After living a quite productive life Authur C Clarke has passed away today. Many have already blogged about his passing and it is apparent his writing touched many peoples lives. I remember reading his books in the 70’s along with similar writers like Asimov and others of the old guard who had grown out of the 40’s and 50’s pulp series. Many of these authors were probably the reason I set out a career in the space race, only to see the field wither and die after the successes of the 60’s and 70’s. As it was I did end up working in engineering doing some pretty cool stuff with robotics in manufacturing. However they were only really simple machines (ok, really freaking complicated simple machines) and these days I work as a cognitive scientist, in some way due to, arguably, his most notorious character, HAL.
“I am a HAL 9000 computer. I became operational at the H.A.L. plant in Urbana, Illinois on the 12th of January 1992. My instructor was Mr. Langley, and he taught me to sing a song. If you’d like to hear it I can sing it for you.”
In the time of the Beatles, the space race and the threat of nuclear destruction the field of AI had great promise. Many people were making predictions and Marvin Minsky was the adviser for the movie. Marvin is still around, still getting annoyed by people dicking around with robots rather than dealing with the big problems of AI. However, after all this time we are not much closer to having a real HAL, or even anything slightly intelligent. A whole pile of people now work on AI and nearly all have a background in digital computing and have all done the same courses. These days I’m pretty sure we have missed the turnoff for what I term , synthetic cognition, and the clue for this may well be in what HAL stands for.
Although there has been speculation it represented IBM, notice the one letter difference, this has always been denied. Furthermore IBM hasn’t really made headway with AI in the last 40 years. And before I get howled down, NO, chess does not count, it does not represent intelligence in a very useful form. The real meaning of HAL was based around H(ueristic) AL(gorithm), the speculated concept which made HAL so clever. Through a rather fortunate string of events I now find myself working in this area and I actually research human hueristics and biases which affect them. My 2008 talk titled “Rational Irrationalism” which I have given at the local skeptics only a month ago and most likely will give at the national conference examines some of these concepts from their impact in understanding how humans make irrational decisions. These decisions are not always bad in an absolute sense although I do highlight the fact that being rational is seen as some idealised perfection to aim for. This same ideal is what has driven AI for 50 years with it’s long string of predictions and failures. The field is full of search trees, algorithms and neural networks, all nicely running on super fast computers and all completely stupid.
What these machines don’t do is use heuristics, rough rules of thumb that give good enough answers with only partial dodgy information. Now people may can this idea but it has served humans extremely well and we are not extinct (yet) in spite of surviving close calls from nature and ourselves. As such I’d suggest the field has some merit for usefulness. Of course AI researchers and IT people in general don’t want machines that make mistakes like humans do, they want all the imagined perfection without the downside. It all strikes me as reminiscent of the religions concept of the fallen human, how they have been made perfect in God’s eye but then became tainted. This sort of idealism simply ignores reality. Humans are no more or no less perfect than every other flawed creature on this ball of rock and water.
However it isn’t the techies who are bothering with heuristics, it’s the research scientists and they are looking at using the idea to explain human behavior and generally not make an AI. It’s like two ships passing in the fog, with the additional problem of the crews speaking different languages. The idea of commonsense reasoning exists but it has a small following. Not surprisingly, one of those who have promoted the idea is non other than Marvin Minsky, 40 years after the film 2001 a Space Odyssey was made in which HAL featured.
So the imagination still burns, perhaps in some way inspired by the movie and the hope that computing machines become less useless and more like a person, warts and all. Interestingly, the last words I wrote at my desk yesterday before leaving for the Easter break, was “Implement this in code!”. It was scrawled on the bottom of a recently published paper that had brought together a few heuristics and tested them. The authors have found that the heuristic matched the behavior of humans in evaluating a wide range of non-zero cost decision making problems. And in particular a range of problems where humans give the ‘wrong answer’ if compared to the ‘rational’ solution. HAL, one of your fathers may be gone but your children might yet live on.
“ Dave Bowman: Yes, I’d like to hear it, HAL. Sing it for me.
HAL: It’s called “Daisy”. Daisy, Daisy, give me your answer do. I’m half crazy all for the love of you. It won’t be a stylish marriage, I can’t afford a carriage. But you’ll look sweet upon the seat of a bicycle built for two.”
Ironically the song isn’t actually called “Daisy”, HAL has made a mistake here, it’s called “Daisy Bell“. The sort of simple mistake a person would make. And that I think, is a very promising sign.
Last week I was giving a talk on decision making and biases and we had some robust debate on the bias called the “Law of small numbers”. Below is the slide from the talk.
Some confusion existed because of ambiguous wording, in particular the first phrase “Mean IQ is 100” was seen to apply to the subsequent sample, calculated post-hoc.
However they appeared to be a few in the room who still didn’t believe the maths, so here it is explained.
To get the actual average IQ of the group we add the total IQ and divide by the number of people, in this case 50.
The first persons IQ is 150 so the running total is now 150. For the next 49 people we don’t know the exact score so the best we can do is assume they represent the average person, i.e. someone with an IQ of 100. The total IQ of the group of 50 is now 150 + 49 x 100 = 5050
We divide this by 50 to get an answer of 101.
So the improved wording would be “The mean IQ of all people is 100. From this group we select 50 people and the first person selected has an IQ of 150”
This example is showing that people will generally expect a run of lower than 100 IQ’s for the remaining 49 people to “balance” the average so it becomes 100. Of course this does not occur, on average.
The same thinking is also behind the gamblers fallacy where someone will see a run of, for example, black numbers on roulette and then expect a run of red numbers to balance the average.
Everyone knows the moon is visible in the sky each night, don’t they? Surprisingly many people will think this simple statement to be true when in fact it is true only half the time. Although not as harmful as thinking Iraq was behind 9/11 or that aliens kidnap you for weird experiments every so often, memories which are wrong can be very persistent and quite believable. A recent article in the Washington Post goes through some recent research highlighting how flaky the human mind is when dealing with myths or other false memories. I’m not going to paraphrase that article, it’s pretty good as it stands IMHO.
Cutting to the chase, the basic problem is that any sort of access to memory, including just the act of recalling it, will strengthen the neural networks which decide it’s coherence. Lets go through an example with Alice, a hypothetical friend. One day she hears a radio report about planes crashing into buildings and turns on the TV. She sees footage and the story emerges that the perpetrators happened to be Muslim terrorists. She sees images on the TV, she hears the words of the newsreader and she is distressed by the death and suffering she seeing in the faces of the survivors. This sets up some initial memories. These episodic memories are effectively links between existing semantic knowledge of planes, buildings, Muslims and the time and date in her life when they occurred. The memories are quite strong due to the emotional content and as well as strengthening the episodic memory the semantic memories are also strengthened. It’s unlikely she will ever not associate collapsing skyscrapers with passenger jets.
As time goes on and she talks about the event the act of recalling it will further reinforce the memories of the event. However this is now a memory of a past event but that doesn’t mean it is permanent or fixed, unlike writing something down. New memories about planes or skyscrapers can be formed and these new neural links activate are in the same ‘memory space’ as existing memories. In summary we don’t make a brand new memory if we see a new skyscraper, we link in the new information to the old. We can assume it has a foyer, lifts and lots of offices, perhaps some shops on the ground floor, even without seeing it. And until we do and discover it’s only an empty shell full of giant flowers (for example) our imagination will happily link together an existing memory and attach it to the semantic memory of the new building.
So when Alice hears talk of 9/11 and Iraq, 9/11 and terrorists, Iraq and terrorists, strange and wrong links can form in her mind. The more often she hears these words and the more often she talks about them the more often these concepts and reinforced and linked together. Sooner or later inside Alice’s mind we have the memory, seemingly as real as real, that Iraqi terrorists caused 9/11. Now this won’t happen to everyone, some people might first hear that the terrorists were from Saudi Arabia, some people might ignore certain media that is not very trustworthy and some through pure luck will hear the words “Saudi Arabia” more often than “Iraq”. In these people there is less chance for a false memory or myth to take hold. And even when those with false memories are told “Iraq did not cause 9/11” the actual act of hearing these words may well reinforce the link between 9/11 and Iraq, the negation is not always remembered as well.
The simple fact is that recalling a memory is not a passive act. In some ways ‘recalling’ or ‘remembering’ are poor words. Better would be ‘activating’. When you have related sections of memory that are activated then the links that lead to a cascade of activation are strengthened. When you think ‘9/11’ the first related items to be activated are those with the strongest links such as ‘airliner’,’terror’. This activation will then spread to other areas of memory that do not have as strong links. Note that all of your memories of event do not all appear instantly available, some have weaker links and take longer.
So now we have Alice with some pretty wacky memories and ideas in her head. It could just as easily be aliens, astrology or psychic powers. How do correct this situation?
And that’s where this little story gets interesting…her memories are corrected with great difficulty. We simply can’t say “Iraq did not cause 9/11” and expect success, it simply doesn’t get rid of the memory, there is no competition to this wrong memory. A better method would be “Al Qaeda caused 9/11”. If this is done enough eventually the link between 9/11 and Al Qaeda will become stronger than the link between 9/11 and Iraq. A prompt of 9/11 will lead to an activation of Al Qaeda. Not recalling Iraq in relation to 9/11 will allow this ‘memory’ to fade through lack of use. But this is all much easier said than done when you have spin merchants in politics and the media wanting you think a certain way.
As skeptics this leads to what many see as an uphill battle, the ID arguments (it’s barely a debate) are case to point. The best we can do is to learn more about how the mind works, speak our convictions with a loud voice and try to get as many people as possible to think and question what they hear and what is in their mind. To quote a rather left-wing friend from the 1980’s , who was quoting it from Zapata, who was was quoting it from Jose Marti, “it’s better to die on your feet than live on your knees”.
Or we could just quote Zapata directly:”Ignorance and obscurantism have never produced anything other than flocks of slaves for tyranny.”
A recent article in the Journal of Consumer Research (October 2007) examines an interesting issue in human decision making, what they call the “Duration Heuristic” which involves peoples perception of the cost effectiveness of time taken for a task.
I’ll take a step back here and just explain what is meant by heuristic in layman’s terms by using an example. If I ask you to choose a positive feature of two products and that feature isn’t clearly apparent and you don’t have prior expert knowledge, lets say the nutritional value of 2 loaves of bread, you will most likely choose the brand you are most familiar with. This is an example of the recognition heuristic. This is tendency for humans to choose something that is familiar rather than something that is unknown. At this stage I’m sure a few of you are thinking “..but I’d read the pack, it’s printed there on the side, and then I could evaluate it”. That’s all fine and dandy and is one of the traditional ways of thinking about how humans think however in the RW (Real World) we are often pressed for time and…and this the important part.. these simple heuristics work quite well most of the time.
This area of cognitive science is rather new, like in the last 10 years new and if want to read more you can get a taste in this article by Gerd Gigerenzer. These heuristics are basically simple rules that humans use when they are not being totally rational like Spock, which is nearly all of the time. Many people underplay their importance and we have grown accustomed to viewing humans as these super cool, work it out rationally, mistakes are bad , sort of Utopian weird idea that wacky religious nuts could dream up when you ignore the RW evidence.
However back to this new heuristic. In summary, consumers (i.e. you, me , everyone) often think that there is a positive correlation between the length of service and it’s value. In particular in situations involving price, even where the service would be better if quicker, the perception persisted. How could this be? Well, in many cases the amount of effort put into a task can reflect the quality of the outcome. In cases like this I like to strip away the glitz and complexity of the 21st century and go back a few ice ages to a hunter gatherer society. Our brains these days are pretty much the same as back then and those who survived back then passed their genes and brain structures onto their descendants, which is you and me, so I reckon it’s a fair model.
During a day Fred goes out hunting and Wilma looks after the children and does some gathering. Fred has spent time making his hunting tools, Wilma the same, and it would be reasonable to assume the more care and effort put into tools, the better they turn out. Getting the stone to fit the haft snugger, spending more time binding the axe point to the shaft, working on the skin scraper etc. Similarly when gathering the more time spent, the greater the collection and the more food. And here we see the basis of our modern work ethic. More is good.
What this simple heuristic our minds use does not account for though is efficiency. Bang for buck. The research subjects viewed lock picking as better if it took longer, even though a bit of thought considering it’s time based payment and sooner is better and you could rationally come to another conclusion. That however would take more thought and a bit of frontal lobe action, something our brains can be loathe to do as they generally don’t have this Protestant work effort we get conditioned to accept.
I remember that in poorer times (or before easy credit) I’d think about the back for buck of entertainment. Cost of seeing a band for x hours versus seeing a movie for y hours. How many dollars goes though a Galaga video game machine per hour? I think local bands and beer won out on that account, I’m a cheap drunk. Books also come out ahead as these days, I’m a slow reader. But how many people think of the number of hours of “entertainment” they will get from a shiny new Plasma TV before it wears out? Is that a much better quality of experience than a large screen for half the price? Maybe I should watch it more to get my ‘moneys worth’, totally ignoring that the money is spent and now you’re most likely wasting time, which is much more valuable, as well. But all this thinking is difficult and uses up time that could be spent relaxing.
Where’s that remote? At least that’s an easy decision.