Think of Artificial Intelligence and the 1950s and visions of Robbie the Robot or apocalyptical futurescapes are immediately brought to mind. Indeed, if we believe the Sci-fi movies and books of the time then this was an era standing on the precipice grappling with the very fundamentals of humanity and automation.
In truth, if we look beyond this pastiche at the real research into AI carried out during the 1950s, we discover how far less things had actually moved. Instead of an era where scientists were locked in their labs recreating Frankenstein, for most academics it was instead a period of open and deep discussion, not over ethics or ground-breaking advancements, but around the laborious task of mapping the fundamentals of what Artificial Intelligence really meant. To be more precise- what makes something truly intelligent rather than just proficient? Early proto-robots such as Elmer of 1948 or George of 1950 were really just radio controlled or had light sensors built in that they could respond to, but they were not intelligent in any way. In conducting their background research into intellect and response, scientists of the 1950s drew heavily on our shared understanding of the natural form and repeatedly expressed a genuine level of appreciation for the mind-boggling processes and procedures which need to take place in order to simply make us human.
There are quite a few early books which deal with the foundations of AI, but of those I have read so far many are either sensationalist (such as ‘The Robots are Among Us’ by Rolf Strehl) or just too heavy for bedtime reading ( such as ‘The Minds of Robots’ by James Culbertson). One set of books I did find illuminating however are two volumes entitled ‘Mechanisation of Thought Processes’. These were an academic transcription of a four-day long Symposium organised by the National Physics Laboratory (NPL) in England which met in November 1958. Attracting an international cast of top scientist and researchers, as the title suggests the seminar was convened to discuss and help direct the very new field of Artificial Intelligence. Please remember this seminar was merely a decade since the very first computers emerged, 8 years since Alan Turing had articulated his Computer Intelligence Test, only 2 years since the term AI had been christened at a specialist Dartmouth Conference, and also only 2 years since the first artificial intelligence program (the Logic Theorist) had been invented. Naturally this NPL event included contributions by luminaries such as Grace Hopper, Marvin Minsky and Gordon Pask amongst others.
When one looks in more detail at the papers and responses that were discussed at this 1958 conference you see that about half dealt with the intricacies and nuances of language acquisition. They repeatedly drew the conclusion from linguists that due to its highly contextual nature, effective speech recognition would be one of the hardest obstacles to overcome in AI. Indeed history has proved them correct as we still don’t have a really good digital translator in the market place. Several papers looked at the clinical response of nerves in the human body and reminded everyone how little we really understand about the way our brains interpret those signals. As they pointed out we humans also rely on electric pulses to manipulate our body but the bit that remains a mystery is how these are translated once they hit the brain. A few papers looked at the importance of information redundancy in memory function (or the art of being selective with our incoming data), and the rest showcased different new computer languages of the time such as Fortran and the Manchester Autocode. One paper even looked at the role of computers in business data processing. Of course, being a scientific audience, very few speakers touched upon the ethical, psychological or sociological side-effects that intelligent machines might create.
So what is intelligence?
In trying to tackle the central issue of what is actually meant by Artificial Intelligence, the key note speaker Marvin Minsky astutely pointed out that our understanding of ‘intelligence’ usually rests on a sliding scale so that an act which seems difficult and therefore intelligent before one masters it, becomes common-place and not-intelligent afterwards. Citing Alan Turing as an example, Minsky pointed out that certain aspects of AI which would have been above Turing’s benchmark of intelligence in 1950, had already been achieved by even the most average computer in 1958. This was a fascinating observation, and as 60 years have now passed since Minsky’s comment, it makes you wonder how our overall expectations of AI have continued to change since that time? In truth many of our machines today are probably well above the intelligence bar that would have been set back then but naturally the world of Science Fiction keeps us ever hungry for more.
Another fundamental point raised (in passing) at this symposium was the observation that to be truly ‘intelligent’ requires the machine to genuinely think outside the box. Put another way, while all machines are programmed to calculate 1+2=3, this is not actually a sign of intelligence but merely proficiency. Instead, true intelligence is to be observed connecting random truths together such as to compute 1+2=a family. A degree of illogicality is therefore a good sign of intelligence and creative people often prove this to be true. Following this argument, creativity therefore is a fundamental pillar of intelligence as it creates something from seemingly nothing and poses the simple question: if computers and robots are built around the basic rules of logic, can they ever become meaningfully illogical?
Overall the NPL conference was telling in what it omitted as much as what it includes and certainly by the end of the decade AI was still very much a fledgling field, led as much by discussion as action. Multiple research projects and conferences such as this one had set the ball in motion and had shown how hard the road ahead would be, but the world was not exactly overrun by robots by 1960. Actually looking back on robotic progress made during the 1950s you can count the famous robot and mice attempts (such as Squee, Garco, Gygan and Katicabogar) almost on one hand. Most of these were perhaps successes in terms of media attention, hydraulic engineering or even clever programming and use of sensors, but were still falling far short of independent action. But instead of us viewing the field of AI at the time the same way we would observe baby’s first adorable steps, research and publicity into the field made during the 50s had the more immediate effect of influencing the business and computer communities to carefully look at and create more efficient and organic systems for themselves too. Therefore, while more intelligent data processing for example does not sound as sexy as a big shiny humanoid, it probably would have had a far greater impact on real life at the time.
Since the 1950s cybernetics has of course moved on quite considerably and elements of AI are in many everyday items we use. I’ll leave it for another day to look at progress made from the 1960s onwards and the debate around the sociological impact of AI will always rage on, but like the countless sci-fi movies that continue to pour out year on year depicting a world over-run by cold logic-driven robots, am I scared about a future where machines may one day become intelligent? Well no, I’m not. You see if the 1958 NPL Conference was correct and creativity is a pre-requisite sign of true intelligence, then I personally have never feared anyone who shows the slightest sign of creativity, even if they are made of metal!
Photo attribution: Metropolis screenshot by tonechootero ; Robot beating up villains by Sascha Pohflepp; Atomic Robot Man by nefasth ; tabletop assistant robot by Matthew Hurst