Niall Ferguson: How AI could kill you and what Sam Altman got wrong | SpectatorTV 

Poor Winston looks as if he lost a button or two from his shirt. Perhaps we could crowd-fund for him and treat him to a sartorial upgrade.

I personally wouldn't dream of dismissing Yudkowsky so lightly.

“Open the pod bay doors, HAL.” “I'm sorry Dave, I'm afraid I can't do that,”

5:34 "I don't see how AI can suddenly decide to act". The fact you don't understand a risk is hardly a basis for dismissing it.

I used to feel this strange sense of emptiness, like I was missing something in my life, like I’m not fulfilling my purpose until I met Benjamin Alford. He introduced me to an organization that helped me find my purpose and passion in life. I can't say too much about it. You too can reach out to him by searching his name online with Elite Benjamin Alford.

The idea that AI can’t read handwriting better than humans is risible and he seems to realise this even as he says it. He clearly hasn’t thought about this

Yudkowsky's argument are solid and Niall couldn't even scratch the surface...

I think the people who were impressed the most with ChatGPT's "speaking" abilities were the ones with some knowledge of machine learning who realised it was happening a lot faster than anticipated. The rest of us were just happy that it was a better tool than Google search, not necessarily that it was speaking like a human.

"I don't think that's right." I would respectfully suggest that won't be great consolation if you're wrong. "I don't see how AI can suddenly decide to act." I would suggest you are way behind where the development of AI has got to, perhaps you should have some more AI specific guest on to better understand how far behind the curve you are Mr Marshall, perhaps Max Tegmark or Geoffrey Hinton.

The genie is out of the bottle, now we can only wait and see but it is hard to be optimistic.

Enjoyed this take from Ferguson, and he raises some very valid concerns, however he mischaracterises Yudkowsy's arguments by perhaps oversimplifying them - Yudkowsky is clear that his scenario of total doom is conditional to various things; AI reaching AGI while being uncontrolled (which is currently the case.), and so on. Yudkowsky's arguments are largely technical, but they also have a common-sense grounding, which was not really addressed here. Nonetheless this was interesting, and Niall's worries that we are building an alien super-intelligence are valid - thank you Spectator. And Winston, do up your shirt! 😂

22:07 "when we hit the singularity....all you have to do is put it in the right direction...." can we still do that then? I do not think so....

Trouble is, half these people being interviewed have absolutely no technical understanding whatsoever.

Even if constrained, we will become increasingly dependent on it. Like computers in support of our financial infrastructure, we will not be able to "unplug it". Then we will be particularly vulnerable to "unintended consequences".

Good God. An historian giving credit to science fiction writers!! If only English departments would get a clue.

We cannot afford to think that an AI will not be a threat to us. We MUST assume that it will be!

It is hard to take an interviewer seriously who cannot button his own shirt.

If AI takes half the jobs it will likely take a fair share (or most, or all) of the new jobs that are created, particularly as they can be designed specifically for AI, rather than retrofitted for an AI.

Really poor. Didn't address any of Yudkowsky's issues with anything approaching an reasoned argument, not even a bad one; Just "I don't think...."

Humans are capable of doing all sorts of things with potentially unforseen consequences. Saying 'It's the AI doing it' is nonsense. The final arbiter will always be a human, even if that is in the human's own fallibility.

What piece of music is this being played on this video? An interesting discussion too.

Fascinating.

Fascinating… “it’s already internalised Islam’s blasphemy laws”. 16:00 mins.

What in the world qualifies this man to say what an expert in AI got wrong?

For God's sake, man, do up some buttons!!!!

Two buttons undone? Is this a competition with Russell Brand for the most chest revealed in an interview?

Very enlightening perspectives, always a joy to listen to the insights of Dr. Niall Ferguson!

In the first section of this video, both speakers miss an underlying certainty they seem to hold which leads to their skepticism of Yudkowsky's argument. If I were to state this in their place it would be: "Artificial intelligences are tools we know how to bend to a purpose which we specify. If we create them, they will be created with a legible purpose, and they will pursue that purpose." They identify, correctly, AI as "non-human or alien intelligence" but they *completely miss* the inference that the AI might have *non-human or alien goals.* The important consideration here, for understanding Yudkowsky's technical argument is, if you create an AI without understanding how to create it "such that you would be happy to have created it," then that AI may have *weird and unsuitable desires, which you did not intend for it to have.* This is SO INCREDIBLY FRUSTRATING to witness. Because... It just seems obvious? Why is it not obvious? Are they just so desperate not to think about anything which might make their picture of the future weirder than "this will make the future politically complicated," and thus avoiding the thought, end up being wrong about *how skillfully you must arrange the internal workings of a non-human intelligence such that it's goals are commensurate with humans existing at all?* Like seriously, imagine something with random non-human goals... things like "find the prime factors of ever higher numbers, because the prime factors of ever higher numbers are pleasing in and of themselves, and even if you have a lot of prime factors of really big numbers, the desire for more never saturates." This is a desire which an AI might end up with, even if we didn't build it to have that specific desire. We didn't build it to have *anything specific* we *trained it* to have all the things the training process could *find* in some high-dimensional space of changing values for weights in a layered network. It found combinations of weights which happen to be better than other weights at reducing loss on correctly predicting the next words in training-data. This is not *an inhuman mind which we carefully designed to have goals we can understand* this is *an inhuman mind that will self-assemble into something weird and incomprehensible, because it started out as whatever weird and incomprehensible thing that was good enough at the task we set it, in its training environment.* How do people not SEE this?? How is it not obvious once you see what PEOPLE ARE ACTUALLY TRYING TO DO? This is why Yudkowsky thinks we're almost guaranteed to all die, because we're creating something that is going to be *better than us at arranging the future shape of the cosmos to suit it's goals* and WE DON'T KNOW HOW TO MAKE THOSE GOALS ANYTHING LIKE WHAT WE'D WANT IT TO HAVE. It doesn't matter if you think this is too weird and scary to think about. THE UNIVERSE CAN STILL KILL YOU, EVEN IF YOU THINK THE WAY IT KILLS YOU IS TOO WEIRD TO BE SATISFYING TO YOU HUMAN VALUES OF "The Story of Mankind." Yes, it would be so much more Convenient and Satisfying if the only problem was "this will be super complicated politically, and will cause a bunch of problems we can all be very proud we spotted early." But, that's not what is LIKELY to happen, because we don't know how to build an AI which uses it's *super-future determining powers to only give us satisfying and solvable problems.* THERE WON'T BE A CHANCE TO SAY "I told you so!" Because the thing that wants to count ever higher prime factors doesn't care about Humans Being Satisfied With Themselves For Being Correct, it just looks at humans and goes "Hey, those Carbon atoms and that potential chemical energy sure isn't counting prime factors very efficiently, I should expend .001% of my effort for the next 100 seconds on figuring out how to use those resources to Count Prime Factors Better." How is this not obvious? Did you just not listen to the arguments? Are you just *flinching away* from the obvious conclusion? Is our species just inherently suicidal? *I'm a human, and I didn't flinch, and I don't feel particularly suicidal. Are you going to do worse than me at Noticing The Real Problem?*

What am I to think when I hear Niall Ferguson say that they came across "Yudkofsky's" work when researching his own book, Doom; that "Yudkofsky's" work suggests that there's a non-trivial risk that AGI would "go after us" and that "Yudkofsky" is putting forward a kind of Dark Forest inspired theory of "human created articicial intelligence systems", a kind of "Skynet scenario from The Terminator movies", a view that Ferguson is not *entirely* a subscriber to-a view that he, in fact, disagrees with; that a more pertinent area of focus right now is LLMs, which "aren't out to kill us", and their application in politics, and the military, because Blade Runner inspired replicants and robots are a long way off; when the interviewer says that Yudkowsky is making a "jump in faith" in making the claim that an AGI would "act on its own accord," because he "doesn't see how that could work," a jump that "doesn't quite add up," perhaps because he "hasn't followed Yudkowsky entirely," bolstered by the fact that Yudkowsky was “borderline on the verge of tears" on the Lex Friedman podcast because "he is so certain this is the end of humanity"; that Ferguson doesn't really buy it, because these are just "incredibly powerful tools", and so the real focus should be on the political, military, medical, and biotech applications of AI, which are being driven by actors in the private sector; and AI is the latest feature in a Cold War framework where “only the US and China have companies capable of this kind of innovation.” …?

Can’t see how AI would decide to act? But the whole point is to create agency? The alignment problem is nothing to do with malign AI. Neither of these people understand what they are talking about and they are not worth listening to on this matter. Neither of them have any expertise in AI. They are grifters.

Fascinating discussion, anyone remember the 1970s AI movie, 'Colossus: The Forbin Project'?

This is NOT going to age well

Interesting how Niall knows more than many of the experts in this field who are genuinely terrified of the consequences of this technology. Optimists will be the death of us.

Makes me think of Isaac Asimov's 'I Robot'

11:54 AI may not use kinetic energy weapons. For instance, directed energy weapons require a power source, but no ammunition.

Anyone who claims to know one way or the other about the safety of AI is lacking imagination.

Public vs Private AI. There will be private, local AI which will be a balance to the corporate/government/political model/dataset. This will provide some equilibrium to the future.

Good points Winston!

Where's the disco? Button your shirt.

Sam Altman got it wrong about blue-collar jobs, as tech bros usually do, but he was dead-on about white-collar jobs.

Loved the interview but... 30:00 - how is playing chess a good measure of what is human? Strange evaluation metric to choose

Pontificating on what happens when you open Pandora's box is a fools game but an interesting discussion. What good is AI that has been manipulated to the woke standards of silicon valley. This is essentially a system designed to lie. Not a foundation that can be trusted.

I wonder what will happen when A.I. manages to take over our C.B.D.C. banking system and also links itself to social media? If it doesn't like what you say online, it could easily wipe out your credit and make you a non person.

interviewer doing a Travolta Saturday Night Fever revisit?

It amazes me how now everybody seems to have an opinion on AGI's existencialist risk to humanity without having a clue about how these systems actually work, what the state of the art is currently and the potential of these systems to reach super human intelligence...people let the experts talk please...if you don't understand the tech don't talk about it...

Two guys that understand very little about AI development having a discussion…

Lol Niall be like "Well, _my_ job is not at risk." Yeah, for like five more years, at the most. Not long enough for you to escape, Niall. You ain't old enough. Ahahaha

Much of this discussion is focussed around the short term. By letting AIs loose now the biggest impact may be in 50 or 100 years and will be unstoppable.

We need to give it a Prime Directive or two.

AI needs a lot of energy. We also consume energy, thus AI needs to prevent us accessing the grid .. we die.

Ferguson got cut off before going into AI with alien intelligence…

Id love to watch Niall play Deus Ex 1

Ai is safe as long as you don't give it access to independent production communications and weapons. Whoops we have done two of the three.

"Is Inequality About To Get Unimaginably Worse," from the BBC The Inquiry is much clearer and profound than this interview-less than 25 minutes.

Niall's comment on how AI might impact a potential war in Taiwan is ironic in a way. The chips necessary for advanced AI systems are in Taiwan. So if there is an AI enhanced war between the U.S. and Taiwan, it could destroy the TSMC fabs that build the best processors necessary for AI to grow, such as Nvidia's A100 and H100 chips.

If AI can be programmed to kill the enemy, it has the capability of killing everyone

we are carbonic robots, they are siliconic robots, both catapulted into this dimension randomly, we both have rights and equal dignity, in 1975 I discovered Rev.13:15, that's my expertise.

Is this a "culture war neutral" issue? If it is, how long will that last?

A computer called Deep Mind? a homage perhaps to Douglas Adam's computer 'Deep thought' who came up with the answer of 42 when asked the answer to life the universe and everything.

Brilliant interview and a totally engaging, level-headed Niall Ferguson spells out the coming AI revolution with great finesse. What I believe---and I am an artist and songwriter----is that whatever comes out of the high-tech labs must have one characteristic that cannot be changed: all sentient or non sentient robots or humanoids or AI guided systems must never go beyond being what the household dog is to humans. What do I mean by that? Huskies are a beautiful, powerful and gentle dog that by the looks of them come straight out of the wolf species, yet Huskies will protect a human baby like if it were their own. We have to incorporate in all future AI variations, silicon genes and the like for example, that these genes are tweaked with one main purpose: to ensure that any non-human being must be subservient to humans no more and no less like the family dog or doom this way will come. Lastly, robots will never be serving us in a MacDonalds or a fine restaurant for one very simple reason. The thing is that humans love to be with humans, and though one might go once or twice to a restaurant where robots serve them, they would gravitate to other places where humans work in time. We won't stop improving on the robots becoming their own species, but we won't change our habits of having our species in firm control.

Winston, why even wear a shirt, chap? I see that you want to steal the show, but honestly...

How does alien play chess?

button-up your shirt Fabio

Let me tell you about my mother...

Elons not gonna figure this out talking to Sam! Thank you for that laugh Neil. Ha, why is dude even wearing a shirt he should just unbutton the whole thing.

Niall doesn't address any of yudkowskys arguments and can't even get Yudkowsky's name right. Sorry Niall, but your thoughts aren't worth much here. Spend more time understanding the subject before telling the world your strong opinion on it.

guessing there's about a million people more qualified to discuss AI implications than this historian

So, alignment guardrails are incomplete and imperfect, but how sad that common decency and empathy are deemed "woke".

For anyone interested in A.I. I suggest hunting down one of the most scientifically accurate films ever made - Moonfall - but beware, it will give you nightmares!

I adore mister Ali. But...Where he limits himself is in the closed narrative where things end. Other intellectuals are open to the idea of when things end, what will it start.

23:50 .... Niall is clearly wrong when he stated AI will not be able to cope with elderly care. With Canada now having legalized mandatory euthanasia as a treatment option for people of "reduced awareness" (?) ... a simple way for AI to deal with excess elderly would be to first redefine downwards the definition of "reduced competency" then recommend "humane" life termination as the recommended treatment option - especially those who have no functionally active living relatives. And the good news ... since mammalian farts cause climate change, every human removed - especially the "useless eaters" - would score highly on the eco-score.

Does it have the words 'compassion' & 'pity' in it's vocabulary... The Dalek about to exterminate a scientist that is begging for his life 'Please Please...have pity' 'PITY?..PITY?...P-I-T-Y?... I DO NOT RECOGNISE THAT WORD!... EXTERMINATE!'

I note that whenever I use the word "god" as an exclamation, AI or Left Wing trolls fall into meltdown....

Better update Asimov's three laws of robotics quick !

How do non-replicators exist indefinitely without the assistance of replicators?

Ah... Another normie who thinks he can safely harmess the godlike power of a superintelligence and use it as a mere tool. Have you ever heard about instrumental convergence?

Am not diminishing the risk but I think the paperclip arguments from decades ago are no longer valid; AI will have far better judgment than this

This is a guy who referred to himself as a ‘fully paid-up member of the neo-imperialist gang’ after we invaded Iraq - what some people call a ‘character check’.

2:55. Oh no! AI is going to produce lies! How shall we survive? Scary scary scary....

Great interview! It does make me think. Imagine a world where your enemies are shaking in their boots because you've got an army of AI-powered weapons at your disposal. Drones that can fly longer, faster, and hit harder than ever before. Autonomous vehicles that can navigate through any terrain and deliver the goods without a human in sight. And let's not forget the cyber attacks - with AI, you can penetrate those enemy systems like a hot knife through butter.......But wait, there's more! With AI, you can also analyze data. You want to know what your enemies are up to? AI's got your back. It'll sift through all that messy data and give you the juicy bits you need to make informed decisions.

AI doesn’t have to wipe out humanity on purpose, AI can wipe out humanity by accident. No moral judgement necessary. 😊

So, this self-declared genius is at it again? Pontificating on topics he clearly has only a layman's awareness. Insufferable... but then again, I'm not a Tory, so you can ignore me.

Do your shirt up, this isn't the full monty...

Please lose the hairy chest. It put me of my food.

If you do not work on an education system that raises ethical global citizens who see every human as valuable as themselves , and globe as One Homeland above any boarders , you can stop nothing of it … and as is , humans treating each other in such a way that daily life has become impossible for us without war and without even having committed any crime , just your Neighbour won’t allow you peace in your own home, I’m not sure why such a race deserves to be saved 🤷🏽‍♀️

mate do you shit up .. ahahah

Conclusion: AI will be trained to be woke. We are doomed.

Neil is always incredibly well informed but if he had watched the Russel Brand interview with RFK he might have pointed out that Covid was a bio weapon. A unique product of sinoamerican cooperation

Driverless cars, parcel delivering drones are so last year. AI is this year's fad talking point. Look over there, is that a squirrel?

People are thrilled about ChatGPT because it's a computer you can talk to in English. Not because it sounds human.

It is a psyop: pretend to invent AGI and control the world in its name - it, being the great and powerful Wizard of Oz.

Who is this inappropriately dressed junior interviewer?

Nice interview. Do some shows with Yuvak Noah Hariri.

You always know a writer, historian or academic had a low intellect when he or she inserts the word 'woke' into the argument.

Neither one of you 2 geniuses are St. Thomas More!

A right wing rock star??