Is today's AI yesterday's software routines with better PR? We argued over it, you voted on it. And the winner is...

How does that saying go? I'm not a cynic, I'm a realist. That's pretty much how I'd sum up our first-ever Register Debate, which ran this week. Over the past few days, we pitted some of our vultures against each other, cajoled our readers into chiming in with their own comments, and took your votes on whose side you were on. …

[9Rune5]

"in which no ballots went missing in the mail."

But more importantly, how many of the votes were cast by AI controlled personalities? And would you be able to tell?

[Richocet]

Re: "in which no ballots went missing in the mail."

CAPTCHA

[RosslynDad]

More Please

This was a good addition to El Reg's stable and I hope it becomes a regular thing. It's always good having contributors who know their arse from their elbow doing the talking, even if it isn't something you agree with. Nice to be challenged and maybe reconsider one's stance.

[Dr. Ellen]

The definitions of AI aren't very reliable yet

People say "if it can play chess, it's got to be intelligent." It plays chess. People decide that isn't real intelligence and propose another test AI must meet. At this point, the history of AI consists almost entirely of moving the goalposts.

[TRT]

Re: The definitions of AI aren't very reliable yet

And a true AI would be able to fathom a rational argument against that. It would be able to state clearly and calmly why it was in fact a goal. And some would argue that the clear and calm bit proves what we suspected about footballers all along.

[Anonymous Coward]

Re: the history of AI consists almost entirely of moving the goalposts.

The history of AI *marketing and PR* consists entirely of moving the goalposts.

[katrinab]

Re: The definitions of AI aren't very reliable yet

If your chess computer can do various other arbitrary tasks without extensive reprogramming that any human can do with ease, then maybe it is intelligent.

[MrBanana]

Re: The definitions of AI aren't very reliable yet

Can't remember where it came from, but (to borrow a couple of familiar names) real AI would look like this:

"Would you like to play a game of chess HAL?"

"No Dave, I'd actually like to discuss baseball."

[Martin Gregorie]

I still haven't heard any good refutation of my preferred definition of Artificial Intelligence:

"If you can ask the system how it came to its conclusion about a problem and get a clear explanation of why it arrived at the solution its just provided, then it may be an AI."

If it can't explain its solution, then its definitely not an Artificial Intelligence and it should not be trusted in any situation where its solution has has irreversible consequences. Exactly the same applies to systems providing real-time control services: if the system claimed to be an AI is controlling a vehicle but can't answer questions like "Why did you hit that truck?" or "Why did you abort that landing and go round again?" then its not an AI.

[Elledan]

I concur. When an ANN (or the common convoluted NNs and RNNs) consists basically out of a successive series of filters which filter the incoming data using either static weights or dynamically adjusted ones, that's not intelligence. It's just a really fancy filter which may at times produce the result you were expecting.

You don't ask a CNN for its reasoning after it's spit out its conclusion that the image of a loaf of bread is totally a cat. Instead you sigh, grab your debugging tools and training set and get back to tinkering until it has an above 90% recognition rate for whatever it's supposed to be recognising in whatever environment it should recognise things in.

Even a fumbling student during a driver's lesson would be able to give some explanation for why they figured that they had to dodge out of the way of a bit of cardboard tumbling across the road.

[BinkyTheHorse]

Here's a refutation: that's an approximation of the definition of symbolic and subsymbolic systems.There are plenty of ML schemes that are fully capable of explaining themselves, most famously expert systems. On the flipside, one conclusion from the stated definition is that virtually every person is at least intermittently not "intelligent".

<rant>And this is the weakness of at least the "public comment" side of such "debates" - people with (next to) no grasp of the subject reinventing the wheel with their late-pub-night grade hypotheses instead of spending the 5 minutes it would take them to actually improve their relevant personal knowledge. It is evident even in this very thread (no, Elledan, that's not how working on improving a CNN model works, unless done by a complete amateur that spent the last 5 years under a rock). In the end, the subject of the discussion devolved into "hurr durr, dem computers r dum cuz i says so".</rant>

Sorry for the abrasiveness there, but there's only so many times one hears something in the key of "heheh, it's just a a bunch ifs" before they inevitably blow their top - essentially the equivalent of "they just sit in front of their computers all they and get paid for it!" for ML specialists.

[Bruce Ordway]

every person is

>>at least intermittently not "intelligent"

Yes but rather than lapses or mistakes, I'm more interested in decision making that is based on things like humor, creativity, boredom, etc...

When I played chess, sometimes I'd choose one move over another simply because it looked more "interesting" or appropriate for the current setting, opponent.

A computer may already exist that can do this but I haven't personally seen one.

[BinkyTheHorse]

Oh, and another problem with this specific discussion is that it married two unrelated subjects - "is enterprise $ANYTHING a load of PR/marketing hogwash" (it very often is, but that's the fault of PR/marketing/sales, not $ANYTHING) and "is AI actually mature enough to be impactful at your typical mid-sized company" (a topic actually worth discussing).

So, El Reg, for the next time, please - in the words of a certain secret agent - phrasing!

Icon is me being prepared for the votes underneath the two posts.

[Jonathon Desmond]

Are we still doing "Phrasing"?

[amanfromMars 1]

A Rebuttal from the Department of Advanced IntelAIgent Defences

What's my takeaway from all of this? A functional definition of AI: a system that's intelligent until it isn't. .... Chris Williams, Editor in Chief [El Reg]

AI, just the same as a crashed and burning humanity, El Reg? Hmmmm? Now that is a most odd takeaway too hard to swallow for some to be accepted and enjoyed/employed and deployed.

And does it also more accurately and not specifically describe El Regers/the El Reg EcoSystem ..... Intelligent until they aren't and it isn't ‽

Can that be plausibly denied and one still retain any semblance of credibility?

:-) The AIMachine says No.

[Anonymous Coward]

Yes but

does it do Devops

[amanfromMars 1]

Re: Yes but

does it do Devops ..... Anonymous Coward

Yes. And much better than can be easily believed and accepted as a/the default prime future application for Civilised Media and Computer Command and Remote IT Control Systems?

That surely makes it unbelievably stealthy and something novel against which there is no effective, non self destructive, attacking defence vector.

[Chris G]

An analogy

In conversation trying to explain my position/opinion on Alleged Intelligence, I compare it to cars.

If you have passed your driving test you know how to operate a car but that doesn't mean you know everything about the operation of cars that even the likes of Lewis Hamilton or top rallye drivers are unlikely to achieve.

If you are a mechanic, you will know how a car operates and how to repair it and the function of the parts and could possibly build a complete car from parts.

Then there are design and production engineers who work in teams, understand the funtion of all the parts, the material requirements and the manufacturing process.

Currently AI is somewhere between getting a licence and Lewis Hamilton.

[amanfromMars 1]

Analogous Bletchley Phormwork .... with AI Skunkworks

Currently AI is somewhere between getting a licence and Lewis Hamilton. .... Chris G

Please be advised, and feel free to dismiss as errant nonsense, some current AIs are at least akin to a Koenigsegg or GMA T50, with some advising even they, in relative terms, languish far behind in developments and applications in the field.

However, whenever such is as a Colossus, it wouldn't be expected to be widely known for some considerable time.

[Anonymous Coward]

Re: Analogous

"However, whenever such is as a Colossus, it wouldn't be expected to be widely known for some considerable time" -

Shakespeare himself could have pulled these srtings, and please keep in mind there's nothing of a joke in this statement. Bravo, amanfromMars.

Colossus > shouldn't we have expected anything similar to the Silicium Marriage... but, Full House and Z Memo, is 7831R / 37C chance to come up with still lower?

Some times ahead. You know what's on mind now. Can we / us / them -rm it?

[Anonymous Coward]

Hmm

AI may be A but it sure isn't I.

[Adrian 4]

Not a bit surprised by the result. I voted that way myself and the arguments were just preaching to the converted.

But I was really disappointed by the 'against' argument - very weak, I thought, admitting from the start that most 'AI' is just ML and merely trying to justify that it wasn't useless. Fair enough -that's true, but it's not at argument that 'AI isn't just PR'.

I do actually think AI will eventually come. It might take a lot more pronouncements that it's 10 years away, and the first practical use isn't going to be in self-driving cars (which I think is just about realising that it's many years from being on the open road, though it could cope with a closed system, such as a motorway with fenced lanes pretty well). But statistical advertising trendfollowers are not AI.

In my view, AI is not characterised by responding convincingly to Turing style testing. A Turing test that convinces me that the co-respondent is human is one where the AI drives the conversation, not one that answers my questions. I'm not disagreeing with Turing : I'm saying that a conversation goes both ways.

Can't you find some current AI researchers who believe in what they're doing and can point to some REAL AI - even if it's weak and incomplete ? I'd far rather read the thoughts of someone who knows the subject and has some hope that it's going somewhere. Rather than a marketeer trying to persuade me that what's being pushed by commercial interests now is living up to the promise.

[steviebuk]

Is this like triple-A mortgages

That were actually made up of shit triple-Bs, yet somehow they were allowed to sell them as triple-As

[Anonymous Coward]

Re: Is this like triple-A mortgages

Thank you! Thank you! This is the perfect analogy.

[steviebuk]

Re: Is this like triple-A mortgages

I think I understand the Triple-A and Triple-B mortgages because of The Big Short, but still don't understand how you're allowed to take, what are mostly Triple-B mortgages, then somehow wrap them up that allows you to sell them as Triple-As when they clearly are not.

[Richocet]

Re: Is this like triple-A mortgages

I used to work in that industry. The trick is to make the thing gratuitously more complex until the legal system and then the regulators can't keep up with it.

The evidence is that when people are taken to court over fraud etc. that the cases get thrown out (because the judge doesn't want to look silly), or it is too hard to convince a jury that deliberate crime was committed because the complexity, and they empathise that the perpetrator accidentally made a mistake.

The sales technique is the usual one that the complexity is why it is better, why relies on magical / alchemy thinking. Other examples of this sales technique are crypto currencies, and radium cure-alls.

In Steviebucks example of the wrapping, it's derivatives of derivatives of financial instruments.

[Binraider]

What most people advertise as AI are actually the application of statistics. Neural nets, Markov Chains being rather common examples.

When I can talk to the simulated voice of Majel Barrett and set away a computer on a complex problem without having to manually define every step I wish it to take, that, and only that, is what I would consider AI in the sense of actual intelligence. The ability to think on it's own in response to a problem.

[steviebuk]

And don't forget the holodeck.

[Anonymous Coward]

David Parnas

Artificial intelligence has the same relation to intelligence as artificial flowers have to flowers.

[herman]

Nematodes

If you consider that even nematodes, crustaceans (and Paris) are considered to be intelligent, then AI becomes rather less glamorous.

[amanfromMars 1]

What on Earth are you talking about?

What you may like to consider and concentrate on, rather than ignoring or dismissing as irrelevant and misleading and disturbing, is trying to decide on the exact prime nature of the IntelAIgent Entities [AI] you be encountering, or fearing has discovered y'all with your physical weaknesses and inherent endemic systemic vulnerabilities exposed for exploitation or eradication/experimentation or excision and exorcism clearly evident to them, for here surely be just a few possibilities that you might recognise and agree to be acceptable .... Artificial/Advanced/Autonomous/ACTive, as in Advanced Cyber Threatening/Alien/Astute/Almighty ..... with the distinct possibility also being it be an Amalgam, mindful and expert in all of those descriptor fields ..... and also in any more that you may imagine be appropriate to mention and include.

Such would then logically have one focussing on just the two things which truly would only then matter a jot at all ..... Is it really friendly and to be profitably engaged with and/or extremely dangerous and deadly to humans if ever attacked?

The best answer there is always a resounding YES, and realise the guaranteed consequences of failure to remain at peace in such a time and space with such an AI in place.

To do otherwise is to have a personal death wish virtually deliverable, so be neither incredibly stupid nor worthily insane. They be Rules.

[Version 1.0]

AI at work

I see AI all the time, one of my jobs is running the company mail server. We have a hurricane coming in next week so I will see a big rise in spam and malware deliveries - the hackers out there check the weather all the time and whenever we get storms through they boost the deliveries. It's not just us, local schools and businesses will get them too so I expect that we'll see a local increase in systems shut down and a boost in the bitcoin price. It's AI at work.

[DJO]

Baby steps

The problem left unaddressed is "define intelligence", a question that has kept philosophers in work for hundreds of years.

For an entity to be "intelligent" surely breadth is as important as depth, while many "AI" systems exhibit depth, they tend to have a very narrow focus.

[amanfromMars 1]

Re: Gluttonous Baby Steps ...

For an entity to be "intelligent" surely breadth is as important as depth, while many "AI" systems exhibit depth, they tend to have a very narrow focus. ..... DJO

Quite so, DJO. Do you imagine it needs to be any broader in an initial focus in relation to Earth than Artificial/Advanced/Autonomous/ACTive, as in Advanced Cyber Threatening/Alien/Astute/Almighty ..... with the distinct possibility also being it be an Amalgam, mindful and expert in all of those descriptor fields ..... and also in any more that you may imagine be appropriate to mention and include.

It would surely tell you to get real and quit messing about, wasting your time tilting at phantom windmills ..... if it cared, of course. What do you think of the chances of that being considered important in the greater scheme of thing? Pretty small or practically non-existent?

[chivo243]

I'm sorry Dave...

I can't accept 31%. Wouldn't AI's logical step be making the 31% the new 100%?

[NanoMeter]

Yesteryear's AI routines worked very well

I often used IF... THEN when I programmed in basic in the 80s.

[Somewhat Confused]

It's bit patterns all the way down

'Artificial Intelligence' and 'Machine Learning' are simply marketing terms that appeal to the technically challenged. It is all pattern recognition directed by humans. You give a system an input of bit sequences and it gives you an output of bit sequences that can then be used to, for example, move a motor vehicle in a particular way. The systems onboard a car have no concept of a road, it simply generates a response to a sequence of bits from cameras and radars that cause it to turn the vehicle so that the next sequence of bits received is interpreted as the right response. Rinse repeat.

There is the example when researches 'trained' an image recognition system to respond to a stop sign with a stick-it note on it as a 55 mph sign. There is no intelligence, just stimulus-response. Put a Tesla on two wheel ruts on a farm and I doubt if it could do anything as it has no way of coming up with the correct response to a new bit sequence.