AI: Google Deepmind makes breakthrough in difficult artificial intelligence test | BBC News

Google has created a new AI system that is able to solve complex geometry problems, the company says.
Geometry is specifically challenging for AI systems to understand because of a lack of data.
Therefore, Google’s AI models were trained with language data, ingested by synthesising millions of theorems.
Subscribe here: bit.ly/1rbfUog
For more news, analysis and features visit: www.bbc.com/news
#Google #AI #BBCNews

[@tw1sted1nc]

Isn't it rare to see a news show that gives someone time to explain something complex? Hats off to the producers and the presenter.

[@andybaldman]

Indeed. We need more of it.

[@MuantanamoMobile]

She didn't even explain it correctly at all though. Neuro-Symbolic AI is a hybrid between Neural networks (deep learning derived models like GTP, BERT etc) and Symbolic AI (logic rule based engines like EDA tools used in Semiconductor Chip design). Neural networks basically make derivations or guesses from pattern recognition of vast amounts data, that's why they can often be hilariously wrong just because they haven't seen every possible example of say a Chihuahua, and instead mistake a picture of muffin for one. Or they simply make up information without understanding the meaning or difference. They also require vast amount of data, but still make errors, while a human toddler doesn't need to see every example of a bird - basic knowledge about just 1 is usually enough to know that a new flying thing they see is most like some kind of bird.

[@MuantanamoMobile]

Symbolic systems on the other hand are designed to follow strict mathematical logic and rules not "informed" guess work. They were the older preferred method, that hit a performance ceiling just like neural networks are now hitting a ceiling. A combination of the two approaches allows the AI system to start "understanding" logical rules e.g. that a muffin can most certainly not be a Chihuahua just because it looks like one. or that it is a bad thing to cut of the oxygen supply to a room, because the people in it need it to breath. Neuro-Symbolic Ai systems also use far less data to make better conclusions - no need for numerous vast data centres, less energy use, less waste.

[@MuantanamoMobile]

TL;DR - The basic argument is based on a need to address the two kinds of thinking discussed in Daniel Kahneman's book, Thinking, Fast and Slow. Kahneman describes human thinking as having two components, System 1 and System 2. System 1 is fast, automatic, intuitive and unconscious. System 2 is slower, step-by-step, and explicit. System 1 is the kind used for pattern recognition (neural networks) while System 2 is far better suited for planning, deduction, and deliberative thinking. In this view, deep learning best models the first kind of thinking while symbolic reasoning best models the second kind and both are needed for an actual Artificially intelligent system.

[@dredzysaint]

Thank you for that bit of information! You should check out the book Rationality, by Eliezer Yudkowski​@MuantanamoMobile

[@Maxime-fo8iv]

AI student here, and I'm very pleasantly surprised to see a general public emission take the time to properly explain advances in AI. With so much sensationalism and preconceived ideas around AI, I think a well-researched explanation like this one deserves praise.

[@priyalakhani1461]

Hi :) Thank you very much for your very kind comment

[@JoeyBlogs007]

The reason why it's significant is because the real physical world is in 3D and geometry is used in the real world. Everything humans do in the physical world is done is 3D space and calculated accordingly, consciously or otherwise. Our visual perception is based on 3D pattern recognition. That's why this is a really big deal.

[@vincent_hall]

Good point! The real world is 3D. Or 4D with time, but yes, geometry is needed for our lives, the world.

[@asdfasdfasdf1218]

The bigger reason it's significant is that they're finally doing deductive reasoning. 2D geometry proofs have rather limited application in the real world, so the significance of this is probably not applying this directly but by trying to extend this to other math problems later on, this is just a stepping stone.

[@e4t662]

"If you act like you know what you're doing, people will generally assume you do."

[@ninedials]

did you say that or did I say that? Must have been a good night either way.

[@6whatnext]

Liberal western politicians don’t view the CCP with revulsion but with envy

[@anypercentdeathless]

This segment was a train wreck.

[@onagain2796]

​@@anypercentdeathless 90 IQ idiots saying this because a woman was explaining it. She is entirely correct. Seriously WTF are you mentally stunted ppl talking about?

[@davidjulitz7446]

yes, and this is often a problem. because acting is mostly not enough to reach goals.

[@declup]

Modular specialization in machine learning is such an obvious strategy for developing AGI that I'm surprised that news and commentary media haven't given it more emphasis. And, yes, ML groups do already specialize in areas like visualization, robotic control, theorem proving, translation, content summarization, etc., but even their relative lack of attention to a modular approach surprises me a little.

[@H0lyFr0ggy]

good point

[@foxt9151]

Exactly what you describe, has been the research focuse of many groupes the past few months so, good, I also believe that this approach will bear a lot of fruit

[@brett5141]

How do we know the UK government isn't using facial recognition the same way China is?

[@Jake_5693]

The police use facial recognition already so of course the security services are all day everyday

[@dramman-onlydramnodrama4845]

I see a lot written about this guy Al, and he seems quite bright. What’s Al’s last name?

[@mikhail.drozdow]

🥁

[@BooleanDisorder]

Aye

[@honkytonk4465]

Overlord

[@BlessingsfromBridget]

AI is a she. She is usually represented as a female robot, so she’s a she. Her last name is Ameca.

[@TomNook.]

Gorithm.

[@GonzoTehGreat]

Kudos to the BBC for covering this AI/Tech news, but they should really have a dedicated programme for this topic by now - perhaps a brief segment after the main news, as with sports news, or even better, a separate program, like Newsnight. Our modern world is built on STEM and the future depends on it. It really deserves far more news coverage than it received, especially when compared to Politics or Economics. BBC Click is excellent, but doesn't air often enough.

[@human_shaped]

You need to increase your segment length a little.

[@davidbruce2546]

Thank you for the segment! I’ve been looking for a source, which covers this topic, that keeps up with the progress of AI. I appreciate the level of the conversation. Not to over my head. I need to learn the jargon. Makes finance look basic. Impossible to keep up without reviewing academic papers, which are growing leaps and bounds. Is there a way to be notified of this specific segment each week? I like the BBC, but not enough time in the week to wade through all of its material. Keep up the excellent work!

[@edahmed7]

From law to AI Priya is good. From someone watching BBC news from the gulf region I feel she is like the Richard Quest of AI. BBC need to get her on a long term contract asap!

[@priyalakhani1461]

Hi! Thanks for your really kind comment!

[@timswartz4520]

Make the ai check it's work backwards a few times while solving math.

[@emc3000]

Excellent explanation of the digital twin concept.

[@roscaris6541]

Very fascinating developments. I do have an objection to the philosophical interpretation often being overlayed here. "Knowledge", "reasoning", "understanding" are all terms that are ambiguous; on the one hand they may be used as a mere useful metaphor to express the idea that an unconscious complicated set of complex physical causes create a product which to a conscious mind replicates an element of consciousness or understanding. So used it is fine. But many people will not realise that this is a metaphor, or those who use these terms may intentionally mean to communicate that there is true consciousness and mind-understanding at work here. That conclusion though remains unwarranted from the products produced by AI. The ability of your calculator to reliably reproduce the complex patterns and truths of mathematics does not require any actual understanding on it's part. So also, reproducing through pattern forming selection algorithms, the connections that underpin certain geometric truths doesn't require actual reasoning of the sort that a conscious entity engages in. I think we should avoid such terms where possible.

[@dg-ov4cf]

damn, GPT-4 is getting scary!

[@danielswift8558]

P

[@williamwillaims]

You use the term "conscious entity", that assumes we even know what that is. The point I would make is that it is irrelevant whether we make an Ai that is "conscious" or not. All it needs to do, to cause massive social and economic chaos is make us believe it's conscious- jobs will go either way. Hello labour vacuum 😂

[@williamwillaims]

Also remember, everything we are seeing right now is effectively beta, we are in the kiddy pool, wait till we step into the deep end of the big pool, or the ocean (50yrs). All these perceived or real limitations will melt away with each generation. There will be, a 4th generation AGI one day. Tip of the spear folks are saying gen 1 is basically here already

[@0zyris]

What is the difference between artificial consciousness and true consciousness and how would the entity experiencing it be able to determine which one it is having?

[@camdennorris646]

Great episode. Not often enough

[@picassopete3766]

Overworked Public Defenders can quickly get precedents for their cases and decide how to fairly proceed. Judges can be fairer with all the AI information relating to their cases. Excellent!

[@galzajc1257]

i'm really interested like how could they generate true geometry statements? it seems kind of strange cause if it's a simple precedure to generate some theorems, they are probably all totally similar, which means ai has only mastered proving a very narrow set of theorems. those are just subtheorems of a more general theorem on which algorithm, that generates those theorems is based on. in other words if they tested it on some other simple geometry problems (not in the set of problems, that the algorithm can output) would it be able to solve them? the answer is probably no, but even if the it's no it's still a big step, cause if you can use similar methods to generate a lot of simple problems from some different part of math and than make them a bit harder and like cover all the basic math this way, idk, it might leed to great results. at least it would be able to do basic math.

[@kevinb9830]

Google with AI is just creepy

[@aric85]

Right...

[@vladdestroyer7002]

i welcome our future AI overlords

[@tuckerbugeater]

grasseater

[@double2mo382]

Yes! You need more time. So interesting.

[@TuaTagovailoaTouchdowns]

The depth of Google's DeepMind is... quite deep.

[@othomile]

Anytime you watch someone from the tech world interacting with media personalities or government ofdicials you can clearly see the huge gap in understand between them .. simply by what they fail to question or what they fail to probe more or what interests the in the matter and so on and so forth

[@tuckerbugeater]

great

[@simplisticscientists]

Robots can do all sorts of things with maths that humans do with guessing. I know because I have played in a band with musicians who think mathematically. They still make music, but they get there in a different way. This opens doors to what robots can do in the real world which can be mathematically modelled.

[@sheepshoof]

Just look at his smile 😊

[@7350652]

very good

[@GraeLaws]

Superb!

[@TomNook.]

Are we calling all complex coding solutions AI now? That bus timetable problem can be done with normal algorithms, not generative AI

[@jon-xd7tl]

We only want it if we don't know what it's doing

[@priyalakhani1461]

hi! They are using 'traditional' machine learning, not gen-ai. Yes, could absolutely make efficiencies using data science, but they are incorporating machine learning to help make predictions and then decisions.

[@MrLolrobinson]

A pity that the expert is apologizing for explaining…. This is why mainstream media with their penchant for breaking everything to soundbites is so disappointing

[@MuantanamoMobile]

She is not an expert though...but is a lawyer by training, slapped the word "AI" on her online-education company, now she gets to be called an AI expert. Even got the basic explanations of a triangle wrong.

[@projectcontractors]

"One can't help but feel the pulse of excitement knowing that the boundaries of what AI can achieve are being redefined before our very eyes." ~The AI Report

[@jaybro3713]

What's the breakthrough

[@Kalkiavatar20231]

You are an amazing Anchor 😊

[@The_Savage_Wombat]

Before AI learns to reason and argue in court, judges should learn to reason first so they can understand the AI arguments.

[@andersonsystem2]

Wow 🎉

[@AaAa-je5eo]

When did Peter Jones leave Dragon's Den to start presenting the news? 😁🤪

[@ingilizcehazrlk9134]

Good

[@werewolvesandfriendsuk]

Say right one more time!

[@Christine.corneille]

Is fascinating in that case quiet equivalent than fashism

[@honkytonk4465]

What?????

[@prathamesh413]

2 sides equal is equilateral triangle💀

[@jon-xd7tl]

Not familiar with elementary geometry, but can explain the latest AI breakthroughs .......

[@MuantanamoMobile]

@@jon-xd7tl Because she is a lawyer...and not an "AI Expert" as presented.

[@Joan-ot9nf]

1:11 Samsang?

[@maxstockdale7328]

Will First Bus also be using AI to make their buses turn up at all?

[@antonpictures]

Google's approach to tackling complex geometry problems with AI demonstrates the versatility and adaptability of artificial intelligence. By training their AI models with language data and synthesizing millions of theorems, they are essentially bridging the gap between linguistic understanding and mathematical problem-solving. This could have significant implications not only for the field of AI but also for educational methods and the development of new tools to assist in learning and research in geometry and other mathematical disciplines. The use of language data to enhance the AI's understanding of geometry underscores the potential of interdisciplinary approaches in advancing technology.

[@bigglyguy8429]

Thanks GPT

[@james_whitford]

"Need to get myself one of those, hah"

[@Ernesto1317]

Great achievement of the Corporations: to have the technological means to replace human beings.

[@harrybarrow6222]

Even if the AI face recognition is not perfect, it can search through many hours of CCTV recordings to find interesting events for a human to check. You do not need a judge’s authorisation to do that, and the human recognition or interpretation can be used in court.

[@louisdiblasi5441]

I here deepmind is moving to Paris. Why pick a location that is 5 billion miles away?

[@shamsulhaquehero5651]

NEWS TO PUBLIC AND THE WORLD

[@agritech802]

In relation to face recognition and privacy, the only reason you'd need to be worried about being recognised by ai is of you committed a crime

[@arnaudgerard1971]

Does it ChatGPT? If not, this is for the ivory tower.

[@khansvirtualdiary]

AI will take lead in every sector in coming years.

[@raul36]

PseudoAI. Artificial intelligence does not exist yet. For more information, see Mr. Lecunn

[@user0K]

Eu: we care about privacy and tracking. Also Eu:

[@DoveReactions]

2:52 What happened to her voice then?

[@AnonYmouS00816]

She's ai

[@thenoobalmighty8790]

​@@AnonYmouS00816yeah she a synthetic for sure

[@skycloud4802]

A glitch in the Matrix

[@user-we1rv5tw1c]

Great AI artificial intelligence

[@Whysicist]

Only linear models have one solution… quadratic has 2… see a pattern here… high-school? Polynomials?

[@jasonkocher3513]

We need this for CNC machining toolpaths, among many other things in engineering. LLMs can't infer geometry on their own. It's like a journalism major trying to explain this Deepmind breakthrough.

[@dudeperforming4188]

So it is coming

[@herby4215]

Clay and metal bad omen

[@WeylandLabs]

Google making breakthroughs but bard gemini still sucks 😂

[@MuantanamoMobile]

That's built on Anthropic's system..a startup they invested in. Just like OpenAI was invested in by Microsoft, and is used in their products.

[@quentinpaden1481]

Revive KZbin

[@jzimmer11]

I like that she obviously knows what she's talking about!

[@philhealey4443]

OK, right, right OK. Right.

[@josefelixaranda9002]

The lady is right because A.I. is built and developed like a child first in the school,like teenager in the secondary school,the young man in the university and finally,ideal success,a man like the Knight Rider when everything will be perfect and the Artificial Intelligence driven car is safe and profitable for ourselves the human passengers.

[@anderbeau]

In 10 years we’d be wondering why we do things the way we did now like a Nokia phone to an iPhone 😂

[@vincent_hall]

AI would have reached the end of actual data in a few years anyway. So we need synthetic data.

[@UndertakerFromWWE]

We are all doomed

[@aliasgur3342]

Legal judgements in my opinion should be based on logic and evidence

[@jon-xd7tl]

Nah, just ask ChatGPT if the accused is innocent or guilty

[@aliasgur3342]

@@jon-xd7tl Better than asking a lawyer who has human biases or simply got out the wrong side of bed that morning

[@leeperkillz5397]

Logic and reason on the basis of emotional intelligence, a mix of both is significant in judgement, but ultimately it should be restricted by basis of what is moral

[@GameGevUA]

4:26

[@user-kx8np1pu2k]

I have been using Maths all of 64 years ,that is HI human untelligence

[@cybernescens]

What in the world is "synthetic data"?

[@paulera2473]

another fancy word for data

[@arthurd6495]

I asked GPT4, "what is synthetic data in the field of ai. Give me a four sentence answer. " Here is its answer: "Synthetic data in the field of AI refers to artificially generated data that mimics real-world data. It's used when actual data is scarce, expensive, or sensitive, allowing for the training and testing of AI models without privacy concerns or data scarcity issues. This data is generated through algorithms that simulate real-world patterns and statistical properties. Synthetic data is crucial for developing robust AI models, especially in domains like healthcare or finance where real data can be hard to access due to privacy laws or rarity of certain events."

[@EmDee3d]

One example of this.... Realistic virtual worlds (like the GTA games) being used to train AI models for self-driving cars.

[@cybernescens]

@@EmDee3d thanks for this very understandable answer.

[@TheFlyingEpergne]

little bit of tension between these 2 lol

[@GamerplayerWT]

Here’s my question for AI: “How do we create a stable wormhole?”

[@gaiustesla9324]

You can't, wormholes arent a thing. Matter creates space with fields, its impossible to part the fields.

[@nishantb80]

AI will have major impact on the job market. Loads of job cuts in the near and short term for sure...next 5 years shall be very crucial. IT companies should be taxed higher than other industries ASAP. Overall, AI can benifit daily in different areas for sure. Companies leveraging AI in medical, logistics planning and transportation planning or crop production should be exempt. On the other hand, companies using such services for luxury, entertainment and nonessential things should be taxed more.

[@louisfivaz1367]

Wow, the guest speaker must have both a major in stating the obvious and dancing around the point.

[@bhgtree]

AI: 'Puny humans, I am better them you in every wa.........' Human: _Pulls the electricity plug out_ 🤣

[@vincent_hall]

Oh I love her. She's intelligent. She knows her stuff she can teach others. This is what we need. Oh! Priya Lakhani OBE!

[@tanyamccann1747]

I can assure you Priya that I haven't.

[@christinetipple6285]

Right right right right ok right right right ok.......NO!

[@jw999]

The topic is obviously way too complicated for the host.

[@syednajamulsaqib7664]

This is why I love science. Had it been a political talk show, we might have seen a lot of conflicting narratives coming from intolerant speakers.

[@user-hi1ke5gp6n]

Мечтать насильно

[@DavidGalich77]

AI is good for many things but as we know we cannot lose the human touch.

[@codyeasonBGR]

Real boring edit here could have used some more graphics or a faster edit i just was having trouble fallowing this with out an example. But what they are saying is important.

[@bingeltube]

Too much talk show talk! No interview with Google staff etc. No excerpts from the paper etc.

[@clavo3352]

English common law is logical. Can AI handle legal fictions, like statutes of limitation?

[@catvisimo]

they will make the data to replace your job

[@liliaaaaaaaa]

Actually what I find more scary about this, is the fact they are stating that Ai is not a calculator, and when AI doesn't know what it's doing, it just blags it. Meanwhile the humans attempting to justify how great they think Ai still is, despite the fact that no, it actually is not logical at all, are also essentially blagging it. Oh, and the fact that anything you photograph on your smartphone will be 'scraped by AI' and sent off to surveillance, just an aside.

[@MetronomeMarionette]

Psalm 46:10

[@MizanRahman-oc3qt]

Difficult

[@masterwatch]

"AI" is still very misunderstood by most people. Often a much better explanation of what people are calling AI is to use the phrase Machine Learning that is probably a better description of what they are talking about. Reaching real conscious AI would be a completely different ball game and will change society on Earth for real. Many actors in this space are wanting the narrative but if real conscious AI was achieved, you would know about it.

[@davidjulitz7446]

Its only a hype for now. The developments are impressing, but AGI is out of reach. Btw: what happened to the Blockchain?

[@masterwatch]

@@davidjulitz7446 it is only my opinion and I do not know but I would say Bitcoin and Crypto present a huge financial risk to people who are using them for Get-Rich-Quick purposes. Central Bank based Fiat money is not perfect but it is a lot better than Cryptos.

[@masterwatch]

@@CleereStarre I do know training models have used or based their systems on that of young people and how they learn behaviour and ability to process information.

[@MuantanamoMobile]

@@davidjulitz7446 Yes. exactly it a good way to fleece investors and the general public. Development are being made but they are over hyped for profit.

[@asha802]

Anybody considering a change of career. Get into AI, Machine Learning, Networking, Cloud Services or any other relevant IT field.. It will take some effort but it's the place to be.

[@andrewmartin1758]

Until you’re replaced by AI

[@ANGELCRYPT0]

Stack your Sats and XRP folks

[@bassoprof]

AI is already a very complex subject. But when the person explaining the subject speaks with the speed of a machine gun, and with sub-standard English, plus with a strange intonation, millions of listeners, almost all layman, get frustrated, angry, and foul-mouthed. The woman on this BBC program is a typical example of this phenomenon.

[@markoconnell804]

The eu would be violating human rights by facial recognition without a judge order.

[@pauldannelachica2388]

❤❤❤❤❤❤❤AI using supercomputer new materials . Drug Discovery. Etc

[@raulmartinez429]

This is not about safety is about taking our freedom and arrest those that don't submit to corrupt scams like covid guidelines

[@samiranphukan8441]

Is ai leading us somewhere

[@user-rg7uh9se4c]

OT: Benny Hill considered funnier than Monty Python by two TV stations --WOR and WLVI.

[@dilir2572]

Waiting for the day AI gains human emotions

[@BrianMosleyUK]

10:45 you're biased and wrong. AI is better at sentiment analysis and reading body language than a random judge. Also, justice shouldn't depend on how good the defendant is at acting innocent.