Artificial Intelligence IV

 

Artificial Intelligence IV

The neural net approach to AI has proven very productive. To reiterate the general idea is to train the AI with lots of examples curated by a human. Thus for example by giving the AI lots of pictures of dogs and saying what type of dog the AI “leans” and can identify the type of dog on an unknown image. This is thought to be the way the brain works by reinforcing pathways which are correct and slimming down those which are not.

The big problem with this approach is that a lot of training data is needed which must be annotated by a human. The good thing about this approach is that it appears completely scaleable  ie. there seems to be no upper limit to the number of examples which is constrained only by the size of the computer. The practical limit is imposed by the size of the training set and the cost of the human describing that set. Big training sets are better but amount of human effort becomes very large and expensive. Computer hardware costs are falling so fast that computer power is not a limiting factor.

A recent development is the self referencing AI. Suppose a text is taken as input. The AI takes a word and tries to estimate what the next word will be. It can check the next word and cycle around until it gets the right answer learning as it goes.. Language contains a lot of redundancy which makes the next word guess much easier. The huge advantage is that the training set doesn’t need the slow and costly human step.

In the jargon of area the coefficient ( weight )  applied to different calculations is called a parameter. To experimenters surprise models with a large number of parameters showed improvements above simply scale. For example through text analysis an AI could correctly interpret a simple addition when that was expressed as a human might as two plus two rather than symbolic 2+2.

Because much larger models with more parameters are more easily possible it is becoming easier to combine into less specialist tasks. These are known as foundation models. Previously AI’s were useful for specific tasks. An AI trained to identify types of dog would be of little use for anything else although excellent and far better than a human in its particular area.. It wouldn’t exhibit enough intelligence to know if it was presented with the image of a cat.

Adding many more and different parameters makes the AI far more generalist. As the number of parameters increases( and we are talking billions ) so the AI increasingly becomes intelligent in a human sense This seems to be improving faster than just linearly with the number of parameters.

One famous test of computer intelligence is the Turing test, named after computer scientist Alan Turing. This imagines a situation where a human communicates with an unseen device by teletype. If the human cannot see any difference between a computer or a human then the Turing test is passed. Quite how flexible AI’s become remains to be seen..

It seems clear that this point is very near. It perhaps needs to be admitted there have been false dawns before and it could turn out this is another., There are many issues and hurdles along the way. If there are problems with the training data  hideous issues can occur. This seems to hark back to the old computer GIGO joke, garbage in, garbage out.

I’m adding a minor personal note. I usually take my comments on science and technology from the specialist press. I try and keep abreast of current developments both in science generally and in particular areas of technology in which I’m interested. In this case however my information comes from a longer article published in the “Economist”. I’m finding that AI is  not being covered very well in the specialist press. The “Economist” has an honourable tradition of specialist “in depth “ reporting  which is in both a weekly science section and a quarterly review devoted to technology. However in this case the item on AI appears as a stand alone article not within its specific sections and I have not seen it reported upon elsewhere.

A curious footnote is that an AI scientist at Google has apparently claimed that a chatbot on which he was working is sentient. His employer disclaims this and his view is not held by other scientists. His mistake is suggested to be because of the human tendency to anthropomorphise ( attribute human characteristics to )  inanimate inventions such as cartoon characters. Pet owners commonly attribute human like characteristics to their pets.

Artificial Intelligence 111

 

Artificial Intelligence 111

To put a context around this technology the term embraces a fundamental change in computer programming. To date computer programming has been based around a detailed understanding of the process to be computerised and a step by step translation into instructions for the computer. The general word for this is that an algorithm is constructed. This is a fancy way of saying that the problem is precisely understood and a solution can be developed.

An example would be the calculation on the interest paid at 2.5% on a principal of £1500 over 2years and 7 months. This is a precise calculation which can only give one answer and the way of calculating it is also known precisely.

Back when I was a student this was contrasted with heuristics where the precise solution is unknown but human intuition based perhaps on analogy can lead towards a solution. Heuristics can lead to rules, by which are really meant generalisations. Heuristics often doesn’t give one precisely correct answer but rather a series of options depending on the precise nature of the question

An example might be a controller of traffic lights at a multi way junction. If the lights can detect vehicles then one might say release traffic until all vehicles have passed on the major route, Conversely only stop the major route in both directions if a vehicle approaches on a minor route. Then suppose we want the lights to behave differently at rush hour. Different rules might be say let the first 50 vehicles pass on the major route before switching. So far an algorithm could be applied but then suppose some more general conditions were applied such as requirement for the lights to perform in the most optimal manner reducing vehicle delays to a minimum. Then a precise solution would be unlikely and heuristics would be needed ( Incidentally I would like to see this no doubt hideously expensive solution used )

Artificial intelligence is fundamentally different. It is based upon the study of the brain and how it works. Essentially the brain consists of neurons connected to others in what can be thought of as layers. When we learn the pathway from neuron to neuron is strengthened for a right answer and not strengthened if the answer is wrong. Artificial intelligence uses a brain like structure often called a neural net. This network can be many layers deep. As it stands the network knows nothing but is trained by giving examples. Suppose a neural net is asked to say when given a picture whether it is a cat or a dog. Many thousands of pictures are presented and the trainer tells the net whether it is a cat or a dog. If a dog then connections are reinforced in one way, if a cat then reinforced in another.

Subsequently if presented with an unknown image the net can respond either down the dog route or the cat route depending on where the neuron connections have been reinforced. The power of AI is that the network can interpolate, ie observing that the image is ( say ) dog like but not cat like. Generally AI cannot extrapolate very well.

It can be seen there are two fundamental steps. Firstly setting up the neural net and second training it on external data of known type. This process is totally different from trying give instructions to the computer which enables it  to differentiate between cats and dogs.

This a grossly oversimplified description of AI but the principle is always the same. AI needs a lot of training data. This explains why such data is so valuable. An AI system may be able to see patterns in data which a human cannot. On the other hand an AI has no common sense. Even a well trained AI can look at a dog and see something entirely different.

One intriguing way in which an AI system can be used is to unleash it on a system and just tell it when its right or wrong. The AI can slowly learn the system and may even spot hitherto unknown possibilities. AI’s are particularly good at games which have precise rules but give rise to very many possibilities like chess. A good AI can beat a chess master.

One heavily publicised possibility is self driving vehicles. As this is written there are prototypes on roads. Often this is within particular boundaries such as the City of Phoenix in the USA. Such prototypes usually have manual supervision. Tesla cars have an extremely advanced version which has the objective of only using camera “eyes” and capable of driving anywhere. Vehicles have been made available in beta test to the public but are not generally available. The self driving AI continues to be tweaked

It is important to realise that AI’s are not intelligent in the way a human is. Humans bring a lot of general learning to tasks and generally avoid the kind of silly mistakes that an AI can make. For example if a human looks at an animal picture with the name tag “Tiiddles” round its neck then the human can probably guess the animal is a cat without any more information from the picture. This is the kind of intelligence which an AI would find difficult.

Some eminent technologists are concerned about the future. They worry that an AI could invent a better version of itself and that this could escalate into a super intelligence far outstripping humans.

One doesn’t go very far without coming across the “Three laws of robotics” Conceived in science fiction by the eminent writer Isaac Asimov they are sometimes seen as the basis of a “thought experiment” when AI has advanced far enough.

The laws are-

1 A robot may not injure a human being or, through inaction, allow a human being to come to harm.

2 A robot must obey instructions given to it by a human being unless this conflicts with First Law

3 A robot must protect its own existence unless this conflicts with First or Second Law.

Although a fictional device these laws are very widely quoted. There have been various efforts to pronounce in advance ethical principles for robotics. Incidentally it should be noted these are for self mobile robots while all existing examples of AI are static “brains”

An immense amount of money and brain power are being invested in AI and the field is moving fast. After previous false dawns it does seem as though exciting tools are on the verge of development. There has been some over excited speculation of mass unemployment. All previous technological developments tell us this is highly unlikely; more leisure hopefully, but also different jobs made possible by AI tools.

On becomimg eighty

 

On becoming eighty

It is often said that particular dates are just that; dates arbitrarily allocated in the river of time. For all that we give particular dates a great symbolic importance. So it is that I give importance to passing my eightieth birthday.

When I was young eighty was extreme old age. I don’t think I had any relatives who were over eighty. I don’t think that I thought of my life as a whole and if I ever did I certainly never thought about reaching this grand old age. In general in this blog I haven’t written much about myself. I have tended  more to write about subjects which interest me and about which I know something which I hope is of some interest to others. So at this milestone in my life I hope I can be excused for devoting myself to my life.

The first thing to say is that I don’t particularly feel eighty. I am conscious of slowing down and becoming quite a bit physically restricted. My balance is poor and led to a recent incident where I lost my balance, fell and suffered a minor injury. This was while visiting my bookclub co-ordinators house. I struck my head, suffered a minor scalp wound which bled quite copiously as head wounds do. A colleague in the group was alarmed, called emergency services and quite convinced them I was seriously injured.  She was so convincing that an ambulance came from Wolverhampton to Lichfield by which time the bleeding had stopped and, although shaky, I was relatively OK. In fact I was quite embarrassed by all the fuss. I was checked over, my wife collected me, and I felt no particular ill effects.

This incident does reinforce my caution about falling. It was a fall while visiting Camberley about 15 years ago which led to a broken hip and prolonged issues I do not wish to repeat. To add to my fears my cousin, recently deceased ,suffered two bad falls separated by a couple of years which led firstly to her being housebound and then hospitalised before she died. Although well into her nineties this was an unhappy time for her.

Thinking about relatives she was the last surviving cousin on my mothers side. I was the youngest but this is a sobering thought.. My children are now middle aged; it gives me a shock to hear my son at fifty looking  into early retirement and changing direction.

In terms of changing direction I have been persuaded by Annette that our present house is now far too large and that we should move to what was our holiday flat This means moving from a substantial house into a two bedroom flat. The effect is not of moving a quart into a pint pot bur rather into a thimble. I have been a hobbyist and collector ( Annette says a hoarder! ) and in preparation for moving we are brutally slimming down.

The actual day started with my card from Annette being a photo collage from childhood to the present day. One was a photo where I’m with childhood friends and also a rather mysterious older girl ( Rachel Quimby ? ). I think its possible she was baby sitter for my friends who were second cousins Janet and Susan. An unexpected pleasure was a card from my niece, Clare. I’m very pleased that I’m the person in our family to keep in touch with her as she is estranged from her father who has mental health issues.

In the evening I attended our Hopwas book club. To my surprise they knew it was my birthday ( from Facebook ) and had brought a sponge cake which we had with our coffee. It was pleasant to see they were surprised that I was eighty

The day after, on Saturday ,we met up for a high tea with Frances and family. Frances had bought a balloon saying 80, a box of chocs and something in a bottle which was as she put it “ a balloon anchor “. The tea was too much and we got a “doggie bag” of uneaten cake for Ben. Since we lived in our  Oxford house,  just travelling back at weekends, when Frances was a teenager we have a special link from that time  I was working for Castrol at Pangbourne. It is hard to imagine that she is nearing her silver wedding. I’m still slightly puzzled that her daughter Alice has now decided to call herself Jaden on Whatsapp.

I feel very fortunate to have such wonderful children and grandchildren. My daughters are both very capable managers after strong academic backgrounds, both got firsts at university. My son successfully battled cancer as a student but just managed a degree. He has been working remotely as an IT networks specialist for about ten years. He managed a very successful family transition  to North Yorkshire and our move will be about five miles away to our former holiday flat in Whitby.

I would be less than honest if I didn’t admit to doubts about our move. Annette is keen to leave what has become far too large a property.  We came here 35 years ago when our circumstances with the children still at school were totally different. We will be making an entirely new type of life and I’m hoping we have the flexibility and resilience to be successful.