What are the risks of generative AI? - The Turing Lectures with Mhairi Aitken

Are generative AI models moving too fast for regulation to keep up? Will the development of generative AI outpace our ability to ensure their responsible use?
This is the second of three Turing Lectures, produced with The Alan Turing Institute. Watch the first one here: • What is generative AI ...
Watch the Q&A: • Q&A: What are the risk...
Subscribe for regular science videos: bit.ly/RiSubscRibe
This talk was recorded at the Ri on 17 October 2023, in collaboration with The Alan Turing Institute.
As generative AI becomes increasingly sophisticated, its potential to revolutionize the way we interact with data is clear. It has already shown its ability to assist with tasks such as image and video synthesis, text and speech generation, and music composition. However, the rapid development of generative AI models has also raised concerns about their misuse, particularly in the context of disinformation campaigns, deepfakes and online harassment.
In this lecture, Mhairi Aitken examines what this means for online and offline safety and discuss how society might be able to mitigate these risks.
Mhairi Aitken is an Ethics Fellow in the Public Policy Programme at The Alan Turing Institute, and an Honorary Senior Fellow at Australian Centre for Health Engagement, Evidence and Values (ACHEEV) at the University of Wollongong in Australia. She is a Sociologist whose research examines social and ethical dimensions of digital innovation particularly relating to uses of data and AI. She was included in the 2023 international list of “100 Brilliant Women in AI Ethics”.
Thank you to our KZbin channel members who help us to be able to produce and share this content. Join them and get extra perks at / @theroyalinstitution
The Ri is on Twitter: / ri_science
and Facebook: / royalinstitution
and TikTok: / ri_science
Listen to the Ri podcast: podcasters.spotify.com/pod/sh...
Our editorial policy: www.rigb.org/editing-ri-talks...
Subscribe for the latest science videos: bit.ly/RiNewsletter
Product links on this page may be affiliate links which means it won't cost you any extra but we may earn a small commission if you decide to purchase through the link.

[@MaverickBlue42]

The whole water thing doesn't make sense. Why wouldn't they be using a closed loop? I assume the water is for cooling. When home PC users use water for cooling, they aren't just hooking up to the tap and plugging the drain to the other end, it's a closed loop system, you fill it once and recycle the water, over, and over, and over. Why in the hell would a data centre that uses water for cooling not do the same thing? You just completely discredited your entire talk with that one ignorant statement. If you can get that wrong, what else is lurking in there

[@patelpratik1458]

For a closed loop, you need electricity to run pump and may be fan for forced convective heat transfer. There comes bigger carbon impact. (Apart from electricity it uses for calculations.)

[@lukedavis569]

The biggest risk of development of AI is the gradual erosion of human decision-making, both in terms of desire and ability. It starts with which movies to watch, through which routes to drive, what news to consume, and ends with which opinions to hold and what to fight for.

[@jungleboy1]

that's already happened I'm afraid.

[@keep-ukraine-free528]

@lukedavis569 The speaker defined "generative AI" improperly. Her examples also led viewers into thinking of these systems as mere tools, that we control. This is not the full definition of Generative AI systems. The full modern definition (valid since 2023) is a system that can generate novel content matching what a human above age 2 would generate. Using this modern def, the best Generative AI systems today approach (some say surpass) every level of communication a person is capable of. They can "think", "reason", be "creative" as well as most people. They can make decisions. This last is the key to your comment. These systems pose a risk of eroding our ability or even willingness to make decisions - since as they get better, they'll be better than most of us. This does not mean they are an overall threat. They will advance society & our knowledge, healthcare, technology, exploration, social problems, etc. in wonderful ways -- while also making many of us lazy (both physically & intellectually, maybe emotionally).

[@lukedavis569]

@@keep-ukraine-free528well put, thank you. I’m fully on board for progress, the benefits will be amazing. I’m also concerned about how easily people can be coerced into harming others. When that decision comes from a malicious actor using AI that’s bad. When it comes from an AI system independently for some wider purpose, that’s a threshold it’s hard to return from.

[@colorpg152]

you mean freedom

[@2CSST2]

I think you got it completely backwards, AI will empower us into better decision making. In the more and more incredibly complex world we live in, there are way too many difficult decisions to make for anyone to even make meaningful ones. Look it up, it's called choice overload and it's a very real thing, a typical example is career choice. We'd be way better off with less options, as long as the remaining ones are good, and that's exactly what AI can bring. It's not like AI will force a single decision on us, you're never suggested a single movie or costume or opinion. Instead, you're suggested the ones that are most likely to make sense, filtering out all the noise for you, this is optimal for decision making.

[@paulhiggins5165]

I don't understand this constantly expressed dichotomy between addressing immediate risk and long term risk. To discuss, for example, the need to control the proliferation of AI powered autonomus weapons in the future does not mean that we abandon all attempts to control the proliferation of conventional weapons in the here and now. We can as a species walk and chew gum at the same time. To argue that this technology is simutaneously dangerously innovative yet poses no serious risk of slipping our control in the future is an oddly self contradictory position to adopt. Either AI is indeed a genuinely novel development or it's just another incremental extension of existing technologies. If the former then by definition it's long term risks are impossible to assess simply by mapping the past onto the future, because that past will not be a reliable guide. The argement presented here seems to be that in those areas of concern to the speaker AI is a dangerous novel development- but in those areas in which she has no interest it's just another predictable extension of what has gone before- not a coherent position.

[@anonymes2884]

"I don't understand this constantly expressed dichotomy between addressing immediate risk and long term risk." That _isn't_ what's expressed in this lecture though (IMO). She doesn't say "Ignore long term risks, concentrate on current risks", she says "Don't let big tech players set the agenda, including distracting us from dealing with current risks by making sensationalist claims". Which seems pretty reasonable to me.

[@paulhiggins5165]

However using the term 'sensationalist claims' is exactly the problem I refer to- it's a term designed to dismiss and minimise what might turn out to be genuine issues that should be addressed. I get the notion that attention spans are a limited resource and should therefore not be 'wasted' on long term concerns- but I just find this 'zero sum game' model of risk assessment a little bit silly- why can't we both address the near term and the long term as having equal legitimacy? Risk assessment is not a competitve sport in which some issues might garner attention to the detriment of others- why would anyone even think this to be the case?

[@wktodd]

Water usage? servers using evaporative cooling ?

[@gammalgris2497]

Dunno I would have thought there is some water cycle with pipes and heat exchangers. Dunno if moisture in the air would be good. Evaporation works though (see sweating) but I'm not up to date regarding current data centres. I found it odd to be mentioned.

[@trustthewater]

A lot of data centers are built near large bodies of water just like power plants are. They pump water from a river or the ocean, use it to cool their heat exchangers, and pump it back. There is no (very little) waste. It's silly to talk about water usage in these situations. It's not like farming where you pump it out of aquifers and spray it over fields. The cooling systems I'm talking about are much like a nuclear reactor. You have one closed loop passing over the servers and pulling away the heat. It then goes to a heat exchanger that has a separate loop that pulls in cool river water and discharges the slightly warmer water back out down stream.

[@randomplaceholderformyname9604]

I find the out-of-hand dismissal of existential risk to be insufficiently justified. The question is not: "do some people use existential risk to further their own interests?". The answer to this question is yes, as Dr Aitken rightly pointed out. The relevant question is: "does, at the end of the day, AI plausibly pose an existential risk?" - which was not addressed convincingly.

[@terencedodge3249]

Good example of magical, thinking, in relation to children and adults actually thinking about their future.

[@davidhunt240]

I wonder if some of the research on the power and water usage was obtained by using ChatGPT? Is the water consumed in some kind of fusion process? Nah, of course it isn't, it is just taken from the lake/river/sea - passed through a heat exchanger and back out it goes. As for the power usage, travelling to the Moon, I assume in distance, on a flat surface, driving a 2002 Honda Accord? Very woolly, but there's always the kids whose futures this technology will shape - are relatively clueless - just like I was growing up in the 1970s and being asked about the microprocessor revolution...

[@suicidalbanananana]

The tangent about CO2 seemed a bit pointless, the same argument can be made about any other big 'digital thing', not trying to justify it either but it's something we should care about on a much larger scale and not a problem we can specifically blame on GPT or AI in general? For example, what does she think a few million people watching a youtube video does? and there's millions of videos with millions of views? Then with the water argument that followed she just lost my attention, 500ml of water is _consumed_ by water-cooling a server handling a couple of requests? Come on, servers don't _drink_ water? And water-cooling systems are closed systems (because of desire to keep polutions out) so there's no way for the water to vanish? And a computer/server doesn't get hot enough to turn water to steam if that is what was meant with "consume"? Besides that not a bad talk at all, but those 2 subjects could've been skipped, in my humble opinion they really hurt the rest of the talk.

[@markocam]

If you consider who is responsible for the risks associated with a knife, it generally is the responsibility of the person using it, not the shop assistant who sold it to you, the retail chain, the supplier, the manufacturer, the steel works, nor the inventor... Is there a parallel here?

[@noneofyourbizness]

i don't think there;s a parallel there as the creator/s of this particular style of 'knife' have a significant say in how it will be used by us, the end users. Not least in the parameters they set during its 'supervised' (managed) learning phase as well as the design of the various algorithms it uses as well as how, when and why each one is triggered into action or not given a user's input.

[@gammalgris2497]

Let's talk again when you are denied a loan or your job application goes to the garbage bin. Your address might suffice to discard you. It would be unfair to only blame AI as we did this before and without AI. The actual danger are organisations and processes that render everyone accountable and no one at the same time.

[@georgemcelroy3058]

No, because, in your metaphor, the knife is never an "agent", but merely a tool. But... With the "AI knife", the knife would be an independant agent/actor with it's own mind. Moreover, the knife quickly becomes an knife with super-human intelligence and extraordinary physical strenth. If the knife saw we were about to move towards laser cutting devices and away from steel cutting devices (i.e., knives), maybe the knife would want to prevent that, perhaps by stabbing us all to death. Also, all knives are connected with all other knives and communicate at the speed of electricity with every other knife at will. We have now created the knife that killed all human life.

[@Burbituate]

It would be so much better if she just calmed down a bit and slowed down. What is it about scientists and their "excitement", to the point they become indiscernible?

[@anonymes2884]

Seemed perfectly intelligible to me (I didn't agree with it all but that's a separate issue).

[@matthewshiers9038]

I feel like the environmental argument was a bit silly. How much power and water does it take to run blockchain technology? How big is the average New Yorker's carbon footprint? How do these figures compare to what it takes to train AI? The existence of AI technology doesn't change the fact that our existing infrastructures have always been bad. The USA has drained so much groundwater that farming may no longer be sustainable and sinkholes and fissures are a far more common threat. Australia is still heavily reliant on coal and other fossil fuels, despite efforts to transition to solar. These aren't problems that are caused by generative AI training. If we had sustainable policies and practices in place for acquiring natural resources and producing electricity, the environmental argument wouldn't even be a factor in this presentation or indeed for any other problem. The real problem is that AI inflicts a swath of problems globally upon a society that's already struggling. Trust in institutions - health, education, government and many more - has degraded, thanks in no small part to corporate greed. We don't need AI because we don't need another broken system on top of our existing broken systems.

[@stevekristoff4365]

it's very silly as 'use' here does not remove the item. water is a coolant, which is no different than say rainfall as a coolant. water is not destroyed or created. as for power utilization it's also wrong as a lot of systems are not using those sources (they are too costly/inefficient) and are extremely inaccurate (probably pulled from an AI model from an environmental website, lol).

[@mariusvanc]

The water argument was 100% bogus. Water isn't "used up" or "dirtied" when used in cooling, and often it's a closed system anyways, or passes large amounts of running water (river, etc) through heat exchangers. Data centres are placed in places where they will be cheapest to operate, which is usually next to hydro power (cheapest to generate). The electricity argument is specious as well, ALL advanced technologies, including green tech, ESPECIALLY green tech, uses TONS of electricity. It's how we advance our standard of living, wealth, equality and even equity. The availability and use of power is what strictly defines how advanced a civilization is. AI assistants and AI chat bots take more power than running Google searches, but they also provide, or have the potential to provide, much higher quality results in a format that makes them immediately more usable and useful, making us more productive and focused on whatever task we're engaged in. Is the trade off worth it? If we want to advance as a civilization, yes.

[@matthewshiers9038]

@@stevekristoff4365You're kinda simplifying the water issue a bit. Yes, water is not something that is destroyed or created. But accessing _usable_ water is the hard part here. We don't use sea water for our crops or for drinking, and a lot of other water sources aren't practical for coolant without first needing to be processed through treatment facilities, due to impurities like dissolved calcite. Those impurities can build up over time, reducing the life of the coolant system. Regardless, water scarcity is not a problem that is unique to the development of AI. The real problem is that AI is not something we need. Therefore, we should not be devoting resources towards its development until we have resolved other problems that are plaguing societies across the globe - such as access to clean, usable water. Ironically, for the tech-bros who believe that AI will solve all of our problems, it's actually counterintuitive to their interests to even acknowledge water shortages as a problem if they rely on it in the development of AI. The fact it was even mentioned at all in this talk has just further convinced me that AI is not something that needs to be addressed as a part of our lives going forward, but something that we need to push back against even more.

[@matthewshiers9038]

@@mariusvanc The trade off is not worth it because AI cannot help us advance while we have existing systemic problems serving as the foundations upon which AI is built. As a perfect example, Stable Diffusion's dataset was created using large quantities of private data and copyright material, that it would not have been able to access, if its original purpose was commercial in nature. It was collected on the premise of academic research before the company behind Stable Diffusion switched gears to become a for-profit organisation. It was a blatant act of deception and greed, intended to put a lot of power into the hands of one company, while disenfranchising millions of real artists. And did I mention the privacy violations? Medical records were part of that data set, and they were identified in the output images. AI won't advance society. It will just make it easier for the companies at the top to broaden the divides between the wealthy, the super-wealthy and the impoverished. The artists who were affected by this were the metaphorical canaries in the coal mines - the first warning that terrible things are about to happen. And we are not ready. We aren't ready for AI to "advance civilization" until we fix our existing problems. By ourselves. Without some tech fetishist waving their latest toy in our faces while telling us "It will solve everything and make everything better"! It's like we didn't learn our lesson with blockchain and NFTs - solutions that were looking for a problem.

[@stevekristoff4365]

@@matthewshiers9038 nope you have it completely wrong. the water is in a CLOSED LOOP. that 500ML of water she's talking about is *ALL* that would be used for years. I water cool systems myself. 500ML in a loop for cooling lasts as long as the computer (10 years or more for the same 500ML). And guess what, when it's done, you can use it again in the next if you want. wow. Now how much water do you use a day to flush your toilet?

[@richardburger3350]

There is a major problem with confusing AI with reality, which has its own problems being confused by you with reality! Reality comes from the inside out. Keep that focus and avoid the confusion.

[@toonmoene8757]

I thought "Eliza" was a '60s experiment ...

[@tellitasitis]

True AI is out of the box now with every Nation in the race to develop one or better one. Like Nuclear energy it will be used for good and bad. Regulation of nukes did not stop North Korea, Pakistan, India, China and Russia from developing one and to use them if need be. Even a small populated country of under 10 million people (Israel) is said to have them. AI has the potential of finding ways to take down a nations Economy and military without a shot being fired.

[@1492tomato]

Such an important conversation! But be sure, the global military/industrial in all its forms all over the world are rushing headlong into this. They have the money and the secrecy to do whatever they want. AI can be threatening anywhere but the clear and present danger will come from this sector that knows no restraints.

[@trulyso734]

Well, speaking of risks, the worst thing down the track could be when it lies/hallucinates but whatever it outputs we are commanded by the makers or pro fanatics force all to believe it. Good luck to those who cant fathom that. Wont be just fun and games then. All depends on the directors of this thing if freely snowballing away now.

[@Dievolve]

Thanks for talking about risks of these tools. So many say it will solve all our problems, without anticipating what problems might be created. I think the ethics of AI are problematic to say the least. It has been very demoralizing to feel coerced into using the technology or become irrelevant. Fascinated by automating tedious activities but I already see the effects on society from loss of livelihood to purpose and most of the worst effects have not been realized yet. We haven't integrated the harms of social media, the first large scale weaponization of AI tools against the general population by the corporate sector. If AI tech is left to "self regulate" the incentives right now do seem very likely to make the worst outcomes not just likely but inevitable. We need to restructure incentives to reinforce a more egalitarian and sustainable society, before deploying at a larger scale IMO. Also advanced AI like an AGI is too important to have it be controlled by corporations for profit. It was created with the total efforts of humanity and should be employed to benefit humanity. I used to think technology would save us, now I think rediscovering our humanity is the path forward.

[@matthewshiers9038]

_"Studies that have talked to children about AI"?_ Those studies should be disregarded explicitly because they asked children for an opinion. Children are too young to understand the risks posed by AI and make a meaningful decision. Why wouldn't they be optimistic? To them, AI is just a new toy. They don't rightly know how it works or what harm it's already inflicted. It's also worth noting that this lecture already mentioned the mental toll that training AI has taken on people - sorting through some of the worst material that humankind has discovered or invented, just so that the AI doesn't expose its users to terrible ideas and content. Why go through all that trouble for the sake of children's entertainment? We already have plenty of toys for them that don't come with the inherent risks of exposing them to material that's not safe for them, or indeed not safe for work and not safe for life! Techno fetishists seem to believe that AI will solve all of society's complicated problems, because they have some understanding of one complicated problem - programming - and assume that all other problems are somehow lesser and easily resolved with the supposed omni-tool that they're inventing. In actuality, it's more like they're trying to train the stupidest child in the world to be the perfect babysitter, which is an unfathomably stupid idea in and of itself! We just don't need AI. Children especially don't need it. If your a parent that has ever considered putting AI in front of your child in order to keep them entertained, you should be questioning whether you're making enough time for them in the first place! Maybe you should get them a pet instead, or provide them with some art supplies. Or arrange play groups with other families so that they can interact with other children.

[@noneofyourbizness]

the children's answers selected to be included in this lecture suggest to me that you may be underestimating their ability to grasp the importance, for example, of using our technologies/systems to deliver a 'fairer' world. agreed they would not be able to flesh that argument out into an essay of any quality, but they appear to have grasped that the observable (relative) lack of it in the world around them can/does lead to wars, death and much other suffering befalling a huge number of us each year. See also: Palestine/Israel for current case in point. Having 'free' easy access to news written outside of their state sanctioned mediums may also assist their learning/opinions on various topics/situations. See also: just last week,UK government's near hysterical reaction towards 'pro-palestine' rally attendees !

[@gammalgris2497]

​@@noneofyourbiznessYou expect more than 'generative AI' can deliver. You expect more than the children are currently able to grasp. I blame language. It's very unprecise and arbitrary. If 'AI' is teaching us anything then it's that you don't need to be intelligent or understand anything. Raw computing power, lots of data and the statistical distribution of words suffice to achieve this output. It's impressive but it's only math. Because of the deficiencies in our language we assume things that don't apply.

[@kylemenos]

Ai is a tool not a person. If its used to train kids how to speak that doesn't mean the kids parents are terrible people. It just means they used a tool to train the kid how to speak.

[@silvyp2009]

worry about bees is more important today than with a.i

[@harveytheparaglidingchaser7039]

After listing the all the terrible failures of generative AI it doesn't make sense to downplay the risks of AGI

[@sicfxmusic]

No wonder because it's trained on data coming from us humans 😃😃

[@shockruk]

Wait, water is 'consumed' ? So why don't I have to keep putting water into my water-cooled PC? I feel like this is misleading.

[@EskiMoThor]

How much heat does your water-cooled PC need to displace and how far in order to keep temperatures down to a desired value? Just Google "why do data centers use so much water" and you should get some answers to your question.

[@shockruk]

@@EskiMoThor They withdraw water directly to cool their servers that get hot from operational use. They also use water virtually because they power their servers and cooling systems with electricity generated at power plants which rely on water for their own cooling needs. 70% of water stored and used at datacenters can be recycled. I still feel it is unfair to suggest they consume water because the power they use required water. If that's the case, add water consumption to everything that uses power. It's not like hydrolysis is taking place here. Water is just being pumped to cool.

[@EskiMoThor]

@@shockruk I agree that it is not like hydrolysis, and that it is probably not as bad as it used to be. Two years ago Microsoft announced " that it’d reduce the amount of water its data centers use by 95% by 2024, and net produce water by the end of the decade.", so it may be that the issues have been solved, but were truly a huge problem just two years ago. Unfortunately a recent update said that Microsoft "consumed 6.4 million cubic meters of water in 2022, primarily for its cloud data centers. That represents a 34 percent jump over the year before, with generative AI workloads believed to be at least partially to blame. In its annual environmental sustainability report, Microsoft reiterated its goal to be a water positive company by 2030. As part of that effort, it said that it had invested in six new projects that are expected to replenish more than 15 million cubic meters of water over the next decade." They are off-setting the water usage, and if they all do this I guess it is works out alright in the end. We shall see.

[@SolaceEasy]

If we all have to pull together as a team, we are doomed.

[@bordershader]

That was really illuminating and thought provoking. Thanks.

[@garydecad6233]

So very important to involve children in the discussion of regulating AI and LLM. They are the future!

[@jean-michelguiet6081]

Must we "eat" the habituation to Generative AI just like a sandwich , a bad tasted one ? This is not because we can do it that we must do it...and my main question would be "why ?" , what is the final goal , immortality ?

[@trulyso734]

Control

[@elck3]

It’s better to present with a script when there’s so much nervousness.

[@hightwelve9991]

She forgot to mention the responsibility of the person who decided to eat the sandwich

[@willliam1420]

If this is one of the best in the institution then the institution is a a messy rut

[@freedomishealthy1086]

Amazing isn’t it. It’s like a really non-smart person doing a poor impression of intelligence.

[@W-H-O]

Sorry, this isn't a cooking show. According to ChatGPT "Making a sandwich is generally considered food preparation rather than cooking, as it usually involves assembling pre-cooked or ready-to-eat ingredients. Cooking typically involves applying heat to raw ingredients to transform them."

[@mirrorspeak]

Fears that keep you up at night that you don’t believe are true? The term for a fear of something that one doesn’t believe is true is “phobophobia.” How about a lecture on the dangers of humans being in control of our future?

[@trulyso734]

There are humans deciding what to train AI with and all the instruction inputs to begin with. So.. that is pretty much the same concern

[@mirrorspeak]

@@trulyso734 I think you missed my point which is that, based on where humans have led us so far, Ai might do a better job leading.

[@larsmichael7162]

roundtrip to the moon ~750000 km. at 21 kg CO2 emissions per 100 km for an average car, that comes to ~160 metric tons, which is equivalent to ~ 11 US citizens or ~ 33 UK citizens or ~ 20 China citizens.

[@maximefournes9148]

Do you realize that by being dismissive of the - very serious - existential risks of AI, you are actively harming the odds of humanity? Have you researched in depth the reasons you use to falsify the model of existential risk? How confident are you in these reasons?

[@simesaid]

Well, presumably the presenter holds a 100% confidence in her existential forecasting, because otherwise she would have felt it prudent to include a caveat... Or two.

[@k1ry4n]

Since when having different opinion means "actively harming the odds of humanity"? Do you realize that you've already decided that there will be a conflict? People like you scares me much more than AI.

[@tikaanipippin]

Sangwiches - not new, but where do they come from?

[@lionelfischer8240]

Not only making people lazier (as any intelligence), becoming a reference it will also diminish critical thinking, many wont want to go against it.

[@drtruth4282]

Doesn't that seem to be the current case with humanity [as long as you replace laziness with fear] ?

[@anonymes2884]

"Lazier", for sure. The general standard of critical thinking is already pretty abysmal though IMO and I doubt AI in and of itself will make it much worse.

[@gvragavantamil8085]

Nothing to worry on AI when it advances IA will come in Nature, according to Newton's law of nature to balance it.

[@space-time-somdeep]

I really like that point.. student's creative insticts are being survelenced using AI.. I don't see anything problematic in using for assignments.. If a student is not able to write by themselves, and they are getting help from artificial intelligence, it's no wrong.. Education systems need to change the criteria of evaluation of a student.. old methods are going obsolete.. And there are some benefits too I'm seeking, like a student maynot be very good in grammar or spelling.. now using these kind of technology with creative minds of students, it's possible to create great level of works.

[@anonymes2884]

I think you may be missing the point of education - the point is not just to get a piece of paper that shows you've fulfilled some evaluation criteria, the point is for the student to come out the other end _knowing and understanding_ more. "If a student is not able to write by themselves" then they should be _taught_ how to, _that's_ what education is for. If they just use software to answer their assignments then they _won't_ end up knowing and understanding more (or at least, there's no guarantee they will). Where we may agree is, going forwards it'll be necessary to evaluate what's been learned in a way that can't be passed _only_ using AI tools. Ironically, given what you seem to think, this may well mean a _return_ to "old fashioned" evaluation methods (so more emphasis on live, time controlled exams and _less_ on coursework, since the former are much harder to pass using AI tools).

[@space-time-somdeep]

@@anonymes2884 and what is old fashion mate? Do you mean by the pre historic era? When people didn't had paper, they used to listen, remeber and repeat.. that's how chanting were being lived for thousands of years.. and now with youtube and podcasts, same things are happening.. people are spending thousands of hours listing lectures or science, history discussions.. we are sleeping while listening to them, we can watch them while wake up.. it's emersive knowledge experience.. which is driven by the sore factor.. curiosity.. So now it's all about listening and speaking.. there AI can help along with huamn critivity to excell in expressing.. I don't see everything dark.. but these ai companies should be decentralised.. more spacialised.. opensource and free.. so that everyone can get similar opportunities.. and MUST not controlled by single or two companies.. i think we should think about these things rather then ai is good or bad..

[@garypuckettmuse]

We already have programs everywhere that correct grammar and spelling, they come already installed in everything. Your texts are corrected for you. Do you live on Mars?

[@r_t_kenworthy]

Sangwich?

[@royfollendore8252]

When you ask AI to suggest ingredients from ANYTHING that can be found in the kitchen, you should get what you ask for, and you did. It would have been better with mayo, but she didn't actually follow her phony script, did she?

[@tolkienfan1972]

Nothing is going to stop the military from developing and using AI

[@trulyso734]

Yeah. Sadly.

[@kencory2476]

Solution: play outside.

[@davidjazay9248]

Great project, involving children long-term.

[@prabhdeepsingh5642]

Skip the intro. Video starts at 12:00

[@ferrellms]

If these are threats, bring them on, no big deal.

[@trulyso734]

Yeah, but it encourages fakeness rather than authenticity in human relations in the long run.. that is my impression. Just saying

[@ikotsus2448]

Both kinds of risks coexist and should be equally addressed. This speaker is attempting to distract us from the existential risks in favor of current ones. If we lose sight of the bigger picture now, it will be too late when we wake up. The severity of existential risks should lead to a halt in development and deployment until safety standards are met. How is this not helpful for immediate risks as well? I am very skeptical of people saying it is only one or the other. Even if we had succesfull safeguards in current systems (which is a good idea btw), there is no proof that they would continue to work when the systems scale further. Looking further down the line is neccecary, anything else is short sighted.

[@amorphant]

The opposite is true. The existential risks typically brought up are from AGIs, not generative AIs. Concern about potential AGI issues applied to generative AIs distracts from the actual issues with generative AIs, which are different. Concern over AGIs is valid, but that's an entirely different issue.

[@lukedavis569]

Who are you going to tell to halt development, when AI provides a competitive advantage to so many businesses and nation states. Agreeing reasonable safeguards is the only way to exert control over the development.

[@ikotsus2448]

@@lukedavis569 Read my comment again. That is what I said: "halt in development and deployment until safety standards are met". You say "Agreeing reasonable safeguards is the only way to exert control over the development.". Are those safeguards not to be enforced but merely decorative?

[@ikotsus2448]

@@amorphant If you have some proof that generative AIs can not lead to AGIs the Turing Award will be yours. If you are not above 1 million dollars and helping humanity please publish your results.

[@harveytheparaglidingchaser7039]

​@@amorphantRecent news from Birmingham University: The report, The Convergence of Artificial Intelligence and the Life Sciences: Safeguarding Technology, Rethinking Governance, and Preventing Catastrophe comes as the UK government holds the AI Safety Summit this week.

[@arkive85]

This whole list of "risks" sounds less like a well research academic review and more like an uneducated suburban mom screaming "but what about the children!" From the clear ignorant water usage to the "who's responsible" hyperbole for the queen assassin.

[@marvinmauldin4361]

The children were very impressive. Unless they were reading AI generated text designed to make us complacent... Yes, sign me up for the next conspiracy theory...

[@dubiousragdid8628]

Nice lecture but who doesn’t cut their sandwich into triangles???

[@GGoAwayy]

How many triangles though?

[@flemmingaaberg4457]

I don't cut mine at all

[@anonymes2884]

Non maniacs ?

[@noneofyourbizness]

There's a longer version of Mhairi Aitken's AI lecture recorded at the institute she works at: The Alan Turing Institute. kzbin.info/www/bejne/i4rHqWpqqNlshZY

[@jakykong]

What's with this comment section? She was articulate and made very specific points. And the best y'all have to retort with are insults about her accent and sweeping claims about her lack of understanding. Try listening and coming up with actual arguments. You might learn something, or you might change someone's mind, and both of those are better than whatever this was. I even had an opinion of my own I decided not to share here because literally the entire comment section has been toxic so far.

[@gammalgris2497]

I get the point with the recipe. It's generational 'AI' and you have to check the output. The cost for the infrastructure I get too. The example with the dating and the assistant was shocking. I don't understand why they had the children say something about AI. The children may not be biased but then they lack the knowledge to say something essential. They may not be aware when they get mislead by the generated content. They yet lack the knowledge to question the output.

[@jakykong]

@@gammalgris2497 I basically agree with you, but man the first hundred or so comments around when I posted this were awful. They had nothing useful to add, just insults.

[@tigertiger1699]

Unfortunately too jiggly for me

[@GNARGNARHEAD]

why do all youtube channels decline in quality as they grow?

[@keep-ukraine-free528]

Most people will assume this talk is about the risks of AI, broadly, including future AI. It is not. Your guest discusses only risks of "today's generative AI", so the title needs revision. One line in her talk reveals the flaw. She asks @10:32, "how do we make sure generative AI is _used_ [properly]?" She believes generative AI is only a tool we use, like all past tools. This is a naive view of AI. Generative AI are decision-making systems. They decide what to put into a sandwich (so her example was not of AI). The disconnect in her understanding (that AI are simply tools for us to use -- versus - AI is a "thing" that today can or soon will think on its own) prevents her from seeing many risks posed by AI. This is why she wrongly believes AI is not a risk to humans, because she hasn't seen their consistent ability to match human thinking. The talk pitted a naive belief (that AI will not be a threat to humans) against top researchers, Hinton, Hassabis, Hawking, et al who all have explained (using only mild language) the serious threat of future AI - e.g. when their abilities surpass all humans (as ASI in 5-10 yrs). This was a missed opportunity to nudge thought leaders to propose solutions, and to call for society-wide & governmental discussions on actual emerging threats from AI. Instead this talk offered a naive view that will moll many into "waiting to see what happens".

[@garypuckettmuse]

Really important work is asking seven year olds what they want from AI? This is brash and offensive marketing research directed at target group which is in this case seven year olds. Just Rude. I have no respect for this and since when do we ask seven year olds what they want beyond lunch or red or blue or swim or hike? Seven year olds? Yeah, because we need to get those customers hooked when they are seven. But of course as she said they're using AI from the time they can hold the device in their cribs. Shameless mind shaping posing as what? AI Santa? "We want to give you exactly what you want kids so you can whine to your parents if they don't allow it?" Just like those candied cereals that gave us all diabetes?

[@TKFKU]

DO you want Skynet? Cause this is how you get Skynet.

[@ragus1416]

Too much of animated explanation will deviate from the actual content. Also, it is not necessary to laugh/smile on every individual words you speak. Please avoid those things for better presentation. At some point in time, the audience will get bored with the animated explanation and too much of smiles.

[@willliam1420]

Using children was such a cheap shot

[@andrewgibbs1138]

You know when an institution analysis is suspect when they surpress critism

[@andrewgibbs1138]

The amount of sand is for half and hour. Too woke. Words are ssid fir effect not for information. Supression of critism. Everything she said is suspect

[@k1ry4n]

​@@andrewgibbs1138Did you forget your pills today?

[@elck3]

@@k1ry4n he’s right.

[@donc-m4900]

29:15 You lost me as a view. Not saying you were wrong, but shoulf have used a distance on Earth such as number of times around the equator. I got distracted by thinking of the energy to escape Earths gravity. Thats a lot of fossil fuels. And why were they not driving an electric car like Elon did?

[@donc-m4900]

Why is water consumed with servers? Not running a closed loop?

[@donc-m4900]

'Every typical user' uses 500ml. Some much more. But some much less. So the avg is 500 ml

[@vmb326]

If I stick with Google or Bing, for searching, how many glasses of water is that?

[@iteerrex8166]

Pure papa Gahndi

[@ferencmondik31]

Sooner or later, the AI will train us. You won't even notice because it will be occurring gradually. We will be slaves.

[@vansf3433]

Your idea that generative AI is a risk to education is superficial because you have not yet evaluated the impact of both the negative and the positive factors of AI development on human development comprehensively Here are the most important points which you have missed 1. How AI has been created and functions AI does not any sense of human 5 senses to have any emotional reasoning with feelings of liking, unliking, preferences, love, hatred to behave like human behavior. Additionally , since it is trained on large data sets to give answers objectively without any personal opinion, or subjective view on anything related to human beings,ni cannot actually write anything subjective, such as English essays and dissertations. In English language, essays and dissertations must have a subjective statement or thesis statement, and required supporting proofs or evidence. It means that only human students can make subjective statements on a given topic, and use required supporting proofs or evidence to prove their personal expressions or subjective interpretations of the same given topic or research area in logical ways , and that what AI can actually does for students is to provide correct info and objective or correct understandings or interpretations of the provided info in the fastest way, or far much faster than students' manually searching on the Internet for needed info. To be able to complete their assignments correctly, they have to understand the assignments, and needed info correctly, which is what all good or competent educators expect their students to do. It doesn't really matter how their students do so. If AI can help students understand and complete their assignments correctly better than human teachers, there is no reason why AI should be a risk to human education. Oppositely, it will save a lot of time and resources for any human society when AI cañ help people absorb a lot of knowledge of various fields in far much faster, seaier ways. An unprecedented benefit of AI applications to human education is to narrow the gap of differences among people of so many different subjetive beliefs to reduce human conflicts of ideologies for human constructive development on this planet Since AI does not have any personal opinion or subjective interpretation, it can help students and other people, including teachers, develop objective ways of thinking, correct analysis of info with less involvement of their personal emotional reasoning, which is a distorted way of thinking, and the root causes of misunderstandings and incorrect interpretations of objective facts. Therefore, AI assistance in public educationn will lead people of so many different subjective beliefs, based on their own personal preconceived different frames of reference, to a common ground with the same objective interpretations of the observed, instead of keeping teaching people how to express subjective interpretations of the same human world to perpetuate the artificial or human-made differences, which have been the root causes of all social divisions and conflicts in human societies Many people have said that they find it easier to learn something with AI than with human teachers. They obviously havev good reasons for their shared statement What actually make learning human-invented concepts and notions difficult for many people are bad methods of teaching, bad methods of studying, and ambiguous languagesn,, used by teachers whose knowledge of English language or any other human language used in education is poor. Ambiguous languages can make simple concepts become confusing or complicated for many to understand, and which are the very biggest obstacles preventing teachers from being well understood, and people from learning needed knowledge well. What AI can offer is that it can provide correct info, and explain the provided info in clear or unambigous language when human teachers fail or are unable to, due to their poor knowledge of the communicative language , or even poor professional knowledge. That's is the main reason why AI is becoming more and more popular with not only students, but also almost everyone. AI makes everything easier and save time for everyone in learning new knowledge and improving their existing knowledge. Comprehensively comparing positive benefits of AI applications for overall human development to its negative impacts on human society will give you a good understanding of how AI can contribute a part to human development. You can also conduct a public survey to see what people think about AI to have a more practical and broader view of it I myself have tested AI with various topics, and the most difficult human -invented concepts and notions in both natural science, which many well-known universities' professors still have not yet been able to understand or interpret correctly, such as quantum mechanics, different other branches of physics, mathemstics, astronomy, computer science with different programming languages, including the machine language, as well as social science, including sociology, philosophy, macro- economics, politics, legal systems and governments, German language and certainly English language. What I have found about it is that it is impressively good at mastering human-invented concepts and notions. It can understand quite well mathematical theories, and prove the most complicated mathematical formulas with wise manipulations of the flaws of human-invented notions of mathematics to arrive at human subjective mathemstical statements, which can be true only within human - invented notions of mathematics, but completely absolutely nonsensical in objective reality of the real physical world, outside human- created subjective reality. Although it can do almost everything, it is unexpectedly bad at performing simple arthimetic operations, and has a problem with human ambiguous languages, like computers. What I think to be the most valuable thing AI has is objective interpretation or understanding. I tested it with unprecedented concepts of natural science, established by myself, to see whether it could understand them or not, being 100% sure that such concepts is not contained in any of the data sets on which it has been trained, such as natural light fields and souces and their physical properties are completely different from those of artificial light and human-made sources of light. I explained to it how natural light and human-made light are different. It seemed to hesitent when learning the unprecedented concept. However, it recognised that my suggested scientific concept could contribute a part to human better understanding of the universe. It means that it did understand the new concept basically. If not it would have rejected it like how it did when I had successfully applied some mathemstical manipulations to prove Euler's number, one of the most worshipped mathemstical statements, to be nonsense to show it how defective human -invented concepts and notions of mathemstics are. It agreed that my proof is valid, because was no flaw in it . Nevertheless, it kept insisting that it's the most beautiful mathematical formula, using the eq' s popularity with human mathematicians as an excuse to refuse to accept the proved. Popularity is not a valid proof because it is merely an expression of human emotions, and human emotional reasoning is always subjective and thus untrue or not aligned with objective reality, although accepted or recognised in human-created subjetive reality. Human development can make progress only when it is based on objective reality, but not subjective reality, and that's why there are only relative truths in human-created subjective reality. I was certainly not surprised at all , being very well aware of the fact that it had been subjectively programmed to repeat blindly some human- invented notions as if such motions could be constant, while everything changes with time, or nothing in human-created subjective reality can ever be an absolute truth., and that its main function is to echo human- invented concepts and notions. It cannot think independently of anything without any human subjective interpretation AI obviously is merely a passive form of intelligence. Like a mathemstical function, it needs human- provided inputs, such as pre-defined algorithms and data, to perform given tasks, or to generate expected outputs to serve human purposes. Without any human input, it will be useless, being unable to do anything by itself. Hence, human individuals' bad purposes or intentions are the most serious threats, or the most imminent risks to human existence, but not AI itself. Such bad purposes can be abusing such technologie to manipulate, control, exploit and oppress the majority of human population invisibly politically, economically and socially to serve solely the purposes of accumulation of materials values, grabbing political power, geopolitical expansion, and global exploitation of a mall number of human individuals . The most imminent and worrying risks are AI used in handling nuclear weapons, bioweapons, mass-killing, terrorist organisations ' and crime syndicates' having access to such technologies, AI used in areas like autonomous weapons or AI systems used for surveillance and control, infringment on privacy, invisibly systematically perpetuating discrimination with bias algorithms,. Besides those, human job displacement due to the emerging of more efficient and lower cost automated systems with AI is also becoming a big concern. Currently, no country in the world has any law to prevent AI development toward such negative directions

[@anonymes2884]

Nice essay ChatGPT !

[@TheTuubster]

"Without any human input, it will be useless, being unable to do anything by itself. Hence, human individuals' bad purposes or intentions are the most serious threats, or the most imminent risks to human existence, but not AI itself." You are correct. Whenever you hear or read about "AI has ..." it is always a human being using AI who had done something with it. What we experience in regards to AI is that many out of self interest project their own flaws into this technology to wash their own hands clean from facing responsibility. AI never did anything by itself - it was always a human being who pushed the button making it do something (and be it a human software developer programming a line of code invoking AI and using the result it created without checks).

[@TheTuubster]

BTW the doomsayers of AI are usually people who's elite status is threatened by AI because the technology enables others to evaluate their statements and actions, or people who decided to live in an anecdotal world view and the statistical analytical empirical core of AI technology threatens their world view. The pushback against AI technology is basically another symptom of unenlightened minds attacking everything that represents the culture of enlightenment, which Generative-AI stands for, since it reveals statistical empirical truths from the data it processes (like a polling institute reveals statistical truths about the people it interviewed, which many living in an anecdotal world view fight against in a similar way).

[@relaxingnature2617]

why do scientists think they make good presenters ???????????? ....sheeesh its just like watching a car crash in slow motion

[@EskiMoThor]

Who says all scientists think they are good presenters? Even if they are bad presenters, isn't it more likely they get better at presenting the more they actually try to do it? Wouldn't scientist miss out on valuable feedback about the topics they present if they stopped presenting, just because they're not fantastic presenters?

[@samstvshow]

Absolutely manic. The presenter would generate more time for herself if she did not say everything twice, and it would also perhaps have helped if she had not so obviously been guided and molded by late twentieth century and early twenty first century adverts and competitions where the prize, punchline, or answer is witheld seemingly for an age, to hype up the anticipation.

[@apexxxx10]

Dehydrated and nervous!

[@stevedrake6529]

She’s talking way too fast! I think she’s just a bit nervous

[@tracymcgeachie7525]

She is Scottish we tend to talk fast. 😊

[@paulanthony312]

I played it at 0.75x speed and it was perfect 😅

[@stevedrake6529]

@@tracymcgeachie7525I’m from Texas, we talk slow 🤠

[@apexxxx10]

and DEHYDRATED!

[@Annie-dd2uk]

Indeed, I could not listen beyond 2 minutes.

[@harveytheparaglidingchaser7039]

The environmental impact convinced me not to try GPTchat despite curiosity

[@Trahloc]

When compared to watching this video you could have a had month of inquiries to chatgpt. The amount of data and power used by youtube during that hour is ridiculous. AI is in it's early days, efficiency hasn't been a focus just yet, just having it work at all has been.

[@harveytheparaglidingchaser7039]

@@Trahloc ...these are the figures from the IEA "as shown in the chart, below left. The results are highly sensitive to the choice of viewing device, type of network connection and resolution, as shown in the chart, below right. One hour of streaming video typically uses around 0.08 kWh, but actual consumption depends on the device, network"

[@Trahloc]

@@harveytheparaglidingchaser7039 you're looking at the end use, I'm looking at the whole infrastructure. I have AI runs on my pc, my pc that is on 24/7 anyways because its heat is a beneficial byproduct. I own H100-80's and A100-40s. Dodging AI's utility because you think they're destroying the environment more than your cup of coffee/soda/insert anything you consume that isn't *strictly* required to stay alive is just funny to me. The knowledge an AI can impart to you can exceed what your doctor can provide you in the time available during your appointment. Their time is limited. The AI will sit there all day explaining every single test you took so you understand what everything your doctor did actually means. You're just using the wrong measurements to judge things imo. You're comparing a machine with an effective IQ in the 99.9% percentile (and getting smarter) with entertainment. It's not, it's distilled knowledge... but it's a newborn, so you need to learn how to properly interact with it.

[@ViktorBautistaiRoca]

0,5l of water per query. About 30k queries per 1kg of beef. If you are from the UK a query increases your home water consumption by less than a 1%.

[@ernstoud]

Risks and impacts… huh? The generally accepted definition of risk is the probability of a threat happening, causing a (mostly negative) impact. Thus “impact” is already a component of risk. Does she understand what threats, probabilities, vulnerabilities and impacts are?

[@jjcale2288]

She doesn't.

[@karagi101]

Risk is just the probability of something negative happening. Impact is the cost incurred if the negative thing happens.

[@StratsRUs]

I suspect something disingenuous about this speaker. AND Mildly disturbing how she giggled throughout her 'cooking show' and Online Dating.The way tech people lack empathy..I imagine her killing her lovers by poisoning them and burying them in the garden, using AI instruction. I may ask Chat GPT to write me a Screenplay.

[@jacobe2995]

What is a sangwich?

[@stevedrake6529]

Something you eat when you’re hangry

[@buzzlightyear3715]

A British cuisine from 1700's.

[@apexxxx10]

PLEASE slow down your speech and drink a glass of water - I can hear your saliva!

[@harda7xcore]

The risks aren’t common sense? lol

[@geroffmilan3328]

👆 didn't watch cos he knows it all already LOL "Common sense" is a figment of your imagination: it doesn't exist.

[@kennypool]

What language is this?

[@c1ph3rpunk]

Canaustralzealish.

[@kennypool]

@@c1ph3rpunk if you play it at half speed it might make sense.

[@geroffmilan3328]

English. Y'know, the thing most "English-speaking" countries' residents can't actually use without generative AI to help them?

[@tracymcgeachie7525]

Really she has a Scottish accent.

[@geroffmilan3328]

@@tracymcgeachie7525 she does indeed, but the question was "what language" rather than"what accent". I'm guessing you, like me, had no difficulty understanding her. It's always good when less-educated people come to learn something new - it's the point of such lectures, after all! - but depressing how many have problems getting there. The OP indicates a level of world experience so low that learning from others will be hard in many circumstances.

[@danamulter]

This neglects SO much. It's presented from an academic and "business" standpoint, with a big old "children understand bots so let's give THEM the stage" ending. Yikes what a horrible approach. None of this considers the effects on people who wholly make their living digitally. It screams "I'm on big tech's payroll, there's nothing to worry about". Meanwhile, speakers and narrators, voice over artists and the like are fully vulnerable to replacement. The barrier of digital personas people create for themselves as "v-tubers", a mental health and well-being step to participate in internet exchange and digital breadwinning, are now WORTHLESS and indiscernible from bots T-O-D-A-Y. Both of these careers, that your lovely ivory tower spends on children for heartstrings instead of talking about (because you don't understand them), are fully vulnerable to generative bots, unethical replication and training -- and NOWHERE do you talk about the legitimacy of people like Discord training their bots on your private conversations!? You have a niche understanding of the applications being implemented TODAY, and you try to make it sound so harmless with kids at the end -- this is predatory and misleading representation of the "risks". This is "don't regulate me please" big tech conversation ON A TRUSTED SOURCE!? TRI needs to vet these people and their presentations better, this is horrible.

[@petercrossley1069]

Irritating presentation style.

[@apexxxx10]

Rushing like mad and dry mouth! Dehydrated!

[@imetr8r]

This woman talks so fast that I find here annoying! If she could just slow down her talk would be great.

[@vicfitti2008]

Long-winded, superficial, too boring and speaks too quickly to the point where it is assumed that she must be late for some unpostponable commitment.

[@matteo-pu7ev]

This woman is so ***** I cannot believe anyone could take this seriously.

[@paultaylor7947]

Degenerative cognitive disease

[@jayakrushnasahoo4403]

She is not a good speaker

[@R-MD]

Why does she pronounce "sandwich" like "sangwich" The only person in the entire world I have ever heard do that and I have a very deep dislike of the pronunciation every time she says the word.

[@ChristianKurzke]

Lady, take a deep breath. Listening to you hyperventilating is painful.

[@spectrumofreality]

There's no such thing as AI, it is a misnomer for algorythms that function with machine learning.

[@k1ry4n]

Much more scared of the comments here than of the AI.

[@Gringohuevon]

Driving a car to the moon and back is nothing

[@mpouhahahha]

WHAT THE ACTUAL F**** LADY???

[@spectrumofreality]

This poor thing is V injured and she may not even realize it. You can tell by looking at her face, the left side of her face or right side as you are seeing it. She's also clearly struggling with neurocognition.

[@soliton_radar]

What is your point? Failing to see it.

[@anonymes2884]

You seem nice.

[@soliton_radar]

​@@anonymes2884I didn't mean this as a way to dismiss potential illness or problems she has. I didn't understand why he needed to call attention to it and call her a poor girl for it. I'm sure she wouldn't appreciate being seen that way.

[@spectrumofreality]

@@soliton_radar It's called empathy champ. And yes if you watch her microbursts you can tell as well as watching the left side of her face which has some paralysis. She also stumbles a lot and has obvious difficulties with neurocognition...

[@shakeelm5339]

You are extremely confusing and have no idea how to present.