14 October 2005

Artificial Intelligence In The Garden Shed

By Rusty Rockets

Have you ever driven yourself crazy by contemplating how it is that you can contemplate contemplating in the first place? While this sort of enquiry might make most people giddy, it's these as yet unanswered questions that go to the very heart of what defines human consciousness. Where lesser mortals fear to tread, however, you can be sure that someone with a pioneering spirit will step in and put it all on the line in the name of science, and in this case it is self-educated and self-proclaimed "digital god" Steve Grand.

Grand contends that starting from the top - human consciousness - is why scientists have more or less failed in their quest at creating Artificial Intelligence (AI). Chess programs, digital dogs and robot vacuum cleaners are all well and good, but they do not represent self-generated thought, as they can only perform the tasks that their programmers have fed them. In short, scientists and engineers can only give the impression of AI, which is no more than an illusion. What Steve Grand believes is that anything that has become intelligent and self-aware has done so by gradual learning; one concept at a time. Enter Lucy, an artificially intelligent android built by Grand that also happens to be one of the most advanced research robots in the world.

Currently, Grand tells us, Lucy is gathering cobwebs in a shed located somewhere in Somerset, England, but we'll return to what Grand calls the "downside of being an independent scientist" later. Grand, by his own admission, has a "lazy and erratic brain" which no doubt contributed to his less than spectacular achievements at school. He eventually found his way into teaching but soon realized that having a shy disposition was not a desirable trait if you were planning on standing in front of a classroom full of children. Fortuitously, the microcomputer came onto the scene at around this time. Grand taught himself computer programming, and then went on to teach computers how to play simple board games. After noodling around with neural networks, evolving organisms, plant morphology and evolved finite state automata, Grand got his first big break. He came up with the concept that was to become Creatures, a smash hit computer game.

Creatures was written while Grand was at a company called Millennium Interactive, but on seeing the potential success of Creatures he and his team set up Cyberlife Technology Ltd. (CTL), which was later to be called Creature Labs Ltd. After four years, Grand and his associates parted company, and Grand went on to set up his current business Cyberlife Research Ltd. Out of his success with Creatures, which won him an O.B.E. (Order of the British Empire), he became fixated on discovering how the mammalian brain works, "out of a desire to make more robust, adaptable, flexible and user-friendly machines."

This is what led to the development of Lucy, an android, or more accurately, an anthropoid orangutan, that currently can execute one action - pointing to a banana. Lucy's limited repertoire has left the media, many scientists and the general public completely under whelmed. But this is hardly surprising, as our expectations have been raised far too high in regard to the capabilities of androids and future technologies. It is a reflection of the current "the future is here today" mentality that people are willing to clamor over one another to witness the latest film featuring androids and other various human-machine fusions, while the real thing is left rusting in a Somerset shed. The problem, according to Grand, is that people just don't get what he has set out to achieve.

Grand realized that to achieve any kind of understanding of mammalian intelligence, let alone human consciousness, we needed to step back a little and start from more humble beginnings. It seems that we have spent plenty of time imagining the future, but less time working out how we are going to get there.

Futurists like Ray Kurzweil have plenty of innovative ideas, but Grand questions whether they are the right ideas. A recent book by Kurzweil, The Singularity Is Near: When Humans Transcend Biology, predicts that that our brains will eventually merge with the very machines that they conceived. This "singularity", claims Kurzweil, means that we will all be living in a human-machine civilization where our experiences shift from real reality to virtual reality and where our intelligence becomes non-biological and trillions of times more powerful. But Grand is less enthused. "It seems to me that people who preach about the union of biology and technology usually do so from the perspective of technology. [People] seem to have it in their heads that amazing, wonderful 'clockwork' is going to come along and teach little old biology a thing or two. Meanwhile, they continue to carry their 17th Century physics mindset with them as if technology will change the way we think, but it won't."

"In my view it's all going to happen the other way round. Biological systems are almost infinitely more advanced and sophisticated than anything mankind has ever made, and it's likely to remain that way for a long time. By the time we are properly able to compete, or rather cooperate, with biology, our technology will be biological - we'll have learned to do things and think about things in a very different way, using materials, processes and concepts appropriate to living systems. Man and machine will undoubtedly merge, not because Man will become more like machines but because machines will become more like Man [or less emotively, more like life]," explains Grand.

Central to the Lucy project is researching how an organism gains and uses knowledge, or more appropriately how the process of acquiring and using data are interconnected attributes. Grand believes that if we are to see any kind of fusion between humans and non-human intelligence then, he says, AI is going to be an integral part of this transformation. "Fifty years ago we started out thinking that thinking itself was mere logic and could easily be automated and streamlined, but anyone who thinks their mind is remotely like a computer has a very impoverished understanding of their own intelligence," says Grand. This is an example of the subtle difference between a machine that does the bidding of a programmer, and a machine that generates its own thoughts and intentions. Grand believes we are machines whose being is greater than the sum of its parts, but we only get to that level through a slow and gradual process of brain and body interaction with the world. Grand spends considerable time focusing on what drives learning and how learned behaviors and actions eventually become learned integral skills that can be performed unconsciously.

"Learning begins with conscious intervention and control," explains Grand, "and only later do these painstaking and painful skills become learned and unconscious. The skill lies in making this transition and it takes a lot of intelligence to observe, practice and self-criticize enough for any action to become natural. A gazelle may be more graceful than a dancer, but gazelles can only pronk and leap - they can't learn to pirouette or perform complex movement sequences. Nor, significantly, can any of today's robots - we simply don't have a clue how it is done. There is a good biological reason why we developed consciousness, and the substrate that makes it possible is critical for understanding the brain - despite the fact that 'the C word' is completely beyond the pale in professional neuroscience circles."

Consider if we had the ability to program our children with knowledge way beyond their years of life experience. Would they be able to utilize the information? That is, is it the data itself, or the experience of using the data, that makes it useful to an organism? Grand's response to such questions clarifies what he is trying to achieve with Lucy. "We already have the ability to program the mind from an early age - it's called parenting! And it's a lot more efficient than we tend to give it credit for. The way the brain stores information is non-trivial - it's not like we've got an MP3 player between our ears that we can simply download new knowledge into. Knowledge is highly distributed across the brain, and every new fact or skill we learn alters the structure of every fact or skill we learned before it."

Grand has learnt much about brain functionality while programming his game, Creatures, which starred some artificial life forms called "norns". Norns, Grand explains, "had relatively simple brains that could learn from experience, but I had a small problem. When they were born they knew nothing at all about the world - they didn't even know they could move around and interact with things. So they just stood there. No experience led to no action, which led to no experience. Obviously, they had to be imbued with some basic instincts, like boredom and curiosity, to give them a nudge in the right direction and start them learning through experiment. But how was I going to program these simple behaviors into their brains? I couldn't wire up individual neurons into the correct patterns because I couldn't know what the correct patterns were - a hundred generations down the line their brains might have evolved to work in a completely different way, and I'd be wiring up the wrong neurons. So in the end I chose to use genetics to 'teach' them. Essentially, genes switched on in the womb that presented the creatures with input signals that happened to mimic real sensory experiences. The same genes then 'forced' them to respond with certain behaviors and rewarded or punished them as appropriate. So I could artificially make them feel bored, for example, and then reward them for moving. The cool thing about this approach was that I didn't ever need to understand how their brains worked - they'd learn the right instincts whatever [happened]."

Grand is using the same basic principles with Lucy as he did with the norns, which is why it has taken so long to teach her to merely point at a banana. As simple as this move may seem to people, it is what sets Lucy apart from pseudo-AI. When Lucy points to a banana she does so because she has learned how to through self generated thought via her neural networks. It's not even just one specific banana, either. Lucy has learned the concept of what a banana is by recognizing the characteristics of a banana using the sensory perceptions available to her. "When real babies are in the womb, their sensory nerves fire in characteristic patterns too, and they kick and cough and so on. These events are vital for the development of our nervous systems and I suspect this is how human beings develop their initial repertoire of instinctive behavior. And it is probably the only way it could be done," said Grand.

Grand provided an example of the kind of intelligence he is attempting to quantify through Lucy at a talk delivered at Applied Knowledge Research & Innovation (AKRI). "Here's a thought experiment; take one super, hyper intelligent, chess computer like Deep Blue, say, and take one domestic rabbit - this is what happens when you ask a rabbit to play chess. They're not very good at it. So on that basis, chess computers are far more intelligent than rabbits, but if you swap the experiment around and try throwing them both into a bucket of water, it strikes me that the one who is really the most intelligent is the one who figures out how not to drown."

What will come of Grand's ideas is yet to be seen, but it's almost certain that he'll maintain his status as an independent scientist. According to Grand, most academics spend far too much time focusing their minds on the wrong things and measuring them by the wrong criteria. "I think academia is unnaturally obsessed with two things: specialization and collaboration. Both have their place, but they do stifle creativity. Specialists are people who've become very good at thinking in one particular way, but if everyone thinks the same way then it's like a search party that's only got one torch between them. Creative ideas almost invariably arise when someone sees a connection between the problem in hand and something quite different that they already understand. I've had more insights into brain function as a result of what I know about electronics, computing, cybernetics and even photography than I ever have from knowledge about neuroscience. And collaboration is supposed to increase this cross-fertilization but actually makes the situation far worse. Ten different people, thinking about ten different aspects of a problem, often in ten different languages, are unlikely to be of one mind about anything," said Grand.

In Grand's book, Growing Up With Lucy, he recalls that since he started the project he's been accused of being an unscientific charlatan who is engaged in a get-rich-quick publicity stunt, a heartless reductionist who wants to demean and demote the human spirit, a misogynist who wants to usurp women's role as the nurturer of new life, a naïve fool who doesn't know any better and an irresponsible megalomaniac with society's downfall firmly in mind. It's little wonder that Grand believes religion and policy driven science are driving us toward a scientific dark age.

"It's hard to believe we allowed ourselves to get here, but I think we do stand on the brink of another dark age. And I think it may get worse before it gets better. Religion and Rationalism are currently at loggerheads, and one of them has to go. Obviously I think Rationalism has to triumph, but paradoxically I think it's science that is largely to blame for this mess. In essence, I think science has focused too heavily on substance to the detriment of form; reductionist analysis to the detriment of synthesis. To the 'man in the street', science seems to reduce us all and explain us away, and religion seems (but isn't) an antidote to this nihilism. To put things right we need to restate the intentions of science, replace the Newtonian worldview with a more modern one, and do a much better job of getting across the sheer awe and respect for life, the universe and everything that a proper scientific understanding can generate."

What makes Grand's story so intriguing is that he has gotten as far as he has on his own initiative, perseverance and, you'd have to concede, a genuine interest and conviction in what he does. If Grand were an android, his current success would be thanks to the sniff of an oily rag. After he became a NESTA Dreamtime Fellow in 2002, Lucy was funded by NESTA for a year. Since then, however, he has had to revert back to scraping together whatever funds he can. What's the father of an anthropoid to do? "I could easily fund Lucy if I made ridiculously hyped claims, or reduced the problem to some tiny little incremental improvement in current AI technique, but that's not what I'm here to do, so unfortunately I have to do it unsupported," concludes Grand. Which seems a shame, given that what could be the key to understanding mammalian intelligence - and perhaps even what constitutes human self-awareness and consciousness - is gathering cobwebs in a garden shed.

Read more about the technology behind Lucy at Steve Grand's website:

Pics courtesy Steve Grand