A. Well, here we are, sitting before our fire, toasting bread in the flames and supping our celebratory cups. It’s a time for reminiscing and looking to the future. Let’s reminisce about our discussions and conclusions over the past five chats. We started, if you remember, with a look at algorithms.

C. Yes, and at fuzzy logic. That’s why it was called ″Yes, No, Maybe″ and we concluded that fuzzy logic is actually nested yes/no selections, not a third state.″

D. Ah, yes, the burned / not burned toast decision – like you are making now but not making very well judging from the smell.

C: Oh, oops. We said a human needs to decide on how dry the bread is and how thick the toast so as to give proper instructions to the machine. Clearly, that’s not something I should be trusted to do!

A: so we concluded that even when a machine is using fuzzy logic, it’s not thinking. It’s relying on data it is given or collects via sensors, and acts on that data without further human interaction on that occasion. 

We also concluded that ″Machine Learning″ is not autonomous. It acts in accordance with its instructions within parameters and we know what happens if parameters are not set: it continues to do what it’s been told until someone unplugs it. 

We concluded that machines do not think like humans because they have no intellectual capacity. They cannot feel, because they have no emotional capacity. But they can be programmed to fake it. In this, they are in some ways similar to a psychopath. 

C. I repeat my favourite quotation from Turing: ″it is best to provide the machine with the best sense organs that money can buy and then teach it to understand and speak English. This process could follow the normal teaching of a child. Things would be pointed out and named, etc.″ He concluded that machines cannot think but only act upon what they have been taught. But in other writing he said that he wanted to believe that machines might, in the future, be able to think.

A. A cynic might say that’s similar to the psychological/sociological arguments over nature v nurture.

C. I think that’s a valid parallel but it’s a bit like comparing a car to a horse: they both move people from A to B but the horse has free will, albeit reined in, whereas the car does as it’s told.

B. But who is doing the telling? Clearly you’ve not tried to make a recent Mercedes do as you tell it: it does as the factory instructed it and you have increasingly little ability to over-ride it. If you put your foot down in some models, it’s been reported, the car sits there while multiple systems debate whether to respond, when to respond, in what order to respond and, eventually, how much to respond by which time it is entirely feasible that the car is under a lorry. 

A. (laughing) I’ve seen one of those reports: the chap who hired a sports convertible hated it so much he left it in a hotel car park for a day because he hated it so much. One thing he hated was the sensors that beeped at the grass on the verges of country lanes, all the time, so when there was an actual hazard hiding in the grass, he hit it. The car that cried wolf!

B. We talked about sloppiness in language, a theme we returned to often. The current over-use of ″impact″ as a catch-all for ″effect″ and ″affect″ and when people say exactly the opposite of what they mean, such as ″I could care less.″ Then we looked at languages, dialects and changes including changes that are not universal, even within a single society. 

B. And with that, we were off and running on what turned out to be the principle theme we followed: that machines follow precision in language therefore they are defeated by imprecision in language. And we examined, albeit in passing, one of the biggest debates that no on want to have: as financial services regulators increasingly put pressure onto financial services businesses to use computerised analysis, can a machine form suspicion?

C. We often spoke of Alan Turing but I think A’s Golf Ball analogy should rank with the explanations by the great man.

A. (chuckles) Well, probably not. But it did explain the fundamentals of why programming is, ultimately, unreliable and why. I was particularly fascinated by C’s discussion over the existential crisis for so-called Artificial Intelligence. The point over the variability of data which is collected, where that data is provided by sources that are infinitely variable, is vitally important. Again, ones and zeros operate only on clean data or treat messy data as clean which means mistakes will happen. 

I especially remember you saying ″Compliance is mechanical: do this, OK, Don’t do it you’re in trouble. It’s clerical work. It’s auditable. It’s measurable. So it’s easy to prove failure and to demand penalties. This is what regulators like. But Risk is not mechanical. Yes, there are a number of established warning signs that might indicate elevated risk and therefore require closer attention to be paid to accounts and those who control them but the whole risk system depends on suspicion and suspicion is not auditable or measurable. ″

C. That’s where we come to the nub of the problem, isn’t it? The judgement of computers comes down to that same old binary issue: yes or no. Even if a computer can report ″it’s no but only by this much″ it’s because it’s been programmed to report on failures plus whether that failure is in a zone where failure is, to some degree, permissible.

An edited version of this article was published in the-yuan.com in 2022
 

B. We talked about language, dialect and accent and we talked in some detail. It was you, D, who said ″If everything is based on language, then machines have to work in a common and standardised language. That’s not a computer language, it’s actual words and phrases, vocabulary″ and we all agreed that if that’s the test, then all computerisation based on the spoken or written word is doomed and we had fun demonstrating how and why that is so and how, context is everything but machines frequently don’t understand enough about context.

D. Oh, yes! Have you watched Disney+Hotstar with the subtitles on? They are hilariously bad. I mean, even worse that those on Iqiyi. The reason Disney's are worse is because they are supposedly producing English subtitles from an English source.  So ″that smells fowl″ is unforgivable.

Also, the Disney subtitles demonstrate a serious problem: the predilections of the programmer.  There’s a film, something about magical crab and a failing restaurant, written by Americans for an American audience and as a way of spending an hour and a half with one’s brain idling, it’s pretty good. In it the writers plainly wrote and the actors said ″aeroplanes.″ We know it was deliberate because elsewhere one of the characters said ″airplanes.″ That difference gave us information.  It was information that the subtitles robbed us of because the subtitles said ″airplanes″ for all of them.

I know this is an extension of the long-standing argument between transcription, translation and interpretation but this shows that interpretation is injected to the effect that it destroys the intention of the writer.

A I agree. Since the early 1990s, ish, I have had a couple of simple tests for so-called artificial intelligence and one of those has been voice recognition for, e.g. dictation. It was rubbish then and it’s rubbish now. Frankly, until ″they″ get that right, I can’t trust anything that is, supposedly, interpretative.

D. We mentioned ChatGPT and other tools and ChatGTP, in particular, interests me. I can see there is a future for it but that future isn’t now.  It produces some fascinating results but if you already know your subject, you can see that it’s basically operating as a very, very good search engine.

But it also produces some abject rubbish. There are many examples where people have asked it very simply questions, biology for seven year olds, for example, ″what is the largest fish″ and it failed but it kept coming back with answers that, ultimately, became circular.

B. I hate chatbots almost as much as I hate call centres. They both work from a script and aren’t allowed or competent to deviate from it. But I read that someone has developed a system that will deal with call centres, using their own tricks against them, to negotiate settlements of disputes. I’ve yet to try it but I want to!

D. Do we see a future for so-called Artificial Intelligence?

C. Absolutely but it’s not as bright as the evangelists say, so we don’t need our shades. Look, it’s a great tool, it will increase consistency of analysis but – vital point – only if there is consistency of programming. And it crunches data in quantity at speed that the human mind cannot do. But is it capable of the type of analysis that the human mind can do? No. Will it ever? Maybe. But only if someone works out what the human mind does and how it does it and works out how to program a computer to do it.

A. will computers evolve?

B. We cannot know. What we do know is that computers cannot, in their present state evolve. But it’s more than 20 years since someone came up with the idea of data storage in plants. If computers can become, in some way, organic – or quasi-organic, then logically, they will be able to evolve but it won’t be quick.

I know there are arguments that machine learning will reach the point that a computer will be able to instruct a robot to produce replacement or even enhanced components but, as we said earlier, machine learning remains dependent on programming so it still falls back on the basic principle: does that component work, yes or no? If no, replace it. Is it in stock, yes or no? If no, make one.  It might even go so so far as to perform diagnostics and to improve the component or identify the external cause of failure and act to prevent it. But only if it’s been told what to do or given a framework within which to operate.

A. What worries me is where failure causes harm.

B. Like watch-lists for banks, etc. There are many cases where such watch-lists are collated by machines but the information is not accurate or is out of date. Livelihoods are lost, reputations are tarnished.

D. Autonomous machines worry me. I’m not talking about industrial robots which, so long as no one tampers with them, are magnificent but their consequences of failure are more likely than not only financial.

A. I don’t agree: have you ever tripped over one of those bloody self-driving vacuum cleaners now used in so many public spaces? They are dangerous to humans.

D. That’s true. I was thinking more of autopilots for ships, vehicles and planes.  Autopilot is far from new in planes but it’s always been operated in an environment where the pilot can retake control. It is now proposed to have pilotless planes and cars.

E. Thunderbirds. In the mid-1960s, Jerry Anderson came up with the idea of massive bulk cargo vessels operated without a crew. Everything was fine until something went wrong. Masterly statement of the obvious. Anyway, in order to prevent a major disaster, the only solution was to call in International Rescue.  We are now at that point but we don’t have Thunderbirds. We are, quite simply, unprepared for the failure of unsupervised automatic processes.

A. Well, well. Will you look at that. It’s almost New Year.

Chinese New Year. Just another way that context is everything.

You thought we were approaching the end of December and that this piece had been delayed but no, the phases of the moon dictate a very early Chinese New Year and, shortly, a very early Hari Raya/Eid. And this piece has been prepared with that in mind.

Would machines have been programmed to differentiate? The answer to that lies in the mind of the reader.  Did you fall for our simple ruse?

B. So what is our concluding conclusion?

I think it’s simple. I think it’s a binary position.

Machines should be carefully and comprehensively programmed before they are deployed.

And when they are deployed, their results should come down to this:  A machine may make recommendations but it should not make decisions.

And that’s why it’s simple: there’s a choice of two options.

To trust or not to trust.

On or off.

One or zero.

When it comes to the final decision, it should always be not to trust, off, zero.

Servant not master.

It’s not an ethical, religious or psychological decision. Law has an influence on it because someone has to be liable when harm is caused to a third party. It’s a pragmatic point. Maybe, one day, after many, many more generations have sat around this fire and talked of many things, machines will be able to reach determinations but the question arises, if we are going to talk about law, if the machines will work according to precedent or according to equity.

A. It is interesting that this debate rages amongst humans: should law be codified, set by the law makers, and applied strictly or should it be interpreted by judges?  It’s a communist/socialist view, that of the misnamed ″liberals″ against the conservative/democratic/ laissez faire view.

Computers follow rules very well. People, unless they are subject to tyranny, don’t.   We are individuals, we are – all of us – a little bit rebellious.  So the end result is that computers don’t understand people because people are messy. People cannot be reduced to ones and zeros until, eventually, the final binary state is reached: alive or not alive.

D. And with that, dear reader, we sign off.

Happy New Year, Gong Xi Fa Chai (Mandarin), Gong Hei Faat Choy (Cantonese) and may your fortune multiply like rabbits.

First published in The-Yuan.com, December 2023
 

<a href="https://worldmoneylaunderingreport.com/publications/web/white_papers/im…">Imaginary Friends 1 - Yes, No, Maybe</a>

<a href="https://worldmoneylaunderingreport.com/publications/web/white_papers/im…">Imaginary Friends 2 - Of Cogs and Golfballs</a>

<a href="https://worldmoneylaunderingreport.com/publications/web/white_papers/im…">Imaginary Friends 3: ″Capital Punishment? </a>

<a href="https://worldmoneylaunderingreport.com/publications/web/white_papers/im…; Imaginary Friends 4 - An Uncommon Language</a>

<a href="https://worldmoneylaunderingreport.com/publications/web/white_papers/im…">Imaginary Friends 5 - ″The first thing we do, let's kill all the lawyers"</a>

<a href=https://worldmoneylaunderingreport.com/publications/web/white_papers/im…"> Imaginary Friends: 6. The circles of an imaginary mind…... </a>