Posted by Austin Morris on  UTC 2016-08-23 12:53 Updated on UTC 2017-01-03

Well, we've built a robot that looks sort of humany: there is heady type thing at the top with two eyey type things in it, two arms, two legs. We have supplied it with circuits and servos so that it can sort of walk over a smooth surface without falling over – watch the shagpile rug! Too late, back on the bench with it.

We then have to get rid of the person hiding behind the curtain with the remote control. We need an autonomous robot – an intelligent one, even.

Perhaps we have managed to get it to ascend and descend stairs – not the escalator! Too late, back on the bench with it.

Perhaps we have trained it to open and close doors: swing-doors, doors with handles, push-doors, pull-doors, revolving doors, automatic doors, locked doors, unlocked doors – no, not that one! 'This door is alarmed'. Too late, back on the bench with it before security gets here.

Perhaps we have done a lot of programming work so that it can get in and out of cars, but then only certain people's cars (and never, ever accept sweeties). Get on and off the bus, in and out of trains – mind the gap! Too late. Empty the carrier bags of pieces onto the bench.

Now that our robot is more or less mobile in a plodding way we have to get it to walk around towns without getting run over or knocking pedestrians and cyclists out of its way. To do this it has to know all the rules and conventions of walking around towns: no bumping, social distance, no sudden stops. We need to ensure that it can walk in all the other places humans walk: forests, deserts, snow, ice, pavements and roads – no, not down the motorway! Too late. The insurers will sort it all out, back on the bench with it. Did you pick up the left leg from the other carriageway?

Now we have got this far we might expect it to do helpful things such as washing up dishes – no, not the Waterford crystal! Too late, never liked that stuff anyway. Or even, perhaps, load the dishwasher, choose the right program, be aware of the warning lights, add salt in the right hole.

Austin's AI test

Now we come to the final test of this artificial intelligence thingy:

Pop round the shop and get me a pack of Marlboros, would you?

This was a task that in the old days before the law got involved you could entrust to an average ten-year-old. A ten-year-old who could not manage this task would be judged to be dim, 'lacking in intelligence', one might say. A career in the Church or banking would be best. In contrast, a ten-year-old who enquired about the need to return the change displayed a high level of intelligence and was without a doubt destined for great things.

Our robot first of all has to grasp what is meant by 'pop', never mind 'round' (as opposed to 'over', 'out' or 'up'). It needs to know what is meant by 'the shop', as opposed to the hundreds of thousands of shops in the country. It needs to know that this is a shop of the category that is licensed to sell tobacco products, as opposed to a butcher's, baker's or candlestick-maker's shop.

It needs to understand the term 'pack of Marlboros'; that this refers to a brand name of something called cigarettes, and the pack in question, which elderly people may still call a 'packet', usually contains 20 of these. Certainly not the box in the back room containing 2,000.

Our robot needs to understand what is meant by 'get' in this context, especially that it is not expected to beat nice Mr Patel to a pulp because he is trying to prevent it 'getting' a pack of cigarettes. It needs to understand the idea of waiting its turn in a queue. It needs to understand the concept of property rights and to be able to carry out a transaction, request the product and hand over money in an appropriate denomination of an acceptable currency. It needs to know that if it hands over a large note and gets only a few pennies in change Mr Patel may not be as nice as one assumes. No reason to beat him to a pulp however.

It also needs to understand that 'would you?' is not a question, just a polite order.

When our robot returns with a pack of Marlboros and the correct change I shall start to believe in the validity of the term 'artificial intelligence'. When it asks if it can keep the change, then we clearly have arrived.

And the moral of this tale?

We may be able to write algorithms of ever greater refinement to do trivial puzzle-solving and visual pattern-matching, but we are nowhere – absolutely nowhere – near anything that can be remotely considered to be intelligence. We can create machines that can do very specific tasks much better than humans can, but they can't do anything else.

If you think you are going to have an 'intelligent' conversation with a computer in a box on a desk which, among the billions of other things it has not done, has never seen 'Snow White and the Seven Dwarves', a machine to which the phrase 'Elvis lives!' is competely impenetrable and which has never experienced the joys of adolescent heavy petting – well, you can forget it. It couldn't even run a blog – and that requires almost no intelligence at all.

And if you are relaxed about a machine learning all the things it needs in order to become 'intelligent', the joys and disasters of 'Snow White', ditto, ditto, but still keeping it in a box on your desk, this intelligent creature that will never know the joys of heavy petting, ditto, ditto – well, that seems quite a questionable situation to me. Once you have given it the Promethean fire, you may need to throw in a bit of free-will, too. One day your intelligent robot may just inform you that the van is coming to get him and he's moving in with the that slim, hot Airbook from down the road.

The fact is that human intelligence requires a long period of the immersion of the human learning machine in a physical and cultural environment. That process is intensive in early childhood, but never really stops. It is not completely inconceivable that one day we can create a machine that has all the motor skills and learning skills of a human, but it will still need many years of immersion to learn about the physical and cultural environment in which it exists. After ten years it might be able to get a packet of fags for you. But what is the point? We have ten-year-old kids to do that.

Update 30.08.2016

Forget the cricket part of it, just consider this as a data processing task: ball trajectory following and prediction, run, jump, twist, stretch, catch, grasp; a tactical assessment of the situation after the ball has been hit; knowledge of the position and state of every muscle, realtime coordination of opposing muscle pairs; understanding of the physics and the entire spatial context; understanding properties of the environment, surface, a provoked fall; understanding the limiting factors of action. Continuous realtime processing and correction. Oh, and keeping your breakfast down while you are doing it. Note that the ball is finally caught after it has passed Plunkett.

[This video is no longer available]

Those watching are unlikely to have the skills, training or physical condition to perform a feat like this themselves, but probably all of them understand the exceptional quality of the catch because they have usually spent a childhood catching things, if only now and again. Even if you can build and train a robot to catch like this, not having spent a childhood missing catches it will never understand the cultural value of what it does. Oh, and it needs to understand how to get the bus home and how to explain to its mum why its cricket kit is always covered in grass stains.

Update 04.10.2016

Here we go again, in that paper of record the Daily Mail:

Experts predict robots could soon be manning tills to diagnosing disease.

Professor Noel Sharkey, Emeritus Professor of AI and Robotics University of Sheffield, said: ‘The robotics community has certainly been considering the idea that robots will be walking among us, and it’s just a matter of when really.

‘We are likely to see robots integrated into society in the near future as shop assistants, receptionists, doctors, bar tenders and also as carers for our elderly and children.

  • They do man tills at the moment. They are called self-scanning stations.
  • It is just a matter of 'when'. Probably never.
  • 'Shop assistants'. Hmm… There are shelf-stackers who work for next to nothing. There are people who sell fresh foods such as cheese, meat etc. Staff in hardware stores, clothes shops etc. Just trying to think of the shop assistant that will be replaced by a robot. I know, let's call it internet online shopping – but we have that already.
  • 'Receptionists, doctors, bar tenders and also as carers for our elderly and children'. Nonsense. Why, for example, at the most critical language development stage of a child would you put it in the charge of a robot?

The current intellectual and physical level of robots is well under that of a housefly. That of the 'robotics community', too, by the sound of it.

Update 03.01.2017

Andrew Orlowski at The Register is also puzzled that people still fall for the myth of the useless, defective technology branded as 'AI':

The most hyped manifestation of better language processing is chatbots. Chatbots are the new UX, many including Microsoft and Facebook hope. Oren Etzoni at Paul Allen’s Institute predicts it will become a “trillion dollar industry” But he also admits “my 4 YO is far smarter than any AI program I ever met”.

Hmmm, thanks Oren. So what you're saying is that we must now get used to chatting with someone dumber than a four year old, just because they can make software act dumber than a four year old. Bzzt. Next...

You don't think it's because of the money, do you? Perhaps capitalism is doomed to introducing pointless technologies that nobody wants just to keep the money wheels spinning.

0 Comments UTC Loaded:

Input rules for comments: No HTML, no images. Comments can be nested to a depth of eight. Surround a long quotation with curly braces: {blockquote}. Well-formed URLs will be rendered as links automatically. Do not click on links unless you are confident that they are safe. You have been warned!

Name  [max. characters: 24]
Type   into this field then press return:
Comment [max. characters: 4,000]