizes of the bullets.

    Adamatzky compares his chemical controllers to conventional parallel
    processors such as neural networks, and believes they can perform any
    function these other systems can. "All algorithms previously
    implemented in 'conventional' parallel processors can be adapted to
    liquid chemical processors," he says. But he admits he has a lot of
    work to do before his chemical computers become useful. They have one
    big limitation too, he says: they are not suited to real-time
    processing because the bullets move at just a few millimetres per
    minute.

    However, Adamatzky and his colleague Tetsuya Asai, now at Hokkaido
    University in Sapporo, Japan, have come up with a possible solution:
    they are making waves in silicon. Asai has created silicon chips that
    generate the solid-state equivalent of BZ waves, and used them to make
    simple logic gates. The key is a form of diode called a p-n-p-n
    junction. When there is a voltage across it, a single "seed" electron
    will trigger the build-up of more and more electrons inside the diode.
    When the charge accumulates to a critical level, the diode "opens",
    releasing a flood of electrons.

    Asai has built a two-dimensional array of these diodes in silicon and
    shown that the electron cascade at one diode triggers electron
    avalanches from neighbouring diodes in turn. The result is a wave that
    sweeps through the array much like a conventional BZ wave, only a
    million times as fast. It is also far easier to get signals in and out
    of a silicon processor, so in the long term this work might produce
    new types of parallel-processing silicon chips, says De Lacy Costello,
    or perhaps even hybrid silicon-chemical systems.

    Could they ever replace conventional silicon chips? Perhaps, if
    researchers can learn how to create waves on a nanoscale, so packing
    the logic gates close together. And there is no reason why you can't
    create the equivalent of chemical waves using individual molecules or
    atoms, says Kenneth Showalter, an expert on BZ systems at West
    Virginia University in Morgantown. "There are reaction-diffusion waves
    on a much smaller scale," he says.

    This idea has already caught the attention of researchers at IBM, who
    are experimenting with a processor that performs simple calculations
    using rows of carbon monoxide molecules on a metal surface. When one
    molecule moves forwards, it knocks into its neighbour and produces a
    cascade effect "very much like dominoes", says Bernie Myerson, IBM's
    chief technologist. By altering the way rows intersect, it is possible
    to create basic logic gates that work in a similar way to Adamatzky's
    colliding bullets. It's still early days, says Myerson, but future
    generations of computers could well be chemical-based.

Building the blob

    Adamatzky is not particularly interested in taking on the chip
    industry, however. He has a far more ambitious plan. He wants to use
    his gooware to create a hugely powerful parallel processor: a liquid
    robot brain in which metal and wire are replaced by a blob of jelly.
    The host material he has in mind is an electroactive gel, a jelly-like
    polymer. Not only can BZ waves travel through electroactive gel
    without being slowed down, but the gel also expands and contracts in
    response to electric fields. The stuff is already used as artificial
    muscles, and researchers have used it to make a starfish that moves in
    response to an electric field (New Scientist, 6 July 2002, p 19).

    The gel also offers another way to create motion. When Osamu Tabata at
    Ritsumeikan University in Kyoto dissolved a BZ mixture in an
    electroactive polymer he found that the polymer swelled and contracted
    in response to the waves of charged ions that diffused through it. And
    when he added 300-micrometre-long hairs to the polymer's surface, he
    found the BZ reaction set them swaying in unison like miniature
    Mexican waves. These hairs could manipulate small objects like cells,
    he suggested.

    Asai and Adamatzky think electroactive polymers are the perfect host
    for their BZ brain, not least because reactants won't spill out if the
    robot makes sudden movements - a problem Adamatzky experienced with
    some of his earliest designs. They plan to copy Tabata's
    BZ-impregnated electroactive polymer, and by giving their blob a hairy
    coat they hope the Mexican waves could propel the blob along, just as
    starfish walk using small "legs". The hairs could even help it sense
    its surroundings, avoid obstacles or find things. And Rambidi has used
    a light-sensitive BZ reaction to create a kind of artificial retina
    that can perform basic image-processing - in particular,
    edge-detection, one of the fundamental abilities of the human retina.

    Gooey on the inside yet rigid enough to keep its shape, their robot
    would be a double for the creeping, wobbling monster in the 1950s
    B-movie The Blob, a film that Adamatzky admits he watched with great
    interest. Without a rigid skeleton, this robot could squeeze into
    tight spaces or change its shape. "It will be completely flexible,"
    says Adamatzky - an intelligent, shape-changing, crawling blob. And
    almost every component they need is in place. The challenge now,
    Adamatzky says, is bringing these elements together, a task he and his
    colleagues have begun in earnest. They estimate it will take them
    about five years.

    And beyond that? Could his liquid brain ever become sentient?
    Adamatzky believes so. He has even begun work on computer simulations
    of emotional states created by reagents in chemical solutions. So far
    the results are impressive, he says. Insert a set of synthetic
    hormones into a powerful parallel processor and a machine might even
    feel or express emotions, he suggests.

    Peter Bentley, an expert in artificial intelligence at University
    College London, thinks that Adamatzky's plans for a chemical brain
    might be over-ambitious - but not completely crazy. "I'm not sure you
    could get the same sort of complexity within a gel," says. "But
    there's a lot of potential here."

    Even if Adamatzky doesn't succeed, he's likely to uncover new ideas
    that could help create better processors or reveal something about the
    way our brain works. After all, says Showalter, BZ-based chemistry is
    one of the best models we have for the processing that goes on inside
    our heads. "Chemistry seems to be somewhere between electronic
    hardware and living tissue," he says. "That's part of the appeal - it
    is moving closer to biology."

Making waves

    The most common recipe for the Belousov-Zhabotinsky reaction uses
    bromide and bromate ions, malonic acid, and a cerium catalyst that
    also acts as a visual indicator for the reaction. Mix the ingredients
    together and three separate sets of reactions start. First, bromate
    ions oxidise bromide ions, forming bromine:

    BrO[3]^- + Br^- + 2H^+ HBrO[2] + HOBr

    HBrO[2] + Br^- + H^+ 2HOBr

    HOBr + Br^- + H^+ Br[2] + H[2]O

    As the bromide ion concentration drops, the second set of reactions
    kicks in, creating BrO[2] radicals that oxidise the cerium and change
    the mixture's colour from red to blue:

    BrO[3]^- + HBrO[2] + H^+ 2BrO[2] + H[2]O

    BrO[2] + Ce(III) + H^+ HBrO[2] + Ce(IV)

    Then the third set of reactions begins: malonic acid, cerium and
    bromine react to create bromomalonic acid and bromide ions. The cerium
    is reduced, the blue mixture turns red and the cycle begins again:6

    MA + Br[2] BrMA + Br^- + H^+

    Ce(IV) + MA + BrMA Br^- + Ce(III) (other products are formed but this
    reaction set is still being investigated)