Flint Read “Dammit, Greg! If you can’t quit cooing in class, I’m going to have to ask you to leave. Unlike SOME pigeons here, the rest of the class are interested in learning oncology.”
Glenn Pollock Sadly, it’s only a matter of time before Greg gets busted self-proscribing opioids and loses his medical license. Despite the high-paying job, trendy nest in Chelsea and young Frillback Pigeon trophy wife, Greg is one sad bird. How many times has his assistant had to pour him into a cab and send him home after too many martinis at lunch, I wonder?
She is not only a good mathematician, but also a brilliant speaker. I watch her explaining mathematics all the time and she always connects with her listeners.
@@goromaster10 only stupid people (including ones with high IQs) ring up massive student debt. Smart people go to state schools, get jobs and study practical majors.
OMG, the final ad, right after Hannah starts to explain the dangers of algorithms... Husqvarna self driving electric lawnmower. Well played AdSense, well played.
So a recommendation algorithm recognized that a biological algorithm was learning about algorithms and responded with a suggestion for a machine with an algorithm?
Ad sense works on your past behavior within Googles ecosystem, so you may have been doing research on stuff or must have been reading mail in Gmail just before you stumbled into this talk. The content of this video influences it only slightly compared to all the activity you had done before this video. Big brother is watching!
As a longtime software developer, I don't trust algorithms to be particularly unbiased or better than humans at most tasks. Software is not magic. It's written by and trained by flawed humans. The best description I can give is this: most algorithms are idiot savants. The software may be good at some things, but it doesn't have a model of the world anywhere near the sophistication and subtlety of your average human being. Trusting a decision because "the computer said so" is a rather naive way of thinking.
He said he enjoyed the talk. Where are you seeing "impressive"? And being knowledgeable about a field and being a good speaker don't necessarily go hand in hand.
@@Ken.- I know what he wrote, i looked through the comments and found 5+ of these saying the same thing. Wow so impressive ect. Knowledgeable part is far more impressive than speaking but the thing that goes everyones head is that THIS IS EXPECTED OF THEM. THEY ARE SPEAKING IN AN AUDITORIUM. And thus it cant be considered wow impressive much love. The bar for that is set much higher. This is not about gender either. I have not watched a single talk in this channel and found it boring. Meaning all of them are more than good enough to convey the subject at hand. This leaves me to believe theres clear bias for a pretty lady talking about complex things somehow being more impressive because she is a beautiful smart woman. I wonder who writes these comments. Cant be men because that thought never came into my head. Must be women power tripping or something lol.
I first saw and heard Dr. Fry in a televison production titled, "Magic Numbers" on the nature and history of mathematics. I was thrilled - especially when she got to quantum mechanics. She is a wonderful, articulate and concise teacher. I find her balanced approach to algorithms and AI comforting. It was heartening to hear Dr. Fry say that the music in the style of Bach was not actually composed music. Programmers are working what what they know about Bach's music. It took the genius of Bach - the person - to create this type of music in the first place. And yet, it can, "...fool a roomful of people." We as the general public are not able to differentiate between content created by advanced AI and the real thing. This has serious implications for societies. Dr. Fry is so right in saying that we need to think about how we use the algorithms and how humans fit into all of this. She is right when she says, "You can't rely on people." One reason you can't rely on people because intelligence is unpredictible. People do not alwyas act in their own best interests. Artificial intelligence will be just as potentially unpredictible. This is a qulaitatively different creation than anything humankind has ever created before. I fear that there are many others who do not share Dr. Fry's balanced viewpoints. Research is roaring ahead from the current levels of AI (which have already been used to cause havoc through spreading misinformation and feeding people inciting content to further keep them engaged online) toward Artificial General Intelligence (AGI) which will be much more powerful and have sentience. There is theoretical potential for later, advanced AGI to be using us instead of us using it. Consider how humanity has treated and used "less intelligent animals" through hisotry. AI will potentially be able to self-replicate, creating mutliple copies, thereby making itself safe from being shut down. It will be able to self-upgrade, self de-bug those upgrades and each time learn how to better self-upgrade the next time. With upgrades, it's architecture will change. Those who created it in the first place may not understand what it has become, making it all the more dangerous. Interfaced with our infrastructures, it will potentially have incredible power and control. It's goals - satisfy what it has been programmed to carry out. And if it perceives people as either unnecessary or worse, a threat to it - what then? What consequences lie in store for us when we have become the "less intelligent animals" from the pont of view of sentient AI? There must be more public education and debate on all of this. I am not optimistic.
What's interesting is, as a trained musician, it seemed much more obvious regarding your example who was who, and I think it comes down to intent. With bachs piece I can tell he is trying to push and pull different things into your attention span because he is aware it is limited and not everyone he is composing for will hear the intricacies in the chords. The computer has no such audience it is playing to so it takes the most complex pieces and rules and throws them at you all at once where the ideas become muddled, it's playing to the rules given, if you further limit those rules I'd imagine it becomes much much harder to figure it out. What each is trying to do seems to be the differentiating factor here. Bach is trying to communicate to the listener. The computer is trying to come up with every permutation within the rules as correct answers. You might be able to trick the audience even better by selecting one of Bach's earlier works.
Algorithmic composition is one of the greatest failures of computer science, so far. We still seem to have have little to no understanding of what exactly determines the quality of human music of any era. We can ask our greatest composers and musicians how they are composing and performing and many of them can explain in great detail what it is that makes their music "special". Our conductors can tell every instrumentalist in an orchestra how to play their part so that the entire orchestra comes together in a way that is far more than the sum of its parts. What we have not managed is to transform these explanations of how music is being created into a set of rules that can compose as much as a simple folk song or that can play a straight forward blues progression in a "natural" way. I find the depth of this failure utterly fascinating.
I'm fond of saying art is self-expression: no self, no self-expression - and the reason we express ourselves is to communicate with other humans. Bach knows people; the algo only knows rules.
@@tantiwahopak101 describing something or someone as "full(y)-fledged" is a metaphor that comes from the life cycle of birds. A fully fledged bird is an adult one which has shed the down it was born with, and grown its adult feathers. A full fledged pathologist is a professional one who has completed his/her training
On the gorilla thing: human brains are actually very good at pattern recognition, so I'd say that human doctors are well equipped for spotting what's wrong on x-ray images. Missing the gorilla is just a case of attention bias. Gorillas normally don't show up on x-ray images so the shape is not registered by the brain.
This talk popped up on my algorithm, which is currently finishing my sentence before I compose it. Hannah speaks to this topic with more salience and forward thinking solutions than I've heard even here in 2024. TY Ms. Fry! 🌎🌍💜
Why can't at have more schoolteachers like this lady? I bet everybody would learn so much more. I could listen to her talk about any subject and I'd be fascinated.
I find the 50/50 Bach result surprising. #2 is so clearly Bach because it contains a cohesive emotion which is something a computer cannot replicate well. I don't know if I recognize that emotional aspect because I'm musically trained, though. Do 50% people just not experience music on an emotional level? My face and head literally tingled when the real Bach played.
So I have to admit I am not quite sure if I understand your point. How was the first one not cohesive? What were the indicators that gave it away for you? I am sure that there is tons of music written by humans that is not emotionally touching. I would also assume that different humans react differently to music. I might get emotionally carried away by some blast-punk and feel nothing while hearing bach. For you it might be the contrary.
@@niklasb6849 Yes, the first one is technically cohesive, but it doesn’t sound like a human wrote it because it lacks emotionality that only a human can give it. There is no “story” to it, it just sounds like Bach. The whole magic of writing music of any genre is to convey emotional stories. Whether or not you get emotionally carried away by it is a separate experience and will vary by personal preference. I just get head tingles when I hear music I like, whether that’s Bach, punk, hard rock, or a pop song.
Interesting - I experienced more of an emotional reaction to the first piece. Just kidding, it's not interesting. That's just how human emotions work - they're completely subjective and contextual.
I’m a new Hannah fry fan. I saw the Berlin talk before this one and saw how uncomfortable she got in the Q and A. She dealt with it really well and was very interesting to hear Hannah talk about that in this talk. …People… make More people like Hannah please.
It's all academic - there is zero chance that democracies will put a computer in charge. People don't trust politicians, let alone a computer program that could be infiltrated and biased.
"man plus machine" was theorised to be why we got the Boeing crashes (I didn't keep up) If there's a silver bullet, this doesn't seem to be it (though I'm all for improvements where we can get them)
Augmentation seems the best possibility in general it takes advantage of human insight and machine rapid processing. In specific domains machines will be vastly superior. Currently games like chess/go and even real time strategy games -- AI is way ahead.
What made it clear for me, was the random high note in the Bach work. I heard once, I think, that they used to put random notes that were out of place so that the audience didn’t fall sleep lol.
Hannah Fry is a British version of a very smart woman at my work, who I have always had a soft spot for. I could listen to her voice for a very long time!
tarquin161234 I agree, humans should change their mentality. But the whole problem is money! Besides programmers aren't all the same so you want to trust a programmer, who's a human But you say humans are selfish and biased? Ever heard of malware? Or ransomware? Or did you think a computer virus is a biological virus that can infect computers? 😂 j/k
@@sendraw4088 Yeah you're right. Computer programs are no different to people because they were made by people. Clearly a human/democratic check is always going to be needed, so the recommendation of the video is a very good one (for computers and algorithms only to be used as tools by people.)
Terrible argument. The Republic was written by a "biased and selfish" human, but his instructions, if they were followed, would not lead to biased and selfish robot politicians. It's akin to saying railway logistics are inherently flawed because they were designed by humans, completely ignoring the fact that humans are only biased and selfish in very specific areas, and in those areas computers would obviously do better.
I'm a musician and I had no trouble picking which piece was written by Bach. I can't tell you why, exactly. I can only say that there was something stilted and mechanical about option #1. It had no life. Now, that could be because the arrangement was too simple or because the orchestra and choir didn't feel it the same way, I don't know. But in the end, it just didn't touch my soul. The Bach piece did.
Relatives of my wife got stuck on a dead end street at a river where once was a military bridge. They were driving by car and completely relying on their navigation system. The algorithm for route planning was not the problem here, but the outdated map. The funny thing is, they got stuck several times at the exact same place.
Almost by the 30th minute I guessed the conclusion of the talk would be synwrgy between machine and human. I remember being part of a debate where I suggested the same. Very inspiring lecture Dr Fry. Thanks.
It's kind of interesting that the talk at re:publica in Berlin might have been one of the worst moment in her professional life at that moment, but in hindsight the audience in Berlin helped her to grow in a way that otherwise might have took much more time if it ever would have happened.
aidan levy, do you think I need to go to the Ministry of Truth for some political correction ? . & why would my saying that preclude me from going to a conference ? . are you a thought leader as well ? I'm not very good at this double think malarky, can you tell ? :)
I am a pathologist, I think the problem here is that the diagnosis of breast cancer by histopathology slides is really an easy task, any medical student who trained in a pathology lab, can diagnose breast cancer in 1-2 weeks of training. I am really interested to see the results in really difficult cases, in conditions where someone have to judge using different resources together (history and physical examination , Imaging data, Lab data and histopathology slides) the results will be different.
When I get my own personal AI, and can customize it, I want it to speak exactly like Hanna Fry. Her voice, timbre, diction, everything... so easy on the ears.
OK, I have a question. It seems to me that based on combining these factors: 1. 32:24 - "now this is the stuff that algorithms are amazing at" - being sensitive; 2. 33:15 - "humans, when it comes to being sensitive, we're rubbish, right, we're totally rubbish at this"; 3. 34:18 - "but specificity, being specific, that's like our [humans'] super-power." ... you (Hannah) then present(s) the idea of having sensitivity done by algorithms, and specificity done by humans -- 35:15: "this is the version of the future that I'm really hoping for". Yes! I think this is absolutely a brilliant way to approach things. Because: When thinking about this stuff, I think about spam (or unsolicited commercial email, or UCE) detection, and I think about how Google's Gmail product has done a wonderful job of keeping UCE out of my inbox... and a fairly poor job of keeping legitimate e-mails (sometimes solicited commercial e-mail, sometimes just stuff from friends and things) out of my Spam folder, and _in_ my Inbox. I get alarming numbers of false-positives (stuff flagged as spam, that's really not), and very few false-negatives (stuff not flagged as spam, even though it actually is). And yet, I think another thing that needs doing here is to try to work on getting the _algorithms_ to be better at avoiding false positives. Because realistically, while I think _I_ am good at finding those false positives in my Spam folder, I'm not so sure that _anyone_ would be - a lot of the ways that I detect them are taking specific knowledge about my own e-mail activities and applying them to find messages that ended up in Spam but shouldn't have. So I think that if, say, Google hired a bunch of folks to go through low-confidence flagged-as-spam messages and pick out the ones they thought were false positives, those folks probably still wouldn't find all that many of them (I'm guessing), compared to what I can do on my own inbox. To the point where I suspect that if I had the proper tools to do so, I think I could probably assist in training an algorithm to detect the false positives much better than other humans could... and perhaps, ultimately, better than I myself could. So, do you (Hannah, if you happen to read this, and/or anyone else who's bothering to read all this and likes to think about such things) have any thoughts on what sorts of things can be done to increase the specificity of algorithms? One thought that I had while writing the above paragraph (two above this one) is to have an algorithm that (for example, in the context of Spam) is trying to detect not spam amongst all e-mails, but false-positives amongst messages flagged as spam. In other words, and in the general case: could we use the sensitivity of a second algorithm to give specificity to a first algorithm? So two separate algorithms are used for a single problem, one to give sensitivity, and the other (through sensitivity for false positives) giving specificity? Whether or not that might work, what other things might we be able to do to get specificity to be better? Because I'm of the opinion that if we're going to be using algorithms for this stuff at all, we'd ideally like there to be a way to have both dials pretty high on the algorithmic side of things, at least for many tasks. For some problems, having humans deal with specificity while algorithms do the sensitivity might be fine. For others, I don't think it would be practical, at least, and perhaps not even feasible.
there are what are called open source software where you can see the algorithm used, only open source should be acceptable when making these kind of decisions. Even a human must write a justification for a certain decision, the reasoning behind a decision should always be public all these problems can be fixed, they do not detract anything from the fact that computers and computing are powerful and preferable tools for justice and public administration they are simply not used enough and used properly. Most of public administration could be totally automated by information technology, mistakes avoided, thousands of wasted hours saved. A citizen should not need to make declarations every year, should not need to queue in government offices to fix and update information that is already available to the state. Because ALL the information is ALREADY available to the state via other mechanisms and registries. List of events like marriages, births, deaths, divorces, property purchases and sale, employment and unemployment, minors in charge etc. In fact ALL is already available to the state but the current system is based on inefficiency and authoritarianism where the citizen is made responsible for maintaining data the government has already even though the citizen pays millions in taxes for a services that is not provided! The state goes to the extent of calling the citizen a criminal or fining them for not doing a job the citizen is already paying the state to do. I would rather use processing power properly and have the state provide me with the services I pay for instead that giving the job to a bunch of incompetent politicians, government officials and judges that demand to run my life and fine me for heir mistakes too.
I wouldn't go as far as to say "all these problems can be fixed" but sure, most of them can. And open-source is definately the way to go with tools as important as these as well as anything made with public money (With the exception of national security). Also an important part of making them open source is a generous bug bounty to actually make people report bugs instead of just exploiting them. There should also be a (properly, unlike most datasets) anonymized dataset of the decisions made by the algorithm to see if they seem valid and if the algorithm needs tweaking.
@@Sibula if we are thinking that we can cure aging and fix almost any disease, we can certainly fix man made algorithms. and there is a very important difference to how I would use computing power to how she and the state is using computing power. She worked to put computing power at the service of the state to control people. I would put computing power at the service of the citizens to control the state. Computing power to control the state can NEVER fall into authoritarianism while concentrating power into the hands of the state it's sure that sooner or later it will fall into authoritarianism.
@@canemcave I don't really know how you're planning to "put computing power at the service of the citizens to control the state". Someone would have to develop and maintain it, giving that person or organization power. And what would it actually control? But there are many ways you can use algorithms (eg. machine learning) Where it doesn't really give any power to anyone, mostly just makes everyones lives easier. Diagnosing illnesses is one such application.
@@Sibula machine learning is just one type of algorithms, we are using algorithms every day for all sort of things, from industrial process automation to document editing. How can you give power to the citizens to control the state? first by making proper use of resources already available to the state (and therefore proper use of the money you paid in taxes) and putting the responsibility for managing and maintaining that information not on you, the citizen but on the state you are employing for that purpose! second by making algorithms public and requiring traceable explanation and reasoning compulsory for any government decision and thus making government accountable and transparent third by applying same rules impartially, without human interference you remove abuse of power and arbitrary judgements and again introduce traceability and accountability in the system. Information systems should record every action and decision made by any government employee and reports should be immediately available. Databases could be made to exchange information only requiring your express permission for the data to be exchanged. trivial example a national and a foreigner couple marry in a country. They should only present their passports, the local authorities should register the act, and transmit it to the relevant authorities. For the foreign passport the event should be transmitter to the passport original consulate or embassy which would than have the responsibility to register it in the relevant registry. Simple, which country does that? No, one! it's a miracle if the information is recorded and transmitted within the country! this type of example can be extended to most government services, they are just not created to serve you, they are created to exploit you. There are many other things that can be easily done by information systems and we do not even need to employ AI systems. From transport systems which should operate Nationwide as credit cards do to sentencing which should be handed down according to a set of legislated rules and averages, irrespective of sex color or religion, information systems could be put to good use to make life easier and fairer, but they are not. Instead the state spends your money trying to control you. it's the time to change this state of affair, government are there as a service to the citizens, not to fill their pockets.
@@Sibula the purpose of open source algorithms is exactly that, removing power from the company implementing the system. An algorithm that, for instance, assigns sentencing, should be open source and any modifications to it should be evaluated and approved by a public college of IT and legal experts appointed for the specific purpose of creating or modifying the algorithm. an algorithm for sentencing does not require AI, it needs to embed the various legal rules properly, it could be as trivial as implementing an expression such as: ((mean period for offense + mean prison population for offense) / 2) + ( Sum period for aggravating circumstances - Sum period for extenuating circumstances ) Do you think a judge on the spot can elaborate all the rules of law properly when a team of scientist, engineers and legal experts cannot properly even defy them as an algorithm? The judge will brush aside most of the rules and go with what intuition or mood will tell within a set of constraints he or she will have developed from experience.
As a guy who started programming computers before she was born, about 45 years ago, I don't trust computers unless there's a human able to override their decisions. Computers are our *TOOLS* , and can never be capable masters. Never have, even though I was more-than-passably good. "To err is human; to really f**** it up takes a computer." My real focus was displaying the output decisions in a way that the human could accept or reject key pieces easily. My goal, if there was some bit of AI in there, was to indicate the "thinking" leading up to the final answer. I think that's nice to have; in some arenas, vital.
Thanks The Royal Institution.☺ Thanks Hannah Fry.☺ "50-50 then. Basically guessing at random. And you laughed at that pigeon...."😀
5 років тому
a great example for an algorithm would be the way change is given from a cash register. Only problem: many academics start to do mental arithmetic, when ask to look after a cash register. and teachers tell pupils they have to be good at 'mental arithmetic'if they ever want to find a job involving a cash registre, so many people that perform algorithms as their daily business, don't realize that that's the case.
It gives me no small comfort that, as a writer with a BA and MA in English literature, I am nigh-on impervious to dementia. Thanks for that, Dr. Fry 😁 Don't correct me I'm enjoying the fallacy.
Intelligent, well spoken, and an absolute babe. She could read furniture assembly instructions and you'd watch the video to the very end with a smile on your face.
Although it would be nice to have a government run by a committee of both humans and computer representatives, which can filter through sound bites and and hyperbole, and runs the country logically.
That was fantastic, thank you Hannah Fry, and RI, you gave me some very intersting insights (like that people and machines should complement each others strengths and weaknesses).
Before watching the video, I'll say this: Computers shouldn't necessarily _run_ the world, but they _should_ make all the _default_ decisions, with humans retaining the option to override those decisions as appropriate. However, the override should have to be on an individual basis, so it's too much of a pain in the ass to override the computers all the time. Statistics don't lie -- only statisticians do.
No, computers are there to help humans, not to take over decision making. There is no 100% guarantee that a programmer can make that there won't be an exploit in the programming that could result in humans being unable to revert decisions made by computers. Just look at how many companies get hit by malware attacks each year, every program can be exploited, safety can't be guaranteed, no matter how much effort you put in.
Binging the RI UA-cam channel, video starts to roll, "wait, I know that voice..." It's only blahdy the funniest half of Curious Cases! I love the podcast, listen to it all the time at work :D great to see you're as edutaining on the screen as you are in audio 😁👍
About the Japanese people trusting their GPS... what about the recent jetliner crashes, where the pilots didn't know how to override the computer? Some one trusted the computer over the humans.
Talking about the 737-MAX situation? In that case I do think that the idea of that system is great, with the changed configuration the aircraft is prone to stall* under certain circumstances pilots might miss out on or aren't used to on the entire 737 family. Proper detection of that and overriding the pilot to avoid the stall and potentially falling out of the sky would be good. Now obviously a lot went wrong, the implementation was quite flawed and the training/certification guidelines/requirements from Boeing about the new feature was less than ideal. It should've been communicated better, such a scenario should've had at least one simulator training session for pilots flying that to experience it and go through the motions of manually overriding it. A bit baffling to me how such a rather essential new feature didn't get that kind of attention. Most concerning though is how they managed to implement it using only one of the angle of attack sensors creating a single point of failure. Of all industries the commercial airline sector is about redundancy everywhere with hardware and checklist procedures. Fairly certain a second angle of attack sensor is on that plane what it should've been designed to do is to turn the system off and give a warning as soon as a significant discrepancy shows up between those to sensors. I'm a bit afraid they were mandated by regulations to put a system in place to prevents accidents due to stalling, jammed it in there somewhere late along the design process. Did didn't test and evaluated it deeply enough due to time pressure to bring the planes to market. I sometimes do wonder what the damage is by now, obviously the two crashed planes but the financial and logistical impact on airlines with grounded planes is very significant. Boeing itself suffered a lot too in that department but that somehow bothers me less they are somewhat at the "too big to fail" state and it was their own fault to end up in the situation. I do hope though it won't affect too many regular Boeing employees and that it helps them to be more careful in the future in their corporate culture. *: Stall here is the fluid dynamics sense of the word. Basically there's an angle of attack referred to as the critical angle of attack at which the lifting potential of the wing is optimal. Going beyond that angle of attack will result in too much airflow separation behind the wing causing reduced lift force. It'll typically result in at least some loss of altitude but it could be severe and if not corrected control can be lost of the airplane. Having this happen in general is dangerous and at low altitude or busy airspace even more so.
The Bach comparison was interesting. I guessed number 2 not because I have any great knowledge of Bach, but because it was more intricate and sounded more like the Bach I love. (yes, I'm a fan of Bach music but with no education on it). But my main point here is that they sounded totally different. I honestly didn't know whether i had guessed right but I'm glad the real Bach really shined through! :-)
Funny thing about algorythms is that is costs massive computer power to refine the algorythm, but when it finally is, it costs way way less computer power to actually use it for useful purposes. Brilliant really. PS Hannah really has a passion for her work, go girl :)
I got the Bach wrong because I reasoned that #2 more closely matched Bach's pretty structured rules for composition. I thought that one was more likely to be algorithmic.
It's a weird paradox but we think of Bach as the epitome of human composer because of the intricate structures he uses, but these structures are the very thing that makes his work exceptionally guessable for a trained algorithm!
I kinda wish Hannah had gone into more detail to explain why the woman that believed algorithms to be unbiased was actually completely incorrect. The data you feed an algorithm can introduce extreme bias. Imagine say we had one of those judgebots, and the data you served it was from a place with systemic racism problems, so it's trained on those pre-existing cases, it's going to inherit those same problems, not get rid of them.
love the idea of the economy being run by computers not to sure about the rest of the world tho, perhaps humans being supported by computers but computers have one massive downside, they are programmed by humans. Edit: i posted this before watching video to see if my ideas were inline with the talk givers, glad to see we are pretty much in sync with each other.
My brother in law is a salvage truck driver and at one point I joked about my satnav going haywire and saying turn right here! And I look to the right and I see a marsh without any roads. He said, kind of worryingly: "I drag 4 to 5 people a year (mainly obedient women), out of fields or medians and even once out of a canal and their defence was, the satnav told me to do it." My reaction was: "maybe these people should not be allowed to procreate for the sake of humanity! I don't even trust my calculator I always do a rough estimate what it would need to be. As us engineers are taught." He said: "and us truck drivers, there's lots of tech to make driving safe. But we are taught to never rely on it and always check for ourselves too. And every year we need to retake the test and if we don't do it correctly our operational license is revoked until we get right."
Pigeons were not only used as messengers in WW1 and WW2, but they were even trained to guide missiles in WW2. Sadly the latter was a "suicide" mission though. :(
I'd like to know if the performers who played/sang the 2 "Bach styled" pieces knew which was which? While I'm no great fan of Bach (I find his work a bit too stiff for my tastes), the second piece had something in it that made you want to hear the rest while the first piece felt more like a TV commercial.......it was beautiful, but I couldn't have cared less if it continued. My question is, would that have been due to the actual piece or the performance? Beyond that, this was an incredibly interesting lecture by an intelligent and beautiful woman. I found this fascinating.
I could tell that option 1 was not Bach because the voice leading was really bad, especially in the middle voices. The second one sounded a little bit like it had computer-synthesized instruments as a ploy to make it sound like it wasn't Bach, but the voices themselves had much better voice leading. It kind of reminds me of the Google doodle that came out a couple weeks ago - any example you had the doodle do had bad voice leading, because it is very difficult for a computer to figure that out.
Who would ever think that a PHD in Mathematics would have a biography that spans Pigeon peaking pictures to the mathematical probability of falling in love. Dr Fry is an amazing orator and a pleasure to listen to. In addition, she has a wonderful sense of humor that even us Muggles understand :-)
Her version of the future is exactly what has happened in chess. Chess programs can beat any human. Did humans stop playing chess? No. Top players now use chess programs as part of their prep, computers are even used in tournaments for the live commentary. The chess programs became an extension of players' brains in a sense - not a replacement. Moreover, humans became good at intuiting in what situations computer analysis is useful for a human and when it'd only be useful for a computer player. Would a GM say "I didn't play the computer move, I must be crap"?
I don't think the analogy works very well. Humans play chess for entertainment purposes. The goal is to provide entertainment for the spectators, and the spectators prefers watching humans play. If the goal was merely to win the game, then there wouldn't be any human chess masters any more, because the engines are just so much better. Serious stuff, like diagnosing cancer, is not a matter of entertainment. We will stop using human diagnosticians as soon as computers do it better. What Hannah is talking about is a situation where computers are better at some aspects of a given task, but humans are better at other aspects. In terms of chess, this might be a situation where, for instance, chess engines are better at openings and humans are better at the mid-game.
At 16:00 or so, she asked the audience if they would prefer A.I. or human factor to judge them as a criminal in court. My opinion would be the human. The reason is that when you're up there all nerve wracked etc. the worst thing that can occur to you usually, is the honorable person to 'throw the book' at you. Thus, I'd risk my judgement being affected by how his day was going, or the previous defendant's insulting demeanor. Likewise, he could have got lucky last night, or all the previous cases were straight forward with less heinous defendants. I feel if the book is recommended to the Machine, the book is what you get.
"An algorithm would be less biased" says the audience member. Cough Facebook Cough. Google etc it's how the algorithm's programmed that matters. Biased people program biased computers.
*Less* biased. You don't get the accidental, undesirable biases to anywhere near the same degree, and it's simpler to fix when found. Not much you can do about deliberate bias, but at least it removes an excuse from such companies.
there are all kinds of amazing women in tech. lisa su is my favorite person in technology today. don't let them tell you you're being held back. don't let them take away your confidence
Option 2 was clearly the one that was most likely composed by a real composer. The first option completely lacked over arching structure while the other one was written the way a good composer would have. It used chord progressions and melody in such ways that made sense.
I know nothing about music especially classical music. So I don't know what anything you said about chord progression and melodies means. But yet I could tell confidently which was the computer and which was human.
Not always. Computers are doing a decent job of generating algorithms already. In the simplest case, they can randomly generate multiple models, hundreds or thousands of them, and choose the ones that closest match a goal. To the video, it's imperative that we understand the limitations of this process and not treat it as a flawless system.
What a brilliant public speaker!
public intellectual
She sometimes increases her talking speed too much and slightly stumbles on her own words but overall very enjoyable speaker yes
And I so very much envy her husband!
also hot af
I've seen several of her presentations and thoroughly enjoyed every one.
Poor Greg the Pigeon. He only went into medical school because his parents pressured him. His real dream was to become the first pigeon composer.
Flint Read “Dammit, Greg! If you can’t quit cooing in class, I’m going to have to ask you to leave. Unlike SOME pigeons here, the rest of the class are interested in learning oncology.”
:(
True, but Greg's dad made a huge donation to the hospital and Greg now runs the pathology department.
Glenn Pollock Sadly, it’s only a matter of time before Greg gets busted self-proscribing opioids and loses his medical license. Despite the high-paying job, trendy nest in Chelsea and young Frillback Pigeon trophy wife, Greg is one sad bird. How many times has his assistant had to pour him into a cab and send him home after too many martinis at lunch, I wonder?
In 2019 we know his parents bribed him and the teachers to get into medical school
She is not only a good mathematician, but also a brilliant speaker. I watch her explaining mathematics all the time and she always connects with her listeners.
Damn pigeons don't even have any student debt to worry about either.
Just one of the advantages of not using money; something which is advocated by The Venus Project and something which I believe we should adopt.
JustOneAsbesto, they are stressed by the need to get food. But that is a more basic need than how to pay your student loan.
@@goromaster10 only stupid people (including ones with high IQs) ring up massive student debt. Smart people go to state schools, get jobs and study practical majors.
The one doing it wrong probably got out of the cage a lot sooner. Some deep learning right there!
@@gabydewilde Finally a response from someone who actually knows what a joke is.
Should Computers Run the World? - with Hannah Fry -> Nervous.
Should Computers Run the World with Hannah Fry? -> Yes. :)
simp
@@radioanon4535 the meme is dead
Should Hannah Fry Computers 'run the World?
no :(
OMG, the final ad, right after Hannah starts to explain the dangers of algorithms... Husqvarna self driving electric lawnmower.
Well played AdSense, well played.
So a recommendation algorithm recognized that a biological algorithm was learning about algorithms and responded with a suggestion for a machine with an algorithm?
@Phi6er A lawnmower that eats grass for energy. That would be amazing.
Ad sense works on your past behavior within Googles ecosystem, so you may have been doing research on stuff or must have been reading mail in Gmail just before you stumbled into this talk. The content of this video influences it only slightly compared to all the activity you had done before this video. Big brother is watching!
I got an add for painkillers at the end.
As long as it's not a self driven Husqvarna chainsaw, that might classify you as a tree and try to cut you down
4:10 Great, now we'll have to deal with the threat AI cyborgs proclaiming, "I'll be Bach".
nice one
Algorhythms
I see what you did there…
A Classical compusition for an orchestra of strings.
No no, it is:
I'll be batch
As a longtime software developer, I don't trust algorithms to be particularly unbiased or better than humans at most tasks. Software is not magic. It's written by and trained by flawed humans. The best description I can give is this: most algorithms are idiot savants. The software may be good at some things, but it doesn't have a model of the world anywhere near the sophistication and subtlety of your average human being. Trusting a decision because "the computer said so" is a rather naive way of thinking.
Hannah Fry is a great communicator. I enjoy and learn from everything she does.
Also I’m completely in crush with her. Le sigh.
she’s party to untold counts of menticide & democide
Her fiance is the luckiest person on earth. I love her voice, her face, her brain, and that spark that brings it all together.
Had a huge crush on her as a teen
Why dont you send her a gift?
She is a really talented speaker! Such a calm but enthusiastic voice😍
Hannah Fry voice alone is completely engaging. When she combines it with knowledge, charisma, and humor she absolutely inspires.
I really enjoyed this talk, she speaks so well, you can really hear her passion for the subject matter.
Loosely rehearsed and a script. Talking about a subject she knows very well. How is that impressive lol. That is expected rofl.
He said he enjoyed the talk. Where are you seeing "impressive"? And being knowledgeable about a field and being a good speaker don't necessarily go hand in hand.
@@Ken.- I know what he wrote, i looked through the comments and found 5+ of these saying the same thing. Wow so impressive ect. Knowledgeable part is far more impressive than speaking but the thing that goes everyones head is that THIS IS EXPECTED OF THEM. THEY ARE SPEAKING IN AN AUDITORIUM. And thus it cant be considered wow impressive much love. The bar for that is set much higher. This is not about gender either. I have not watched a single talk in this channel and found it boring. Meaning all of them are more than good enough to convey the subject at hand. This leaves me to believe theres clear bias for a pretty lady talking about complex things somehow being more impressive because she is a beautiful smart woman. I wonder who writes these comments. Cant be men because that thought never came into my head. Must be women power tripping or something lol.
@I'm the captain now I have no idea what it is you're trying to convey. Too many basic errors for me to decipher it. Have a nice day.
@@Ken.- want tl dr version? Her talk is no more impressive than any other ive seen. People need to calm down this isnt an entertainment medium.
I first saw and heard Dr. Fry in a televison production titled, "Magic Numbers" on the nature and history of mathematics. I was thrilled - especially when she got to quantum mechanics. She is a wonderful, articulate and concise teacher. I find her balanced approach to algorithms and AI comforting. It was heartening to hear Dr. Fry say that the music in the style of Bach was not actually composed music. Programmers are working what what they know about Bach's music. It took the genius of Bach - the person - to create this type of music in the first place. And yet, it can, "...fool a roomful of people." We as the general public are not able to differentiate between content created by advanced AI and the real thing. This has serious implications for societies. Dr. Fry is so right in saying that we need to think about how we use the algorithms and how humans fit into all of this. She is right when she says, "You can't rely on people." One reason you can't rely on people because intelligence is unpredictible. People do not alwyas act in their own best interests. Artificial intelligence will be just as potentially unpredictible. This is a qulaitatively different creation than anything humankind has ever created before.
I fear that there are many others who do not share Dr. Fry's balanced viewpoints. Research is roaring ahead from the current levels of AI (which have already been used to cause havoc through spreading misinformation and feeding people inciting content to further keep them engaged online) toward Artificial General Intelligence (AGI) which will be much more powerful and have sentience.
There is theoretical potential for later, advanced AGI to be using us instead of us using it. Consider how humanity has treated and used "less intelligent animals" through hisotry. AI will potentially be able to self-replicate, creating mutliple copies, thereby making itself safe from being shut down. It will be able to self-upgrade, self de-bug those upgrades and each time learn how to better self-upgrade the next time. With upgrades, it's architecture will change. Those who created it in the first place may not understand what it has become, making it all the more dangerous. Interfaced with our infrastructures, it will potentially have incredible power and control. It's goals - satisfy what it has been programmed to carry out. And if it perceives people as either unnecessary or worse, a threat to it - what then? What consequences lie in store for us when we have become the "less intelligent animals" from the pont of view of sentient AI? There must be more public education and debate on all of this. I am not optimistic.
What's interesting is, as a trained musician, it seemed much more obvious regarding your example who was who, and I think it comes down to intent. With bachs piece I can tell he is trying to push and pull different things into your attention span because he is aware it is limited and not everyone he is composing for will hear the intricacies in the chords. The computer has no such audience it is playing to so it takes the most complex pieces and rules and throws them at you all at once where the ideas become muddled, it's playing to the rules given, if you further limit those rules I'd imagine it becomes much much harder to figure it out. What each is trying to do seems to be the differentiating factor here. Bach is trying to communicate to the listener. The computer is trying to come up with every permutation within the rules as correct answers. You might be able to trick the audience even better by selecting one of Bach's earlier works.
Algorithmic composition is one of the greatest failures of computer science, so far. We still seem to have have little to no understanding of what exactly determines the quality of human music of any era. We can ask our greatest composers and musicians how they are composing and performing and many of them can explain in great detail what it is that makes their music "special". Our conductors can tell every instrumentalist in an orchestra how to play their part so that the entire orchestra comes together in a way that is far more than the sum of its parts. What we have not managed is to transform these explanations of how music is being created into a set of rules that can compose as much as a simple folk song or that can play a straight forward blues progression in a "natural" way. I find the depth of this failure utterly fascinating.
I'm fond of saying art is self-expression: no self, no self-expression - and the reason we express ourselves is to communicate with other humans. Bach knows people; the algo only knows rules.
I'm a simple girl. I see Hanna Fry, and I click.
I'm not a simple girl. I see Hanna Fry, and I curl in to a fetal position, tumble over and wishes she could hold me in her arms.
These replies are pure gold :'D
Just like Hannah Fry
I'm a simple guy. I see a copypaste comment, I dislike.
I'm a simp. I see a girl, and I click.
A flock of pigeons are as accurate as a "fully fledged" pathologist... I think everyone missed that one,, including Hannah.
That or she is a witty genius lol
😂
What does it mean?
@@tantiwahopak101 describing something or someone as "full(y)-fledged" is a metaphor that comes from the life cycle of birds. A fully fledged bird is an adult one which has shed the down it was born with, and grown its adult feathers. A full fledged pathologist is a professional one who has completed his/her training
I got that too and immediately started worrying if all humour had left the Earth.
I have barely every seen such a sympathic speaker that just drags you along the topic. Very charismatic, excellent lecture.
On the gorilla thing: human brains are actually very good at pattern recognition, so I'd say that human doctors are well equipped for spotting what's wrong on x-ray images.
Missing the gorilla is just a case of attention bias. Gorillas normally don't show up on x-ray images so the shape is not registered by the brain.
Very good talk, as always with Hannah Fry. But this delightful voice! Gosh, that's really something to hear, for a non-native English speaker 😌
this Royal Institution is like TED Talks but more fun !
The Royal Institution has been giving TED talks since 1825
I would hope that The Royal Institution does a better job at vetting their speakers than TED.
TED talks are commercial and PC biased talks with a lot less intelectual value.
TED Talk started great but then devolved into a hot mush of interesting speakers and snake oil peddlers.
@@PapiJack With TED talks proper being interesting speakers and TEDx talks being snake oil peddlers, or has the legitimate article also degraded?
This talk popped up on my algorithm, which is currently finishing my sentence before I compose it. Hannah speaks to this topic with more salience and forward thinking solutions than I've heard even here in 2024. TY Ms. Fry! 🌎🌍💜
It doesn't take a very sophisticated algorithm to understand that I want it to serve me every last Hannah Fry video on UA-cam though...
It looks like you are trying to avoid videos with "Hannah Fry" in them. Would you like help?
☐ Yes Please ☐ Yes ☐ No (Yes)
Your avatar looks like sheep jumping a wooden fence on a lush green field.
*you* interact with this video
*youtube algorithm* would you like some minecraft video from the year 2010?
Why can't at have more schoolteachers like this lady? I bet everybody would learn so much more. I could listen to her talk about any subject and I'd be fascinated.
She has great charisma, really enjoyed this!
This has been so educational and Dr. Hannah is a brilliant public speaker
Great talk! Very interesting and very funny. More Hannah Fry please!
Hannah Fry is one of the most intelligent, wittiest, and most beautiful person I have ever seen.
I find the 50/50 Bach result surprising. #2 is so clearly Bach because it contains a cohesive emotion which is something a computer cannot replicate well. I don't know if I recognize that emotional aspect because I'm musically trained, though. Do 50% people just not experience music on an emotional level? My face and head literally tingled when the real Bach played.
So I have to admit I am not quite sure if I understand your point. How was the first one not cohesive? What were the indicators that gave it away for you?
I am sure that there is tons of music written by humans that is not emotionally touching.
I would also assume that different humans react differently to music. I might get emotionally carried away by some blast-punk and feel nothing while hearing bach. For you it might be the contrary.
@@niklasb6849 Yes, the first one is technically cohesive, but it doesn’t sound like a human wrote it because it lacks emotionality that only a human can give it. There is no “story” to it, it just sounds like Bach. The whole magic of writing music of any genre is to convey emotional stories. Whether or not you get emotionally carried away by it is a separate experience and will vary by personal preference. I just get head tingles when I hear music I like, whether that’s Bach, punk, hard rock, or a pop song.
Interesting - I experienced more of an emotional reaction to the first piece. Just kidding, it's not interesting. That's just how human emotions work - they're completely subjective and contextual.
I’m a new Hannah fry fan. I saw the Berlin talk before this one and saw how uncomfortable she got in the Q and A. She dealt with it really well and was very interesting to hear Hannah talk about that in this talk. …People… make More people like Hannah please.
Hannah Fry could narrate paint drying and I would still be fascinated :x
I just stumbled across Hannah and her videos. I’m in LOVE! I could listen to her all day every day!!
Another question - should the people who control the computing platforms run the world?
and vice versa
It's all academic - there is zero chance that democracies will put a computer in charge. People don't trust politicians, let alone a computer program that could be infiltrated and biased.
Stephen Sinclair it’s actually the same question
The talk is only four years old at this point, but with all the recent developments in AI it feels more like 10 to 15 years old.
'Man plus machine' seems to be the winning strategy for the foreseeable future. Enlightening talk!
Transhumanists would agree, in the area of technological augmentation.
But but but... what about the pigeons?
Let's not discriminate against crows, either!
define the procedure "plus" first
"man plus machine" was theorised to be why we got the Boeing crashes (I didn't keep up)
If there's a silver bullet, this doesn't seem to be it (though I'm all for improvements where we can get them)
Augmentation seems the best possibility in general it takes advantage of human insight and machine rapid processing. In specific domains machines will be vastly superior. Currently games like chess/go and even real time strategy games -- AI is way ahead.
What made it clear for me, was the random high note in the Bach work. I heard once, I think, that they used to put random notes that were out of place so that the audience didn’t fall sleep lol.
Useless time stamps:
6:34 she says, algorithm
17:09 she goes jogging
30:32 brainz
cheating the youtube algorithm I see...
Slightly arousing.
"Useless" is a matter of opinion :-)
Hannah Fry is a British version of a very smart woman at my work, who I have always had a soft spot for. I could listen to her voice for a very long time!
Absolutely brilliant and fascinating... her knowledge and ability to communicate it in a simple and concise manner is fantastic.
How delightful to watch this after reading the book "Coders", esp. with that chapter on biases inherent in data sets models may be trained on.
People who put a lot of trust in computers don't have a lot of experience with programmers.
Still better than a biased and selfish human.
The whole world has put their faith in programmers because computers make the world go around.
tarquin161234 I agree, humans should change their mentality. But the whole problem is money!
Besides programmers aren't all the same so you want to trust a programmer, who's a human
But you say humans are selfish and biased? Ever heard of malware? Or ransomware?
Or did you think a computer virus is a biological virus that can infect computers? 😂 j/k
@@sendraw4088 Yeah you're right. Computer programs are no different to people because they were made by people. Clearly a human/democratic check is always going to be needed, so the recommendation of the video is a very good one (for computers and algorithms only to be used as tools by people.)
@@tarquin161234 You do realize that the computer's instructions are going to be written by one of those biased and selfish humans, right?
Terrible argument. The Republic was written by a "biased and selfish" human, but his instructions, if they were followed, would not lead to biased and selfish robot politicians. It's akin to saying railway logistics are inherently flawed because they were designed by humans, completely ignoring the fact that humans are only biased and selfish in very specific areas, and in those areas computers would obviously do better.
I'm a musician and I had no trouble picking which piece was written by Bach. I can't tell you why, exactly. I can only say that there was something stilted and mechanical about option #1. It had no life. Now, that could be because the arrangement was too simple or because the orchestra and choir didn't feel it the same way, I don't know. But in the end, it just didn't touch my soul. The Bach piece did.
1:01 - "fully-fledged pathologist..."
Was this pun intended? If so, it's the greatest ever uttered.
Pls explain the joke, i dont want to fake laugh when i hear it in future
fledged means covered with feathers
Relatives of my wife got stuck on a dead end street at a river where once was a military bridge. They were driving by car and completely relying on their navigation system. The algorithm for route planning was not the problem here, but the outdated map. The funny thing is, they got stuck several times at the exact same place.
thank you for this amazing lesson. Although I am not a native english speaker I could follow about 90%, thanks to Hannah.
Almost by the 30th minute I guessed the conclusion of the talk would be synwrgy between machine and human. I remember being part of a debate where I suggested the same. Very inspiring lecture Dr Fry. Thanks.
I love Hannah Fry's videos. She's very informative, interesting, and easy to understand.
It's kind of interesting that the talk at re:publica in Berlin might have been one of the worst moment in her professional life at that moment, but in hindsight the audience in Berlin helped her to grow in a way that otherwise might have took much more time if it ever would have happened.
Thought leader and public intellectual Hannah Fry :)
something distinctly Orwellian about the term "thought leader"............
@@freezatron then don't visit american edutainment conferences :)
your thirst is showing
aidan levy, do you think I need to go to the Ministry of Truth for some political correction ? . & why would my saying that preclude me from going to a conference ? . are you a thought leader as well ?
I'm not very good at this double think malarky, can you tell ? :)
I am a pathologist, I think the problem here is that the diagnosis of breast cancer by histopathology slides is really an easy task, any medical student who trained in a pathology lab, can diagnose breast cancer in 1-2 weeks of training. I am really interested to see the results in really difficult cases, in conditions where someone have to judge using different resources together (history and physical examination , Imaging data, Lab data and histopathology slides) the results will be different.
I'm a simple person, I see Hannah, I press like
I really is just that simple, isn't it?
that's not simplicity. it's called thirst.
When I get my own personal AI, and can customize it, I want it to speak exactly like Hanna Fry. Her voice, timbre, diction, everything... so easy on the ears.
Hannah, you're everywhere.
I mean, mostly just that one island, but like... _virtually_ everywhere.
OK, I have a question. It seems to me that based on combining these factors:
1. 32:24 - "now this is the stuff that algorithms are amazing at" - being sensitive;
2. 33:15 - "humans, when it comes to being sensitive, we're rubbish, right, we're totally rubbish at this";
3. 34:18 - "but specificity, being specific, that's like our [humans'] super-power."
... you (Hannah) then present(s) the idea of having sensitivity done by algorithms, and specificity done by humans -- 35:15: "this is the version of the future that I'm really hoping for". Yes! I think this is absolutely a brilliant way to approach things. Because:
When thinking about this stuff, I think about spam (or unsolicited commercial email, or UCE) detection, and I think about how Google's Gmail product has done a wonderful job of keeping UCE out of my inbox... and a fairly poor job of keeping legitimate e-mails (sometimes solicited commercial e-mail, sometimes just stuff from friends and things) out of my Spam folder, and _in_ my Inbox. I get alarming numbers of false-positives (stuff flagged as spam, that's really not), and very few false-negatives (stuff not flagged as spam, even though it actually is).
And yet, I think another thing that needs doing here is to try to work on getting the _algorithms_ to be better at avoiding false positives. Because realistically, while I think _I_ am good at finding those false positives in my Spam folder, I'm not so sure that _anyone_ would be - a lot of the ways that I detect them are taking specific knowledge about my own e-mail activities and applying them to find messages that ended up in Spam but shouldn't have. So I think that if, say, Google hired a bunch of folks to go through low-confidence flagged-as-spam messages and pick out the ones they thought were false positives, those folks probably still wouldn't find all that many of them (I'm guessing), compared to what I can do on my own inbox. To the point where I suspect that if I had the proper tools to do so, I think I could probably assist in training an algorithm to detect the false positives much better than other humans could... and perhaps, ultimately, better than I myself could.
So, do you (Hannah, if you happen to read this, and/or anyone else who's bothering to read all this and likes to think about such things) have any thoughts on what sorts of things can be done to increase the specificity of algorithms?
One thought that I had while writing the above paragraph (two above this one) is to have an algorithm that (for example, in the context of Spam) is trying to detect not spam amongst all e-mails, but false-positives amongst messages flagged as spam. In other words, and in the general case: could we use the sensitivity of a second algorithm to give specificity to a first algorithm? So two separate algorithms are used for a single problem, one to give sensitivity, and the other (through sensitivity for false positives) giving specificity?
Whether or not that might work, what other things might we be able to do to get specificity to be better?
Because I'm of the opinion that if we're going to be using algorithms for this stuff at all, we'd ideally like there to be a way to have both dials pretty high on the algorithmic side of things, at least for many tasks. For some problems, having humans deal with specificity while algorithms do the sensitivity might be fine. For others, I don't think it would be practical, at least, and perhaps not even feasible.
there are what are called open source software where you can see the algorithm used, only open source should be acceptable when making these kind of decisions.
Even a human must write a justification for a certain decision, the reasoning behind a decision should always be public
all these problems can be fixed, they do not detract anything from the fact that computers and computing are powerful and preferable tools for justice and public administration
they are simply not used enough and used properly.
Most of public administration could be totally automated by information technology, mistakes avoided, thousands of wasted hours saved.
A citizen should not need to make declarations every year, should not need to queue in government offices to fix and update information that is already available to the state.
Because ALL the information is ALREADY available to the state via other mechanisms and registries. List of events like marriages, births, deaths, divorces, property purchases and sale, employment and unemployment, minors in charge etc.
In fact ALL is already available to the state but the current system is based on inefficiency and authoritarianism where the citizen is made responsible for maintaining data the government has already even though the citizen pays millions in taxes for a services that is not provided!
The state goes to the extent of calling the citizen a criminal or fining them for not doing a job the citizen is already paying the state to do.
I would rather use processing power properly and have the state provide me with the services I pay for instead that giving the job to a bunch of incompetent politicians, government officials and judges that demand to run my life and fine me for heir mistakes too.
I wouldn't go as far as to say "all these problems can be fixed" but sure, most of them can. And open-source is definately the way to go with tools as important as these as well as anything made with public money (With the exception of national security). Also an important part of making them open source is a generous bug bounty to actually make people report bugs instead of just exploiting them. There should also be a (properly, unlike most datasets) anonymized dataset of the decisions made by the algorithm to see if they seem valid and if the algorithm needs tweaking.
@@Sibula if we are thinking that we can cure aging and fix almost any disease, we can certainly fix man made algorithms.
and there is a very important difference to how I would use computing power to how she and the state is using computing power.
She worked to put computing power at the service of the state to control people.
I would put computing power at the service of the citizens to control the state.
Computing power to control the state can NEVER fall into authoritarianism while concentrating power into the hands of the state it's sure that sooner or later it will fall into authoritarianism.
@@canemcave I don't really know how you're planning to "put computing power at the service of the citizens to control the state". Someone would have to develop and maintain it, giving that person or organization power. And what would it actually control? But there are many ways you can use algorithms (eg. machine learning) Where it doesn't really give any power to anyone, mostly just makes everyones lives easier. Diagnosing illnesses is one such application.
@@Sibula machine learning is just one type of algorithms, we are using algorithms every day for all sort of things, from industrial process automation to document editing.
How can you give power to the citizens to control the state?
first by making proper use of resources already available to the state (and therefore proper use of the money you paid in taxes) and putting the responsibility for managing and maintaining that information not on you, the citizen but on the state you are employing for that purpose!
second by making algorithms public and requiring traceable explanation and reasoning compulsory for any government decision and thus making government accountable and transparent
third by applying same rules impartially, without human interference you remove abuse of power and arbitrary judgements and again introduce traceability and accountability in the system.
Information systems should record every action and decision made by any government employee and reports should be immediately available.
Databases could be made to exchange information only requiring your express permission for the data to be exchanged.
trivial example a national and a foreigner couple marry in a country. They should only present their passports, the local authorities should register the act, and transmit it to the relevant authorities.
For the foreign passport the event should be transmitter to the passport original consulate or embassy which would than have the responsibility to register it in the relevant registry. Simple, which country does that? No, one! it's a miracle if the information is recorded and transmitted within the country!
this type of example can be extended to most government services, they are just not created to serve you, they are created to exploit you.
There are many other things that can be easily done by information systems and we do not even need to employ AI systems. From transport systems which should operate Nationwide as credit cards do to sentencing which should be handed down according to a set of legislated rules and averages, irrespective of sex color or religion, information systems could be put to good use to make life easier and fairer, but they are not.
Instead the state spends your money trying to control you.
it's the time to change this state of affair, government are there as a service to the citizens, not to fill their pockets.
@@Sibula the purpose of open source algorithms is exactly that, removing power from the company implementing the system.
An algorithm that, for instance, assigns sentencing, should be open source and any modifications to it should be evaluated and approved by a public college of IT and legal experts appointed for the specific purpose of creating or modifying the algorithm.
an algorithm for sentencing does not require AI, it needs to embed the various legal rules properly, it could be as trivial as implementing an expression such as: ((mean period for offense + mean prison population for offense) / 2) + ( Sum period for aggravating circumstances - Sum period for extenuating
circumstances )
Do you think a judge on the spot can elaborate all the rules of law properly when a team of scientist, engineers and legal experts cannot properly even defy them as an algorithm?
The judge will brush aside most of the rules and go with what intuition or mood will tell within a set of constraints he or she will have developed from experience.
As a guy who started programming computers before she was born, about 45 years ago, I don't trust computers unless there's a human able to override their decisions. Computers are our *TOOLS* , and can never be capable masters.
Never have, even though I was more-than-passably good. "To err is human; to really f**** it up takes a computer." My real focus was displaying the output decisions in a way that the human could accept or reject key pieces easily. My goal, if there was some bit of AI in there, was to indicate the "thinking" leading up to the final answer. I think that's nice to have; in some arenas, vital.
Thanks The Royal Institution.☺ Thanks Hannah Fry.☺ "50-50 then. Basically guessing at random. And you laughed at that pigeon...."😀
a great example for an algorithm would be the way change is given from a cash register. Only problem: many academics start to do mental arithmetic, when ask to look after a cash register. and teachers tell pupils they have to be good at 'mental arithmetic'if they ever want to find a job involving a cash registre, so many people that perform algorithms as their daily business, don't realize that that's the case.
It gives me no small comfort that, as a writer with a BA and MA in English literature, I am nigh-on impervious to dementia. Thanks for that, Dr. Fry 😁
Don't correct me I'm enjoying the fallacy.
Another thing that would really help people trust algorithms is open sourcing the code so people can actually see what is going on...
28:31 Nobody laughed at that brilliant joke because everyone in the room really had a favourite dataset.
Intelligent, well spoken, and an absolute babe. She could read furniture assembly instructions and you'd watch the video to the very end with a smile on your face.
My country is currently run by a "Garbage In Garbage Out" algorithm with a twitter account as the medium.
There's worse. Enjoy it while it lasts, whatever they replace it with likely won't be an improvement.
@@EggBastion likely the twitter account will cause it
Although it would be nice to have a government run by a committee of both humans and computer representatives, which can filter through sound bites and and hyperbole, and runs the country logically.
@@wyndhamcoffman8961 far too complicated for modern AI, there are some great tools though
She's amazing! Is she a TV presenter as well, she has a very compelling presentation style, hard to turn off
She has been in some documentaries, and on the numberphile channel
"a fully fledged pathologist" I see what you did there
That was fantastic, thank you Hannah Fry, and RI, you gave me some very intersting insights (like that people and machines should complement each others strengths and weaknesses).
"I love the idea of there being 'unconstitutional maths'.... " . That line got me laughing.
She’s such a good communicator !
To be entirely fair, though, _people_ trained the birds to diagnose breast cancer.
An excellent talk. This is the first time I've come across Hannah Fry. She's a really impressive presenter.
Before watching the video, I'll say this: Computers shouldn't necessarily _run_ the world, but they _should_ make all the _default_ decisions, with humans retaining the option to override those decisions as appropriate. However, the override should have to be on an individual basis, so it's too much of a pain in the ass to override the computers all the time.
Statistics don't lie -- only statisticians do.
No, computers are there to help humans, not to take over decision making.
There is no 100% guarantee that a programmer can make that there won't be an exploit in the programming that could result in humans being unable to revert decisions made by computers. Just look at how many companies get hit by malware attacks each year, every program can be exploited, safety can't be guaranteed, no matter how much effort you put in.
Educating + interesting + entertaining = Hannah Fry
Binging the RI UA-cam channel, video starts to roll, "wait, I know that voice..." It's only blahdy the funniest half of Curious Cases! I love the podcast, listen to it all the time at work :D great to see you're as edutaining on the screen as you are in audio 😁👍
Just binging on it now! lol Her BBC 2 docs are awesome too if you've seen those?
I started with this in the background, and it quickly gained forefront attention, bravo!
About the Japanese people trusting their GPS... what about the recent jetliner crashes, where the pilots didn't know how to override the computer? Some one trusted the computer over the humans.
Talking about the 737-MAX situation? In that case I do think that the idea of that system is great, with the changed configuration the aircraft is prone to stall* under certain circumstances pilots might miss out on or aren't used to on the entire 737 family. Proper detection of that and overriding the pilot to avoid the stall and potentially falling out of the sky would be good.
Now obviously a lot went wrong, the implementation was quite flawed and the training/certification guidelines/requirements from Boeing about the new feature was less than ideal. It should've been communicated better, such a scenario should've had at least one simulator training session for pilots flying that to experience it and go through the motions of manually overriding it. A bit baffling to me how such a rather essential new feature didn't get that kind of attention.
Most concerning though is how they managed to implement it using only one of the angle of attack sensors creating a single point of failure. Of all industries the commercial airline sector is about redundancy everywhere with hardware and checklist procedures. Fairly certain a second angle of attack sensor is on that plane what it should've been designed to do is to turn the system off and give a warning as soon as a significant discrepancy shows up between those to sensors.
I'm a bit afraid they were mandated by regulations to put a system in place to prevents accidents due to stalling, jammed it in there somewhere late along the design process. Did didn't test and evaluated it deeply enough due to time pressure to bring the planes to market. I sometimes do wonder what the damage is by now, obviously the two crashed planes but the financial and logistical impact on airlines with grounded planes is very significant. Boeing itself suffered a lot too in that department but that somehow bothers me less they are somewhat at the "too big to fail" state and it was their own fault to end up in the situation. I do hope though it won't affect too many regular Boeing employees and that it helps them to be more careful in the future in their corporate culture.
*: Stall here is the fluid dynamics sense of the word.
Basically there's an angle of attack referred to as the critical angle of attack at which the lifting potential of the wing is optimal. Going beyond that angle of attack will result in too much airflow separation behind the wing causing reduced lift force. It'll typically result in at least some loss of altitude but it could be severe and if not corrected control can be lost of the airplane. Having this happen in general is dangerous and at low altitude or busy airspace even more so.
The Bach comparison was interesting. I guessed number 2 not because I have any great knowledge of Bach, but because it was more intricate and sounded more like the Bach I love. (yes, I'm a fan of Bach music but with no education on it). But my main point here is that they sounded totally different. I honestly didn't know whether i had guessed right but I'm glad the real Bach really shined through! :-)
Funny thing about algorythms is that is costs massive computer power to refine the algorythm, but when it finally is, it costs way way less computer power to actually use it for useful purposes. Brilliant really.
PS Hannah really has a passion for her work, go girl :)
I got the Bach wrong because I reasoned that #2 more closely matched Bach's pretty structured rules for composition. I thought that one was more likely to be algorithmic.
It's a weird paradox but we think of Bach as the epitome of human composer because of the intricate structures he uses, but these structures are the very thing that makes his work exceptionally guessable for a trained algorithm!
@@andymcl92 That's what I thought as well. He liked structure. Also paradoxical: that affinity for structure made him a terrific improvisor.
@@erfling1 Oh without a doubt the best people at improv are those who know their scales!
Should Computers Run the World? Yes, but I get to program them.
Only the ones that I did not program first. :P
I kinda wish Hannah had gone into more detail to explain why the woman that believed algorithms to be unbiased was actually completely incorrect.
The data you feed an algorithm can introduce extreme bias. Imagine say we had one of those judgebots, and the data you served it was from a place with systemic racism problems, so it's trained on those pre-existing cases, it's going to inherit those same problems, not get rid of them.
love the idea of the economy being run by computers not to sure about the rest of the world tho, perhaps humans being supported by computers but computers have one massive downside, they are programmed by humans.
Edit: i posted this before watching video to see if my ideas were inline with the talk givers, glad to see we are pretty much in sync with each other.
My brother in law is a salvage truck driver and at one point I joked about my satnav going haywire and saying turn right here!
And I look to the right and I see a marsh without any roads.
He said, kind of worryingly: "I drag 4 to 5 people a year (mainly obedient women), out of fields or medians and even once out of a canal and their defence was, the satnav told me to do it."
My reaction was: "maybe these people should not be allowed to procreate for the sake of humanity! I don't even trust my calculator I always do a rough estimate what it would need to be. As us engineers are taught."
He said: "and us truck drivers, there's lots of tech to make driving safe. But we are taught to never rely on it and always check for ourselves too. And every year we need to retake the test and if we don't do it correctly our operational license is revoked until we get right."
i dont know what weapons ww3 will be fought with, but ww4 will be fought with pigeons!
Pigeons were not only used as messengers in WW1 and WW2, but they were even trained to guide missiles in WW2. Sadly the latter was a "suicide" mission though. :(
I think the invisible sheep have an edge.
ww3 will be fought with Nukes, no need for any ww4
I'd like to know if the performers who played/sang the 2 "Bach styled" pieces knew which was which? While I'm no great fan of Bach (I find his work a bit too stiff for my tastes), the second piece had something in it that made you want to hear the rest while the first piece felt more like a TV commercial.......it was beautiful, but I couldn't have cared less if it continued. My question is, would that have been due to the actual piece or the performance?
Beyond that, this was an incredibly interesting lecture by an intelligent and beautiful woman. I found this fascinating.
Oh Hannah fry? I'm here for it
Same
@@davidschmidt5507 Likewise.
I could tell that option 1 was not Bach because the voice leading was really bad, especially in the middle voices. The second one sounded a little bit like it had computer-synthesized instruments as a ploy to make it sound like it wasn't Bach, but the voices themselves had much better voice leading. It kind of reminds me of the Google doodle that came out a couple weeks ago - any example you had the doodle do had bad voice leading, because it is very difficult for a computer to figure that out.
Who would ever think that a PHD in Mathematics would have a biography that spans Pigeon peaking pictures to the mathematical probability of falling in love. Dr Fry is an amazing orator and a pleasure to listen to. In addition, she has a wonderful sense of humor that even us Muggles understand :-)
Her version of the future is exactly what has happened in chess. Chess programs can beat any human. Did humans stop playing chess? No. Top players now use chess programs as part of their prep, computers are even used in tournaments for the live commentary. The chess programs became an extension of players' brains in a sense - not a replacement. Moreover, humans became good at intuiting in what situations computer analysis is useful for a human and when it'd only be useful for a computer player. Would a GM say "I didn't play the computer move, I must be crap"?
That's what computers should be used for: a bike for the mind, as would say Alan Kay.
I don't think the analogy works very well. Humans play chess for entertainment purposes. The goal is to provide entertainment for the spectators, and the spectators prefers watching humans play. If the goal was merely to win the game, then there wouldn't be any human chess masters any more, because the engines are just so much better.
Serious stuff, like diagnosing cancer, is not a matter of entertainment. We will stop using human diagnosticians as soon as computers do it better.
What Hannah is talking about is a situation where computers are better at some aspects of a given task, but humans are better at other aspects. In terms of chess, this might be a situation where, for instance, chess engines are better at openings and humans are better at the mid-game.
At 16:00 or so, she asked the audience if they would prefer A.I. or human factor to judge them as a criminal in court. My opinion would be the human. The reason is that when you're up there all nerve wracked etc. the worst thing that can occur to you usually, is the honorable person to 'throw the book' at you. Thus, I'd risk my judgement being affected by how his day was going, or the previous defendant's insulting demeanor. Likewise, he could have got lucky last night, or all the previous cases were straight forward with less heinous defendants. I feel if the book is recommended to the Machine, the book is what you get.
It is funny, but in Russia most people prefer an algorithm to judge them. It is a last resort to get true justice in Russia.
Павел Макушев
So basically:
if( personjudged.LikesPutin() )
{verdict = "good boy!";
release(personjudged);}
else
{verdict = "bad boy!";
delete(personjudged);}
I love her enthusiasm
Ms. Fry is just wonderful. Very English, very Southern England. Thank you Hannah! Very cool!
This was a most refreshing talk.
"An algorithm would be less biased" says the audience member. Cough Facebook Cough. Google etc it's how the algorithm's programmed that matters. Biased people program biased computers.
Just code in the bias it's simple!
If(bias > 0){
bias = 0;
}
*Less* biased. You don't get the accidental, undesirable biases to anywhere near the same degree, and it's simpler to fix when found.
Not much you can do about deliberate bias, but at least it removes an excuse from such companies.
I called it as to which music was legit. First one has no flow. Second one is a coherent piece of music.
Hannah Fry make me more confident about being a woman.
there are all kinds of amazing women in tech. lisa su is my favorite person in technology today. don't let them tell you you're being held back. don't let them take away your confidence
Good for you!
Option 2 was clearly the one that was most likely composed by a real composer. The first option completely lacked over arching structure while the other one was written the way a good composer would have. It used chord progressions and melody in such ways that made sense.
I know nothing about music especially classical music. So I don't know what anything you said about chord progression and melodies means. But yet I could tell confidently which was the computer and which was human.
The answer is: 42
Such a soothing voice! Plus we agree on the opinion.
Can we please remember that humans write the algorithms.
Not always. Computers are doing a decent job of generating algorithms already. In the simplest case, they can randomly generate multiple models, hundreds or thousands of them, and choose the ones that closest match a goal. To the video, it's imperative that we understand the limitations of this process and not treat it as a flawless system.