Artificial Intelligence and the Next Industrial Revolution?

if you have an organization that's not using artificial intelligence it's going to go out of business and that's going to be a widespread job loss when some other company some other organization uses artificial intelligence and the reasons are not you know particularly evil or they're just it's a cost-based argument that if they can do things faster um cheaper I mean this we're we're in the context of faster cheaper standardization and and if another company can do things faster and cheaper incl

uding knowledge work I what we're talking about here is is largely in physical mechanical work what's interesting is my knowledge work here but people are going to lose their [Music] jobs good evening and welcome to the Charles River Museum of Industry and Innovation my name is Bob Perry and I'm the Museum's director um I'm very happy to welcome both an in-person audience and an online audience that's watching this live uh this event is being live streamed by the WGBH Forum Network and uh it als

o will be uh something that you can watch again later so if you missed something want clarification or you just loved it so much you can't get enough I know that's going to happen um so tonight's me talk is about artificial intelligence or Ai and of course AI is not just about research and writing as our speaker will well explain but in preparation for my role this evening I decided to take an intentional AI plunge myself for the very first time uh intentional because we're all actually using AI

uh all the time unintentionally you probably knew that but if you didn't you're going to learn a lot about that tonight I registered for Microsoft's recent acquisition open ai's chat GPT application which is now a wildly popular AI tool that helps people research and write I've heard the Wonder the superlatives the astonishment over chat GPT and with tonight's event on the schedule it was clearly time uh for me to give it a try and I too am fairly amazed I must say um I gave it four topics to r

esearch and write and what I'll read to you next other than the headers all of it is unedited product of chat GPT and honestly I think it's pretty good um to begin here is chat gpt's 100-word introduction to this Museum nestled along the scenic banks of the Charles River in walam Massachusetts the Charles River Museum of Industry and Innovation stands as a tribute to the Ingenuity and entrepreneurial spirit that fueled the Industrial Revolution housed within a historic 19th century textile mill

the museum offers visitors a captivating Journey Through Time exploring the evolution of technology and its profound impact on society though through immersive exhibits interactive displays and engaging programs the museum Chronicles the Innovations of yesty year inspiring a deeper appreciation for the Relentless pursuit of progress and the transformative power of human creativity in shaping the world we inhabit today wow really it's mostly true um I said please uh give me an introdu a 100-word

introduction to the Charles River Museum of Industry Innovation that was the almost immediate response uh you may know that uh and we are very grateful that our Mill talks are underwritten by a generous Grant from the L Institute here is chat gpt's 50-word introduction of the L Institute the L Institute a beacon of learning and Enlightenment in Boston Massachusetts has been fostering intellectual curiosity since its founding in 1836 dedicated to providing free public public lectures and educatio

nal programs it continues to enrich minds and Inspire lifelong learning serving as a Cornerstone of cultural and educational Excellence it's true and accurate and um I could not have written it better uh when you signed up for tonight's event you may have learned that our speaker co-hosts an excellent podcast Me Myself and AI here's chat gpt's 50-word introduction of that podcast Me Myself and AI is a groundbreaking podcast that delves into the profound impact of artificial intelligence on our l

ives and Society hosted by Leading experts in Ai and Technology each episode explores the ethical social and practical implications of AI offering insightful analysis and thought-provoking discussions for listeners worldwide and finally our speaker uh you were able to read a lengthy bio of Dr Sam ransbotham on our event page uh so I know that you've had full access to his remarkable background here's chat gpt's 50-word introduction of our speaker Sam ransbotham esteemed professor at Boston Colle

ge is a luminary in the field of information systems and digital Innovation with a wealth of expertise in business analytics and technology management his research illuminates the intersection of technology and strategy guiding organizations towards sustainable competitive advantage AG in the digital age wow um interesting no mention of AI but plenty of that to come please welcome Dr Sam ransbotham [Applause] hey y'all is this good all right thanks for coming and braving the the rain tonight um

this is a relatively small group I'm happy to talk about really whatever you want to talk about uh especially if it uses words like esteemed and luminary in it which you know I'm not sure if I've ever rated esteemed or luminary I think we should reemphasize those um I got a bunch of slides here of stuff that I thought you might be interested in but I didn't know what this audience would be so if you if your eyes start rolling like it's 8: a.m. and you're an undergrad then I'll just move faster a

nd if you're if you perk up in your seat we'll flow down and and delve a little bit um I'm about to start you've all had this experience of of calling your bank okay it's always wonderful experience uh you call your bank and immediately what you get is this person trying to stop you from doing what you want now um I like this picture because it look like a jeand and I like to say jeand because it makes me feel sophisticated and um luminary and Elite and all the all those good words um but they a

sk you things like what's your mother's made name what's your pet's name what's your etc etc all the things we keep on social media um but why does it ask you that stuff to annoy you I don't think that's their primary objective that may be a secondary accomplishment huh gatekeeping why do we want to gatekeep I mean wouldn't it be easier if we just let everybody call up and get straight to their money I'm worried about people like you calling up and pretending to me me and getting to my vast reso

ur sources right so we we endure this because we understand why it's happening but it's expensive it's expensive for you it's expensive for the organization for the bank right well let's think about what happens with artificial intelligence here and I'm going to say we're going to have a new process here with artificial intelligence and my first option is this option A and if you can tell I've replicated my tiny little jeand mostly so I can say jeand again um a whole bunch of times and this equi

valent is the chatbot that you're used to you call in and they say what can I help you with today and you say withdrawal you say transfer you say whatever it is you want and then because as you've noticed I have a beautiful southern accent they ask me again did you say reservations res I put apparently I put too many syllables in these words the agent Smith approach duplicate a bunch people like it people like to do this approach the banks like this approach why what's it cost them to do this ve

ry little they cost them while it's running it may take them some expense to get it set up but while it's running it costs electricity and they can answer the phone almost immediately it may not be helpful but they can answer it immediately so that's one approach and that's pretty attractive for people and I'll mention that as we go forward uh this transition from increasing the fixed cost to reduce the variable cost okay here's another option though a couple of other Banks HSBC Fidelity Barclay

s are trying a different approach which is as your call you get a human immediately and they say what can I help you with and then you say I'd like to transfer money while you're doing that they're checking your voice print and they're authenticating you through your prior voice interactions I think that's interesting now it's all opt in and in the you know requires requires them having a voice print of you from a prior interaction um but I like it because well they know my beautiful Southern vo

ice it's nice too because if you think about what's happened here it would be impossible to train every Bank customer service agent to recog niiz the voice of every customer right that's just not going to it's not going to work in scale but the machine can do it and the Machine can do it well and do it fast but we also have if you're paying attention here a separation of the task they're really were two things going on here there's the authentication process verifying you're who you say you are

and the what you want to do it turns out the machine is extremely good at the authentication part but it's extremely poor at the general what do you want to do part they can give you about five options if you don't say one of the five options that they've chosen then you have to say them again and repeat them over and over again um we've all been been through that um machine can do other things it's a little more subtle I don't want to spend too much time on this but it's pretty fascinating um i

f I claim to be calling to into my bank account and I say oh I'm in Boston I'm calling about my bank account they can check the phone line latency this is how long that phone line you know the you know the little Gap when it gets too long and you start interrupting each other uh when when people take too long to respond they can check this and if your phone line latency is showing that you're more profiling like you're thousands and thousands of miles away versus 20 miles away I can't hear it in

my voice we can't hear it as humans but the machine can pick up on it and you may be traveling there may be a perfectly legitimate reason that you're not in Boston calling about to your account but it can make a flag for that customer service agent to say all right let's check into this a little further because something is a little bit out of sync the voice isn't matching the I think that's a really interesting and I'm you know I'm going to say I like option b here it's letting the machine do

what the machine is good at while the human does what the human's good at and that's what can I help you with taking a general abstract problem and understanding it we're still beating the machines on that one um and this is not all sort of touchy feel and free they're not doing they're saving about 10 to 20 seconds per call on authentication times so so there's there's a financial incentive for them to do this as well my overall point here though is that tools are agnostic we can use a tool in

many different ways people decide how we use tools people decide and I think that's a really interesting an important thing to pay pay attention to here so that's my first point I got six total points so some of them are longer points though so you don't don't get don't get thinking you're going to get out early here um let's talk about these technology decisions for a minute think about the first caveman that picked up a rock option A could Zog could CLA Grog on the head with the rock right and

so this is a picture cartoon picture of caveman clonking other caveman on the head with rock another option would be could build a house with it use it as a hammer use it as a as a stone The Rock didn't know the difference my analogy here is that these tools these artificial intelligence tools that we're talking about they don't know the difference either they don't know if we use them for good and they don't know if they use them for bad and that's still a huge role for human agency all right

you have a protest beautiful yes yes what's your clue that they're generated he's got three legs yeah it's a stable diffusion yeah our friend here has got three legs all right so am I actually although I got to say even with that it's better than my artistic skills um but I would even I would have known not to put three legs he also has fewer fewer toes than we might but the toe thing might be sort of a stylism um guessing the three-leg thing is not it's a choice it's a technology choice that we

all have uh sharp eyes there um and actually that also brings up if you look at the bottom here almost every when I've got slides or numbers or reference material at the bottom I've got URL where they came from if you're curious later okay so we've been making these decisions for a long time about how to use technology and this is we have another big case coming along with artificial intelligence um I think about this person who made fire the first time or discovered fire invented I don't know

what you do with fire I don't know what the right verb is for fire discovered fire cuz the fire existed um prob pretty excited about it pretty warm pretty happy about it um I don't think that they were sitting there thinking man we're going to go to the Moon someday because I found this fire thing it is really difficult for us to imagine where Technologies are going to lead when when they come along because we we think about them from our own frame of reference um probably using for warmth but i

mmediately you can use it for light right so I don't think we'd be reading and writing like we do now if if we didn't have some method of using candles uh we see cars here left and right um I don't think that cars were going through caveman zog's uh mind at the time or I don't think Rockets were these are just it's really hard to more imagine what we're going to be able to do with Technologies here and I think that's where we are right now with these technologies that are coming up with artifici

al intelligence uh we can guess but we can't really know all right point two it's really difficult to know what these new technologies will spark anybody complaining well I'm we onethird of the points here but the my my later points get bigger so don't don't all right this is a chart actually Bob sent me this chart and I liked it a lot um so I decided to incorporate it here um what it is here is the the growth in the in the global gross domestic product so kind of measuring how much economic act

ivity there is um and we have time scale across the bottom so from the year thousand I guess the research didn't go back any further than a thousand uh to right about now and it's trillions of let's say dollars uh this is a giri Hamman dollars which is an economic thing um and it's a logarithmic scale so this this little shooty up thing you see here it's even more shoty up it's it's orders of magnitude shooty up shooty up is technical term um all right so what you see here is some revolutions th

at have come along some technology revolutions uh you're I mean I'm going to point out a couple of them printing press came along that's kind of a big deal um you know see they've got some nice charts here about the you know the magnitude of GDP increase over the next decade from when these Technologies came along um I can't help but mention Steam engine sitting in this this context here um big deal uh moving the to where we can run the belts off the steam engines versus the direct drive from th

e mill right outside um transistors is a fascinating for me transistors is fascinating for me because it's this 1955 and Bell Labs it's a big deal and I am not terribly sure that people like students in the year three ,000 when they're copying Wikipedia for an essay like they're going to be doing are not going to be thinking about man that transistor and the same word that they talk about wheel and fire I mean this is a a huge deal and it's happened while we're paying attention to it and what pe

ople are excited about is artificial intelligence and is that in here now this is a little I'm snark on this graph a little bit you can put any technology in here on this line when it was invented it's not establishing a causal relationship here but we can squint and think about the things that these Technologies enabled and allowed them to go forward all right so the question is is this going to be even more hockey stick are we going to have to double log this y AIS here to to make it fit on th

e screen good okay keep going here I'm a professor as Professor we have to give quizzes so here's my first quiz for you there'll be more coming later um how much are you personally using a uh right now I say on your job and I say on your job because I asked this to about 6,000 people around the world and I have their answers here in a minute um think about it for a minute you have to Anchor in key in asking a question in class is forcing people to Anchor in if you were in class I'd make you writ

e it down a piece of paper so that you you instantiate it and you commit to it um that's exactly what the I'm I'm G I'm gonna let you off the hook here and just say here's what our respondent said 66% of the people said that they were not using artificial intelligence minimal or no use does that feel right why not what's the just right there there's the tricky part fortunately we had the sense to come along right after this question and ask another question and say hey well you know what about t

hese things have you used these things have oh well yeah I use that all use that all the time so 43 half those people immediately recognize that oh no my first answer was completely wrong and I think that is part of the Insidious nature of a technology Revolution is that it just gets pervasive pretty quickly how many people here came with GPS I did I mean it's like I'm not sure how we navigated before right um I'll mention the SE later as a way of navigating but the Point here is that these thin

gs are just Insidious and we take them for granted if you if I gave you if you let's say you're someone living in 1700s and I gave you a quill and your quill you were writing and the ink would turn red If You misspelled a word witchcraft right I mean I think up in Salem they did some stuff about that right that's craziness does spell check strike any of you as craziness right now that's not even AI right that couldn't be AI is it's just so normal we get very used to Technologies uh and how quick

ly they go yeah only things that don't work exactly exactly um one of my actually I don't have a slide on this but uh I cut it out for time which you're going to be surprised at when I talk for three hours here but one of the fun things is most of the definitions of artificial intelligence say something like AI is doing things that normally didn't require uh that required humans without humans doing intelligent things one it always has Intelligence on both sides we're not very good at defining i

ntelligence so we put it on both sides of the equal sign which is a mathematical file um but we use the word normally a lot because if it works it can't be AI all right yeah so we ask that if you're curious about there's a whole bunch more questions that we ask here and got some URLs at the bottom if you're curious um the problem is actually I think one of the phrases using AI is a problem like I know what it means to use a stapler of course a red stapler because the Office Space movie people kn

ow when they're using a stapler it's pretty obvious you got a stapler in your hand you don't have to understand how a stapler works you just squeeze a thing and it puts a staple in it um and stapling kind of means the same thing everywhere to everybody all the time so that I think that's a problem like we don't think about using AI in that way because it's not like a stapler even though we use the same verbs I did an interview um Bob mentioned the podcast we've done some interviews with people a

nd this is Fiona tan who's at the time was a CTO of warfare and I've noticed as I've done these slides that I think everybody that I have in the slide no longer is in their position that they were when I interviewed them I'm not sure if that's causal we're putting people out of their jobs maybe they're getting promoted because of it um but what she said was hey people don't come in every to work every morning and go hey I think I sure would like to use some artificial intelligence today you know

they're looking to to do things they're not looking to use tools all right so my third point is that there's widespread use of artificial intelligence but widely varying types degrees of awareness and how people are using it good all right only four more hours okay as we think about this most of the time people will say oh my gosh the robots are coming to take our jobs that's a common fear um I think that fears makes sense but I think it's misguided because there are two things you should be mo

re afraid of right now than the AI coming to take your job um so in case you weren to sleeping right now here's two things I think are more important first other humans that are better at using artificial intelligence are going to take your job before the machine takes your job that that's just path of least resistance I mean we know about the diffusion of Technology Innovation from ever Rogers and thousands of books on this but the point here is that we doesn't have to be replaced by a machine

for you to get affected by artificial intelligence it can be a human using it um there's a great quote uh Pedro Dominguez who's a professor of Washington um I think he's still a professor so that's okay I didn't put him out a job um AI will not replace managers but managers with AI will replace managers without it um a similar quote here from Eric bson who's no longer at MIT who's at MIT when I when I I talked to him about this people using AR are starting to replace people who don't and that tr

end is just going to continue I think that is a Salient worry and I have to put a little slightly thing in parenthesis here that I love both these quotes that they're not talking about technology is the actor I have a particular peeve um one of the things that we edit out of our podcast is people saying AI does X because the technology does nothing it's like saying the rock does something or the hammer does something no the person holding the hammer the person using the technology does something

and then most of these quotes are are spot on with that human agency all right so that's the first thing that you ought to really worry about is other people using the technology better than you second thing if you have an organization that's not using artificial intelligence it's going to go out of business and that's going to be a widespread job loss when some other company some other organization uses artificial intelligence and the the reasons are not you know particularly evil or they're j

ust it's a cost based argument that if they can do things faster um cheaper I mean this we're we're in the context of faster cheaper standardization and and if another company can do things faster and cheaper including knowledge work I what we're talking about here is as largely in physical mechanical work what's interesting is knowledge work here but people are going lose their jobs one of the things our research found though we also ask people people and this was an afterthought that we asked

that I'm so glad we did 60% of the people responded to our survey felt like the tool was kind of like a cooworker that is not the sort of feeling that people have with people who are going to the robot who's going to kick them out of their job and I think part of the reason for that is that most people are overworked they've got too much to do they have we are all from Jay-Z and Oprah Winfrey down to me we are fundamentally limited by 24 hours a day there's just nothing that we're going to be ab

le to to do to to change that and this is offering some opportunities to get more done per unit time all right that was perhaps a little too optimistic though I've got this picture and this is actually a relatively dated picture now um it's about making sneakers produces 2400 shoes in 8 hours and that would require 200 people that's exactly the sentiments that people had when they were assembling the watch screws upstairs that we were just looking at this is the sentiment there was a strike I wa

s just reading about uh I lost my track where I am there a strike over here um was around these types of ideas and what are we going to do if all those people lose work who's going to buy those sneakers these are not trivial tasks because it's not going to be the same as saying all right last week you were a sneaker assembler this week you're going to be an AI expert that's it's very that's not going to be a easy transition for people so while there may be a a net no change in jobs as these Tech

nologies create new job new opportunities for people it's it's not always going to be the same person that loses a job and gains a job um I'm I realized now I'm mentioned my son twice today and I only mention my son he really likes plain he used to like plain hamburgers like not hamburgers he liked bread and cheese and bread that's what he wanted and if I go to the kiosk if I go to the person here and try to explain that to them they roll their eyes because they don't understand why would anybod

y ever do that and it takes about five minutes and as I'm waiting and doing that and telling his order I can feel the eyes of the people behind me like like what is this guy's problem he needs to get going I got important stuff to do now I go to this kiosk and I put in exactly what I want and get it I don't know if you've been somewhere that's uh like been traveling somewhere if you walk to this kiosk there's a button in the corner that switches it to English and Spanish and French and hundreds

of other languages and that again is a second evidence in our hey we're switching fixed cost and variable cost because you can't have this person trained to understand hundreds of languages but if you make a fixed cost investment in the technology you can scale that technology out cheap there's I mean people like this because these things don't go to the bathroom they don't have health insurance they very appealing here um really should work this in better the the truth is I just kind of like th

is picture um and I really don't have a good point with it other than I just love the idea of how sheepish this robot looks called looking and at cats versus doing his important charts um I said that the robots didn't go to the bathroom or take breaks well maybe maybe they didn't all right this is not a hypothetical problem in Amazon warehouses right now they have these little gizmos they go underneath the stack of or what you call that shelves pick it up move the Shelf to the person and they mo

ve the right shelf to the right place all the time I find that fascinating because one of our very first jobs was in a lighting store and I would go around and fulfill orders for chandeliers and stuff and knowing where things were in that store was a huge like for the first three weeks I was lost I kind of had to walk down every aisle looking looking matching part numbers um this thing doesn't look down part numbers is productive on day one Amazon has exploded in their use of robots but not so m

uch with people yeah in my lighting job they when I was a kid and and tinier they used to make me climb up to the top of the shelves because I was probably more Expendable and less more pliable yeah so this is happening you know this is you got to think that if Amazon was going to grow like Amazon had grown they would have had to hire a lot more people uh to achieve that same level of growth um without robots it's not hypothetical this is right now um what's interesting though is it's also switc

hed from being them to us I don't know about you I I was not hoping to become a sneaker manufacturer so that sneaker manufacturer assembler slide didn't really hit home for me I you know I did work in a warehouse but I did not think of that as my my career but being a judge that's a human intellectual activity and this artist technical article talks about how they're just as good at predicting with models what a judge will do um as a human will do that's a human task not a not one we think of as

typically machine um and then I found this and got a little sad it's not just them it's not just us it's me so in 2017 I think yeah the world economic Forum came to this prediction that by 2030 students be learning faster from robot teachers than from from Human teachers and in 2017 when I first read this that was that was a long time ago way in 2024 that's feeling a lot closer than it than it did at the time I can see this I'll come back to this in a second um because it it the the fundamental

truth of it makes sense to me uh that it is really difficult peer-to-peer individual teaching works but it doesn't scale you can't do it in a classroom and I'll talk about this in just a second all right I think you've heard a lot about generative AI seem seemed like Bob has spent some time playing with Chad GPT recently here um what's been interesting computers have always been good at math so this is a sequence of numbers what's the next number wait to see if we have any Rainman type people h

ere in the all right it's tough we could do it we we we could do it but it's it's difficult actually that number is a million 48576 yeah done it before I had to look that up ahead of time you can check me to see if I'm right maybe I mistyped it and I can assure you I didn't do it by paper and by hand um computer's really good at this always have been very good at this not surprising we're not very good at it um now what about this problem so this is a sequence this is the Leland Stanford's horse

sequence this is this first one of the first moving pictures and your goal is to predict what picture goes here is it this picture or this picture you got it immediately right I mean it it wasn't even taxing in fact it's so easy that you're doing the undergrad thing right now where you're wondering if I'm trying to trick you right nobody answer a simple question because it's the Propet is going to going to no I'm there there's no trick here this is the easy dead simple answer and we got it inst

antly much much harder for a machine now machines actually can do this at this point but it's not like the mathematical program that I just showed you before we we are as beings amazingly good at predicting patterns and sequences if I'm talking saying a bunch of words you immediately know what the next it's going to be it didn't tax you at all did it right all right that's exactly what's happening here with this new these new tools of generative AI that uh um I'm going to make fun of Bob's uh U

Chad GPT use here slightly in that it all felt a little generic to me it all felt like stuff that it probably could have just Googled right it didn't have to generate that text as much as it just had to find that text um now here's our exercise here I've started this is my first sentence it's a very first small sentence here um it has a blank here if you had to guess based on your years of experience what the next word is what would you guess actually out of this total I'm limiting our dictionar

y to seven words the open closed muddy lonely unicorn or veryy which what's the word you're going to pick here no other information the why sentences in English start with the statistically speaking if you have to you know bet on something you're going to bet on the so if I was going to think about the predictive probabilities of the next word here I'd say well 90% of time it's going to be the there's some sentences that open here you know that have these things but on the and I've made up these

numbers don't don't these are these are not real numbers these are stylized numbers here all right so let's take this I'm going to show you a series of other sentences the bank was Bloody all right okay I hear a lot of opens and closed right feels like it fits we recognize that pattern okay and yeah I think I agree with you I mean it could be the bank was the worst experience I've ever had in my life but also very likely it's open or closed um maybe I'm showing a positive bias here towards bank

s by being open um it probably not a unicorn that's what's happening when what's that it could be very crowded so very very still in the running here all right what about so frustrating the bank was all right so what happened was your basium update you did a basium update you took the prior the posterior prior what you had heard before and you said I'm going to update this to say all right now because of this frustrating because of this tone because of a look back that we knew in the sentence we

can Upstate and say all right closed is pretty likely m m is possible yeah unicorn is still not in the running here my shoes are ruined so frustrating the bank was muddy yeah yeah that's previous one yeah so here's the point this is what these tools are doing right now now I have obviously grossly simplified what's happening here but these are predictive updates of what's happening and what the next word is most likely to be now they've gotten incredibly fancy their dictionary has more than sev

en words of course um the patterns are that they're learning are in the trillions of parameters but this is the the core idea about what's happening actually they're uh there's memory you know the closer something is the more likely it is to be Salient um actually some of the recent ones Bert is bidirectional that's what the B and Bert is uh bir Direction actually for some reason all these models are named after Sesame Street characters there's Ernie and Bert um bird is bidirectional um looking

to say okay if I put this word in how well would it fit going the other way anyway all right that's a a little bit of a story here about the bank with and so what happened was you know statistically speaking Bank probably meant um financial institution when you first think about it but it could easily be Riverbank and because we have cognates there probably other words for Bank all right but that's what's happening when you're looking at Chad GPT series of blanks series of blanks series of blank

s over and over again backwards and forwards trying to get a cohesive set of of thoughts together okay how many people here have used chat TBT played with it all right this is not the greatest audience for that question because uh you you're come to spend a rainy night talking about artificial intelligence so so you're you're a little bit biased here but there's a Pew study that came out October this year last year um that said that most people have not used it and few think it will have a major

impact on their job I would love to see these numbers updated um I pull one there's several interesting charts in this and the URL is at the bottom here um one of the things I'll point out is let's look at this one they don't survey people less than 18 but if they did I can I can extrapolate as to how that would go and actually based on my sample size of two uh 13 and 15 year olds uh they are all over this technology it it's amazing and I and certainly we see it in the classrooms all the time u

m when I see charts like this I immediately think about the divide that going to happen in our society between the technology Haves and the technology Have Nots and that's Technologies are just never universally benefiting everybody somebody's always benefits more somebody benefits anyway this fascinating study um what are people using generative for um I probably have too many slides here for this this is an interview I did with mcdad Jaffer actually ironically who just today announced that he

moved to open AI so yet another one in my list of people they no longer though he work exactly yeah you might need a new a job and come talk to me they're all employed by the way there's not they've lost jobs um so Shopify may not quite be familiar with it but it's a it's a plumbing and infrastructure for e-commerce and so what happens when people set up an e-commerce site they have to fill out lots of product descriptions what Shopify did early Chad GPT came out in something like October Novemb

er they had this product in place in February where it was when you were filling out your description of your product you'd show it a picture and it would generate the text it'd give you five or six different versions and it would let you pick which one you liked and edit and adjust it massive value I save people 30 seconds over and over and over and over again as they set up a giant website they're saving 30 seconds a time I think that's what's pretty fascinating here so that's an example of uh

one product that that was literally in production here go ahead W but see yeah so actually I have something about human areas and vision coming up here in a second but we're not perfect this is part of the thing we we make mistakes too um and so it reminds me of there used to be some studies about Wikipedia versus encyclopedia branica and those did not go well for encyclopedia branica I mean just because anybody can edit Wikipedia doesn't mean they don't but I don't have a good like concrete nu

mber for you on that' be interesting um what else are people doing um generating audio I you saw that I generated these images here with stable diffusion generate audio um perfect for elevator music if you're need some background music um code I don't know if you coders but so much code is on the internet and it's very easy to scrape and copy and place in and tools like Copilot are doing that right now um it I've done this actually I did it with a with a p torch Vision model and the code was oka

y it was bad it wouldn't work but it was good enough to where I didn't have to start with a blank screen I don't know about you but I'm very good at telling about things I don't like about something but the blinking white cursor is intimidating to me and so it generated a bunch of code that I knew was not right but I started to rearrange it and fix it and I think it did so save a bunch of time um video you want videos of your sales product out there companies will do it let say I'd like a a woma

n X age X you know look and they would generate it talking about your product so what it means is if you come to a website and this to type of person is going to appeal to you they can make that the sales person just a matter of figuring out what which which is going to give you a better sales lift uh um style I need this one marks and Spencer they they'll tell you what to buy um we've got a podcast episode coming out um about this not with marks and Spencer with a with the different company Sti

tch fix they're talking about you know how they're predicting what kind of kind of clothes you might like um feedback listening to call center employees giving them a feedback on all right you're a little Annoying there you're a lot annoying there you're um here's how you can improve in real time improving in real time is fascinating I actually have a much larger study we don't have to talk time to talk about about how people are improving playing chess through immediate coaching oh no that's a

bad move that's a much better to know in that moment where you can take it back and think about it and and try and assess different paths versus the model that we're used to in school which is we teach you a bunch of stuff then we we give you an exam and then we disappear for a week while we slog through raing and then we give it back to you stuff like this is the idea of real-time feedback is is fascinating all right um engineering I'm going go a little quicker here so there's a bunch of exampl

es here but generative tools are showing up everywhere and it was hard for me to find you know to cut down this list not hard to find them this is an audience check to see somebody ought at least say hey you're already there buddy what are you talking about but no you didn't okay well we'll treat you like a hostile audience here what if I wanted to get fit we have two options here I always about the options here I could hire a personal trainer and this is if I was Oprah or Jay-Z that's what I'd

do anytime I was ready to exercise I would have that person ready to go whip me into shape whenever I decided to quit eating chips on the couch right um this is my Lifestyles of the Rich and Famous approach not practical for me yeah phot Shop's easier um option b you just came out a product from pelaton they'll put a device in your home that will watch you and say that plank looked bad you said you were going to do 12 situps that look like four to me you know th that's what this tool is doing th

at's tool I mean call him a tool uh that's what this person is doing um and we can have this done uh by a technology in the in our living room ready whenever we're ready to peel ourselves off the couch um little device and exactly it could happen see if if they work hard the difference here is the goal is not to get a predictive model for the machine to do something it's for me to improve and I think that's a a distinction that we've lost um up till now um and I'm pretty excited about it this an

interview we did with Sanjay niani who was no longer at pelaton of course um but realtime feedback and metrics driven accountability that's what they're trying to do and put that in place for people I'm guessing this product doesn't work great the first time I bet it doesn't work great the second time one of my favorite quotes from our podcast is Gina Chung you said the first day is the worst day because these systems get better and better over time and improve the first day that this person wa

s a personal trainer they probably didn't do a great job either we don't fire people their first day on the jobs because they're not perfect and I think that's what's going to happen with these Technologies all right the point here too is that this is not where we're trying to teach the machine to do something something for us we're not trying to have the machine imate and do what a human can do we're having to the machine help us get better all right this isn't hypothetical Adam's my son I love

Adam if you're listening Adam I love you um Adam and I during the pandemic assembled a Lego chest set and it was really cool and fun we had a great bonding experience I showed him how the pieces move and I was the cool dad and he thought IID knew something and um you know we're playing and I I'm letting him take back moves I'm saying all right Adam are you sure you know think about what I'm going to do if you do that right we're had a great time suddenly I started having to pay attention becaus

e he's getting better you know I can't be sort of zoned out and halfway paying attention then the little snot disappears for a month or so and he finds this online site that is a aib based web uh chess training and he goes through a bunch of puzzles and exercises has a bunch of critique done of his games and the story got worse this was so fun when I'm saying oh Adam are you sure do you want to take and now he's doing the thing like Dad do have you thought about that are you and you know it actu

ally really kind of is a sad we don't play much anymore because he's just got I like to think that I could if I would go spend time get better but it hasn't happened yet um but the point here is that this is an exercise where he went he didn't hire an instructor he didn't hire he didn't play with any other bodybody else he went online and found some tool that helped him get better that's big um are we still qualified to teach um I'm going to give this a little short shrift in the end of in the i

nterest of time but there's a giant contest for computer vision that's happened over the last let's say 2010 to 2016 where kind of shut down you'll see why in just a second but what happened was they were using computer vision algorithms to recognize pictures the challenge was here's a database of 14 million pictures what is in this picture and these were the submissions in 2010 huge dispersion lots of different techniques a lot of code written that said something like I'm kind of extrapolating

or being intuitive here but you know if it has four legs then it's you table or is it a cat well if it's furry then it's a cat if it has a tail you know it's just this B giant thing and it's an internet success story because what happened was after the first contest all the people here who did well everybody looked at what they were doing and learned right that's great that's the point of a contest um but in 2012 something fun happened Alex De out of University of Toronto guy came up with a deep

learning approach and this was hey I don't really know what's happening we have this system that will mimic a human brain and we're just going to throw it and teach it repeatedly show it the same pictures over and over again until it learns and out of the gate Alex took him behind the witch head and spanked them next year we had a few couple holdouts trying to you cling to the prior thing but everyone saw what Alex did and pushed up the accuracy to the mid 80s by 2014 everyone had given up on t

he old approach and gone to this deep learning approach these deep learning approaches are exactly what's behind Chad gbt uh and these other models um we have a little dispersion here happening in 2015 where I think people are exploring trying to do something different you know hey let's try something see if it works if it doesn't work let's try something else and that one did work and so we're we're exploring and learning um but here to your point about errors this error rate right now on this

network is about 3.6% um human error and looking at the same images is about 5% um and the difficulty comes down to if you have a picture of a mushroom uh these models will call it in inoi which is a type of mushroom and the right answer might be mushroom so it's it's it's more than right all right so we're not really still qualified to teach people yeah that Che are you are you a robot or are you a human yeah yeah they can I don't think those are those are not going to sit around for much longe

r I'm really kind of surprised they're still being used because that I you going to give me on a tangent here but much like copy protection I think it's attacks on the honest and it doesn't affect the dishonest what I mean by about that is that if you're trying to get past these things you've got a system that will beat these things these if you're trying to beat the captas You've Got A system that will beat them if you're just trying to get into the website you're getting annoyed uh right so it

's it's not doing a good job of separating out the the Myans from the non- Myans um actually there's a whole different tangent like catches about like anyway okay let's keep going all right the second thing is I think that we need to move away from this model of directed learning I'm really big on a directed learning model this is again Chad DP sorry do e which is a different diffusion model uh asking for a professor College classroom bald male um that hurts but that's that's that's part of why

we feel like these images are right is they reinforce the stereotypes we have in our mind and if if it violates a stereotype then it doesn't seem right and that's a whole giant problem um but what I used to have in class was a very narrow distribution and some students would come in they were mostly gathered around this mean I would teach in this blue band here and there were some Advanced students that Knew Too Much I was going to bore them there were some students that didn't really have the b

ackground I was going to lose them and I was just trying to pick up the middle you know just trying to get that sweet spot um because I have to teach the same to 35 people and 40 people what's happening now I think is people students are we're seeing more and more dispersion in students we're losing more people because my what I can do in class doesn't change but the the tailes I'm losing and the tailes are getting bigger and so it's an argument for personalized learning personalized learning at

scale and there's a whole philosophy here about the pedagogy of the oppressed from Pao frier um it it rails against a banking model of Education where we treat students as empty vessels and like I the Learned Professor tell them everything rather than co-learning I think artificial intelligence gives us a chance to move towards co-learning individuals it worked well beyond the trasman I don't know if you watch the good place this is a uh an episode where cheaty on the on the left is breaking up

with his girlfriend Simone on the right and he's trying to figure out the best way to break up with her he goes through a virtual reality simulation he breaks up with her like a 100 thousand times all trying different ways because he doesn't want to hurt her feelings this is an episode where he hands her a puppy at the same time he says he's breaking up he hands her a puppy and it it turns out actually for the you know the moral of the story is there's no good way to break up with someone but t

he point is Chey was able to practice a difficult task and it's hard to practice difficult task I have a daughter who's about to start driving I would like her to go through a lot of simulations I would like for her to have near wreck experiences or wreck experiences without actually having to have wreck experiences there's a surgeons who' separated conjoin Twins and they did it in a in a digital simulation to practice a bunch of times because you don't want to practice on a bunch of twins becau

se one they don't come along it's not like you get hundreds a day to practice on and you want to get them every single one of them right anyway goes beyond the classroom so that's what we can learn from a machine but I want to tell you the story about the fosberry Flop which was a high jump so dick apparently he died recently dick Fosbury everybody before Dick Fosbury wasn running up to the bar and jumping forward over it he started jumping backwards over it and ghire people saw him doing that a

nd within two years every single high jumper was going backwards over the bar we saw an improvement it diffused through the population and people everybody jumps backwards now carve this is a we'll put a sensor in your ski boot and tell you how you're skiing and tell you to put more weight here more left weight and it'll do it in real time in your ear what's going to happen is somebody's going to do something crazy on their skis carve is going to be watching and listening and and most of it will

result in yard sale crashes of ski equipment on the slopes but something will work something will be amazing and when it is amazing the machines are going to notice and then tomorrow it's going to be telling all the rest of us how to do that better and I think about this particularly with language you know if you're learning language um you might have a good teacher that over 30 years gains a lot of ways of you know teaching well and but if they discover some way that people are learning langua

ge better these people do a lingo have a ability to do it at scale that's unprecedented and individualized I think they're they're having amazing product right now um heavily heavily used artificial intelligence point five we're already good at teaching the machines we need to think about what happens next all right we're kind of running along I'm going a little quicker here um my last point is more about things I'm worried about this is Janice the two-faced gu so I'm going to be a little bit tw

o-faced here we can have ai tools do a ton for us and that's awesome they can get rid of the dirty dull dangerous that we do they can get rid of the tedious they can get rid of they can give you a a draft of a document first um at the same time I'm worry about how we're going to progress from the basics if you don't know how to write a bad document do you ever learn how to write a good document um and we don't know the answer to that yet we don't have an understanding of the long-term effects I

worry that this is becoming a race to mediocrity we lose skills this is a seent was on the image here there's a sexon upstairs here anyone here can navigate by the stars of the seon I'm guessing no we don't we've lost that skill and I think it's okay GPS works better we were talking about hand cran cars apparently that that car back there in 1912 has a rear crank instead of an electric ignition has a rear crank anyone here know how to do that it's okay it's to you break your arm apparently doing

it um gear shift Can people drive a stick yeah okay see now that feels good here that's going to be the same thing here in a bit longer here you're you're thinking oh no I could drive better than machine it's going to get you eventually yeah it's called intangible cultural history that's like class of things is like our tangible cultural history like we're losing like through like you said the six shed or like looking at the stars and we legitimately lose l like if you look back they just can d

o I I have tried very hard to start a fire by rubbing sticks together I put actually I even got my cordless drill out to spin the streck for me and I still can do it and me I'm guessing most everybody could for a while we're losing some skills it's not clear which skills we want to lose and which ones we don't and I'm not sure if this these Technologies are giving us a head start or they're helping us be mediocre we had to figure that out what do you think about the idea about not losing necessa

rily skills but like ability so like there was a study done on London C pre and post CPS and they like literally had different Trin structure and I think amount the brain matters like they they literally have less cognitive ability because they don't spatially navigate regularly yeah tie that back to my Chad GPT chart that showed who the young people were using this oh sorry um using these Technologies changes the way we think in general fundamentally and it's not a spefic specific skill it may

be a more General thing and her example was London cabbies whose brains are different before and after GPS the point here is that we may be not just losing the ability to do a skill we may be kind of losing some general skill abilities I don't think we know I don't think we know that yet um but but I worry about that especially with the age distribution that that Pew chart showed um it's all changing really fast really really fast sorry is that a hand out Wonder might want to touch on one thing

and that is to some degree we've been talking here you've been talking about like an AI for a particular instance but now you have several hundred of these guys going at once and coming up with slightly different perfectly good answers to the same question is there is this problematic and I think ways it happens to me uh just in in in what this these very tools what they think I what they think my phone what they think address yeah I mean there we get a lot of answer if we ask about here about s

omething we get a different answer too so I'm not sure that that's necessarily an AI specific thing that they all give different answers I mean yeah and it may be good if more things are thinking I mean that that it's a positive um it's changing fast we can get these benefits pretty quickly we can scale we can correct actually I really like stories where tools find bias and do because then we can change it like there's a ton of bias happening there's a ton of like hiring and racial and misogynis

tic bias in our society and the machines didn't create it we created it and the machines would give us a chance perhaps to correct that at scale if things are in an algorithm and the same things things can explode before we know what to do with them I think I'm worried about the pace of that um data science is the sexiest job of the 21st Tom is still a professor bson so that's good um but his philosophy next should we be philosophers and not are you okay Bob do you what we should I'm acting I wa

s just checking in with frederi here about the online audience and whether there are questions oh okay all right right you we can we got I got one more concern here and then we can kind of wind down um I have a big concern about large technology companies that benefit from this scale the cool thing is that we get crazy valuable tools for almost free we can go to chat GPT and use it for free even though it cost you know Millions hundreds of millions I think Microsoft's investment was a billion do

llars that's cool that we can get access to these tools but it's we have very little control or knowledge about what's happening behind the scenes in these tools who's who's running them whose objectives that they're thereafter um we had a great example in the 80s of Sabers technology where they put put technology in every travel agent and it happened to show all their American Airlines tickets first and it was a giant uh lawsuit about that um I'll go quicker on that I think they're disturbing p

arallels in 1906 Upton Sinclair wrote a book called The Jungle about the meat packing industry in Chicago and meat packing was gross um I think we might be similarly horrified if we knew what was happening inside technology companies right now we don't know what's Behind These algorithms we don't have any good ways of finding them out we faced similar things before we solved it with food we put in the Food and Drug Administration we put in the CDC we put in restaurant inspectors we we don't hear

a lot of news about bad food in fact when ecoli outbreaks happen it's news because it's so unusual um we've done with nuclear um biotech we've done some things we have a way we have we are not new to this um we can restra there's there's lots of things we can do you don't have to have a perer have to be perfect we can we can change how these Technologies play out here but I think we need more transparency to be able to do that we do that with CPAs who go in and audit the books of a company and

then we trust the CPAs and so then we buy the company right they're mechanisms I don't know which one's perfect but um it's hard the digital digital hard digital makes things hard and the big question here is not so much about what we can do we can do a lot it's a question about what we should do and that's fundamentally fundamentally changing here and so that's my final point is that when when we combine an agnostic tool with hidden back room processes happening at speed and happening at scale

we're likely to end up with trouble that bom feels like a good closing argument I have a bunch of content here if you're curious about it all these stories come from something but that is a good place to stop and get grilled by the online audience so I've got the second mic I to stand behind the speaker we get feedback um learned this is our first um hybrid event hosted with the gbh Forum Network and one of the things we've learned is that um the online audience couldn't hear your questions beca

use you weren't speaking into the sound system I should have repeated that mate right so um sorry to the online audience for that technical error on our part because um we could have prepared for that um there were a couple uh questions that were submitted in advance that I'd like to POs to you uh one is um how can young professionals get ahead in the AI space uh ahead is a tough can I um everyone's doing this so that getting ahead is really difficult um but most of these Technologies are relati

vely easy to access you went to Chad GPT and played with it right me you know more about Chad GPT than you did before you won't get wise with the sleep in your eyes right I mean that's the the classic Rush lyric um you got have to play with these Technologies you got have to to use them and so much of this that I'm excited about is open source and free and available on the internet now usually it's limited in some way in terms of like number of times you can do it or whatever but it's not a seri

ous limit that's going to stop so I would start playing great thing about an alline question says that they can't complain about trans and uh Tomaso prop or excuse me Sharon posed the question as a teacher I'm wondering how do you check if a student is using AI to do his or her work and um the followup was what impact does AI have on education yes so I'm I'm big on that or I'm concerned about that I think there's a couple different ways to answer that question if you think about a calculator thi

s let's make an analogy to a calculator if you're teaching someone addition that you need to prevent use of a calculator because they got to learn to do the the addition on their own but once you get to an upper level class or statistics there's no value in having the person manually do it might as well give them a calculator so I think you know when I you hear a question like that I I immediately want more context about what they're trying to teach and what they're trying to to learn I mean it

may be perfectly viable as a tool and I teach a class and machine learning people can use it to do whatever they want to and I think that's part of the skill that I'm now trying to teach is how do you take the output from some tool like this and improve it and make it better I think that's the important but that may not fit for whatever the context is I'm wondering if you've found yourself suspicious that the work that you've assigned to students of your own is coming back not having been done b

y the student no actually I'm not suspicious because I'm okay with in my context I'm okay with it that it's completely fine what I what I worry about is that they don't know how to get a better I can see the the output from Chad GPT or the co-pilot that is STO there and it doesn't work and they don't recognize that it doesn't work and they don't recognize how to improve it and that's where I think we have to work let me twist the question to answer something that I want to answer which is um I t

hink we're we're missing out on opportunities for if you think about uh experiential learning can happen in four ways if you go back to col's theory of experiential learning concrete experience you can ractice by doing things you can have reflective observation people can watch you do it you can get feedback you can experiment so there's lots of different ways that we can learn and I think there's a lot of ways we can map use of AI into each one of those four ways we can put people through simul

ations we can use the tools to critique them we can use the tools to understand better what they know and what they don't know diagnostically there's a lot of opportuni here could be also kind of a game of intellectual tennis where you hit the ball in the form of a question it gets hit back in the form of an answer you realize maybe my question wasn't so good so you improve the question and hit it back and you keep challenging the student who is now owered with the modern AI technology you have

to up your game as the questioner as Challenger yeah Socrates rebirth so um I'm going to walk over here and just see if there are any questions that are in the um chat on the zoom call um there are I'm hearing a knot there are two one one sec anyway I think all this stuff's fascinating obviously you know you can feel I can talk about it for a long time okay I got a lot of slides couple questions from the online audience um from Jacqueline what about the use of AI in social media and politics to

feed misinformation yeah oh yeah interesting timing for that um let me flip that what about the use of AI for detecting misinformation I mean I think that again that comes back to the tool is agnostic and we got to figure out how to use it I I think the ability to do to to do to do evil at scale is unprecedented now the ability to do good at scale is unprecedented as well and I think that's on us to to push hard on the yeah how my question that is the followup to the um audience memb question is

how can we use AI to rebuild Trust yeah me so you got to think about like misinformation marking investigating things for you finding the sources I mean these are all ways I mean what would you do to trust a piece of content right now is you go find you'd go read well these are the things that the tools can do for you as well find read synthesize yeah and the other question on one I can actually answer are there different levels available of chat GPT free and for a fee and the answer is yes yea

h yeah yeah they're they're they're recovering their money on some things um one of my concerns about well let's don't get into that it's AO broadcast live stream um audience questions I'm going to give you a mic so we can fix the earlier problem so please pause and wait till you receive the mic what do you think about the New York Times uh uh lawsuit uh basically they said they generated all of this intelligence and and facts um and they want in on the GS that these corporations are getting rea

lly fascinating there's a huge question here about whether we're eating our Seed corn so like the idea of eating your seed corn is if you eat your seed corn then you don't have corn to plan next year right well by analogy if we have taken away if we've sucked up all the computer knowledge from stack Overflow and codings websites who's going to add that content later on for the next generation of tools and I think the new New York Times um use of of content is exactly that sort of thing that's a

in the second period of the game where's that content going to come from and I think that's something that we beyond just the New York Times have to worry about is okay let's say we consider it completely fair to do that well then what happens next year when the New York Times doesn't doesn't produce content oh we got to think this thing through more than just one iteration some things that are Le legal individually are not always legal at scale for example let's think about surveillance Technol

ogies a policeman can stand outside your house and watch you right that's been legal surveillance has been legal forever but now they have the ability to surveil at scale and speed to surveil everyone all the time constantly no bathroom breaks um our laws and infrastructure we really built for that thinking because it because what's happened is the cost structure has changed dramatically and we got of for I mean if I had some great answer to that I wouldn't be standing here in the in the low Mil

l it's a it's a big problem I agree I agree with the problem hi so I um I took some AI related courses including one artificial neural networks and it was pretty fascinating and then the AI winter happened so in other words I took these courses in the 1990s and uh on the one hand it you know I have a little bit more informed sort of uh look at what technically what these systems are doing but on the other hand I asked myself why did I waste my time like if I known the AI winter was going to happ

en I might not have bothered because you know I worked in technology thought I needed to know it and that's just sort of a lead into the idea of an expert in Ai and Society and you know you see the headlines and it's like every headline is expert tells you what you need to know about Ai and I just want to ask the the expert how many H and civilizations have you led from pre-ai to post aai that you use to develop your expertise well unfortunately I think your your criteria for selecting an expert

is not going to leave you with a lot of people in your pool to listen from so if that's your your criteria then I don't think you're going to get anywhere um well I think one that's happened since the AI winter is that well three things we've had massive amounts of of data available that we just didn't have before we have massive ability to process and finally we have massive ability to communicate that those results and those three things have all come together and again you know back to my pi

cture is a the fire discover really hard to know how those three things were going to come together to give us this rebirth and it's really hard to know what that next winter might might end up being but I don't think you're criteria for expert I think you going have to relax that a little bit or else you're not gonna you're not you're not going to fight any expert in the media they're portraying that as how yeah I I find it always funny in in uh in in job search you know where people want extra

you know 14 years experience with chat GPT right now I mean but I think that's in some sense liberating that that you're not behind on you know people are not behind on these things uh here's a statistic that I as an expert can quote 0% of the world were born knowing how to use these Technologies and you know I don't I'm not going to side my source on that but it's all learned and it's all anything we know about Chad GPT has been learned in the last you know year or so now people built off of u

nderstanding like yours of Technologies and neural networks before that so there's a head start there um but it's all happening quickly and there's time to time to become an expert good do now are taking it easy here hi what do you think of attempts to regulate uh like the European Union's um attempts and will they be successful what you know yes tough it's tough to regulate any technology um one of my in my past life before I was a professor I work with the United Nations in the weapons inspect

ors program that's that's a way that we tried to regulate uh atomic and you know we've not had a significant atomic explosion since you know 1945 so you know there there's there's some degree of confidence that the things we do to regulate can work now in the case of that organization we shared information about safe uses of power reactors we safe use safe uses of radiation of of corn and you know seed crops so there's a a car and stick approach there you also had the inspection program which wa

s more of the stick approach um what's somewhat Difficult about these is that there are um these Technologies don't require the infrastructure that for example nuclear does uh they don't have the physical presence that for example food inspectors do you know so it's not like we're first time we're trying to regulate a difficult technology but on the other hand there's some things about this technology that make it particularly challenging and quick um but I don't think we can just sort of like r

oll our you know you know say oh gosh we we can't do anything about it um regulation is tough because it ignores National borders in this particular case we don't have a good Customs you know border control for digital Goods U we're going to work against that um some of the things we've heard about regulation for example oh everybody let's just pause well it's no secret that the people wanting to pause were the people who were ahead like I I'd want to pause if I were ahead too um and it's also s

o many incentives for people to defect from a pause like that um but I think you know to the other point of combining you with speed is these things are happening so quickly even if we did get some regulation in place it's not clear to me that it would be effective for more than a year and we've not seen regulation move that fast and we still think about copyright the way we thought about Gutenberg Press uh versus digital media we're we lag that's not very uplifting I'm sorry I didn't know the g

rilling was going to be such a big part of this my question is about the the term deep learning um my understanding is that basically the example you gave us with this forming the sentence about the bank deep learning is basically doing that with a really really large number of repetitions so um isn't it actually the opposite of deep uh so actually the Deep the Deep has to do with more about the create the models create layers um this this goes back to your your you know the original models in 1

980 um it's originally modeled off the brain so you have a bunch of inputs that go in and then you have a bunch of synapses that and it's going to I don't know anything about brain but like they fire they get active and you learn from those and then those trigger other Neons and then those trigger other neurons what's deep about these is the layers of neurons that they're creating to make an analogy when we train some a model to pick a good employee through deep learning we may not give it somet

hing like of attribute like race or gender because we want to not include that in the model but what we find are these hidden deep layers learn what the race and the gender uh are of people from that um and so there might be a node in there that is like male female uh that's that's learned its prior layer has taught it how to recognize male female the next layer then learns how to use male female and so even if male female was not in the original model it picks it up from the college people went

to the sports that they played in college the activities that they do um and so when you're talking about the deepness the deepness has to do with the construct of the model layers uh not so much the depth of repetition Chow depth I'll draw a picture too okay be shocked to learn to have a slide for that later on I mean yeah make more stes if you think about the later um I wanted to get your thoughts on one thing it almost seems like you know many organizations we know they're trying to invest i

nto AIML you know but sometimes it feels like some organizations like not necessarily identifying what problem they're trying to solve what the use case as opposed to Let's invest and see if it sticks so I think all the above are correct here um so I have some great great quotes actually a great quote from um Matt Evans who again at the time was the in the charge of the Airbus production um he's since moved on but um his quote when I asked him about why do you invest in artificial intelligence h

is quote was we don't invest in artificial intelligence we we that's not that's not our objective we were trying to solve a problem um actually we were Google an article with me and George Westman and kiar for arado um we just published um earlier this month about exact at sort of starting with the problem thing um now that's true but there's also some investigation and exploring um and so one of the things we did in one of our surveys a few years ago was we asked if people research first and th

en did or did and then figured out what happened and which path LED because we were thinking one of those paths would work and it turns out to your just do something approach both both pth R viable you know people can learn by either trying some stuff figuring out failing um you can't know to apply artificial intelligence to a problem unless you know something about artificial intelligence but you can't at the same time be looking for artificial intelligence uh to solve you know now I have this

Hammer where's my nail um did an interview with uh I got Orange Theory who is also not there didn't in the trend and I asked asked him what was his proudest moment about in using artificial intelligence and he said when I use logistic regression which is not artificial intelligence he found the tool that worked quickly and solved his problem and they didn't crank up the machinery and moved on to the next problem and it was very problem focused there's also a lot of Mark around it what The Cuttin

g Edge Cutting Edge um yeah so there's a whole um maybe I can kind of shift us a little bit to talk about a concern which is that it's happening largely in Industry versus Academia whereas a lot of Technology the basic technology happened in Academia and research institutions in the past but the money and scale required to do this is frequently now coming from um corporations um you think about things like open AI uh they used to be open now they're they're closed they got a lot of funding from

Microsoft we you think about the names the Googles the Facebooks metas these are the people behind these big models because it's taking a ton of computational power to do these things and that's something that you know rather than specifically telling you like who who's exciting I think that's more of a general trend of uh there's a great science article about um how that shift has moved um Neil Thompson and ahed have wrote about them it's just big Trend and we don't know how to we don't know wh

at to do about that is there a problem is it not a problem we don't we don't really know it's just not how things have happened in the past um the so you hear a lot about Ai and my question is uh is this mostly something taking place in the United States and is there a lot of patent activity behind it or is it just Global actually it's highly Global and um the Asia huge and China in particular big initiatives towards this so it's not even clear that you know I think we're sitting in a US Centric

World here it's not clear to me that we're while we get we generate a lot of headlines and the headlines we read are about our Technologies uh it's it's happening everywhere actually one of the best open- source uh models about to compete with gpts comes out of France at the moment if you would if you're curious actually it's always kind of curious there a hugging face has a website where they rank the accuracy of model maybe this gets into your ears here's an accuracy thing is um they've got a

rank of all the current models that are out there and how they're fairing in terms of uh accuracy on text I believe is what their atric is they it change it I can't keep up with hugging face in your uh co-learning example you talked about these carve boots and I've seen ads for them on YouTube and I suspect they're not capable of giving the kind of insights outside of the box that you suggested they might be able to that's not exactly my question my question is more that people are attaching AI

to all sorts of products powered by so is there anything we can do about that is the sort of trivial question but the deeper question is is AI even an accurate term for any of these Technologies are any of them ever like within a light year of being intelligent man then you're going to get down into what is intelligent which is a a disaster well prag Al in terms of human expectations when they hear it I think that some of these things are truly amazing Technologies now there's not say that ever

ything that says powered by AI is um and there's a joke about it's Ai and the PowerPoint deck and machine learning and you know and Engineering that that that exists mostly for marketing and actually one of the questions we have out in a survey that's we're Distributing right now that we'll publish the results in November is to what degree is all this attention distracting from progress on important stuff is this a shiny object that's luring people away or is this something that's helping lure p

eople towards some some goal but I think you know I'm a I'm skeptical about things in general I mean that's kind of my nature but um I do think that some of the things that that models can do are amazing I've been working with some X-rays I know nothing about X-rays and I've been working with some image data and I cannot look at that X-ray and tell what is a problem and what is not a problem and even I am able to train models to get 75 80% accuracy in detecting a problem through this tool and I

know nothing about I I literally as a human cannot see that now probably some actual trained medicine person could you know could do that U but I think that's pretty amazing that it can learn uh those when and what look like grainy images to me I think we need to get a specific about individual things if you want to you know Case by casee it's baseball season soon and I have one thing to add about baseball before we wrap um ai ai isn't Lao get slapped on anything it seems and is AI one thing lik

e a beagle or many things like a dog and part two is how can I tell whether it's really AI or just technology is so complicated I can't tell it from Magic yeah I mean so you're you said beagle versus dog I mean let's I don't know where we want to go with the kingdom F class order family genus species here but I would go a lot further up the up the taxonomy to maybe Kingdom and film before I think about where that word AI fits it's it's nowhere near those Leaf toads um and you know is it is it AI

are there many AI yeah there there are many different tools doing many different things many different you know there's text based there's image based there's sound based there's just a world of different things that we all lump into this and so that makes it really hard to Define uh what is and what is not um but before we had a survey we tried to put a definition we tried to agree amongst ourselves within the survey about a definition we never get there we just had to use we had to use the ox

er dictionaries definition which we didn't like but it was a at least something to get us out of continuously debating what was and was not there's a certain pragmatic marketing aspect of technologies that I think is fine I mean that people would be dumb not to sort of embrace this funding and marketing that comes along with it um that doesn't mean that everything that they're doing necessarily fits under that umbrella um but we did that four 10 years ago with word analytics and data and I'm gue

ssing there'll be a new word new word in in a month or two or a year or two yeah actually that's kind yeah I would not want to be in charge of putting that label on something um GMO non GMO AI non AI wow I don't think I want to see our government resources going towards towards the creation of that agency because actually by the time they labeled it it would change too I I'll drop in the baseball reference um uh this past weekend someone was here and told me that a mother load of film of old bas

eball games has been discovered and is being digitized I mean hours and hours and hours and hours and hours and hours of film of baseball games played in the early uh 20th century and um AI being used to review all the footage to identify who's in it and what happened so the they're it's able to figure out who the players were what position that they're playing and when something happens a pitch is thrown a ball is hit uh who the ball is going to how it was fielded so the statistics are being re

constructed about these old games for which there are no physical there are no written records and so they're able to and and and they're able and computers through image recognition technology you know parsed with some identification characteristics are able to reconstruct all of this that humans probably would never be able to get to in time so it's a it's a pretty interesting Arcane not important but interesting um way of an application of artificial intelligence but in that case the humans c

ould do it if they had the time and resources right yes yeah well you do the writing that to the questions about do we call this AI or not is gets it's really tough to say but it certainly allows it at scale you think about uh actually I'm looking here is the World Wildlife Fund they got cameras pointed out there looking at trying to find Critters and they identify the critters that's true that they could send a whole bunch of people out into the field to identify those Critters but if they can

have the machine do it that they can do it at scale yes you talked about uh these algorithms consuming their own seed is anyone doing this experiment where they just take vast amounts of chat GPT output put feed it back into the model for Generation after generation after generation to see what happens yeah I'm not sure the result but there are some people that are they're doing exactly that sort of exercise of generating some stuff and putting it back in generating it you know with our own self

-trained models I don't know what the right answer is um I did go to research presentation today where they were trying to give a shout out to to mang she was doing a presentation on Zillow and zillow's algorithms and whether or not this estimate leads to agents agents leads to the estimates and how those two are interacting and um her evidence was that there not a lot of sort of amplifier squelch going on in that particular context but there I'm sure there's a bunch of other research on that to

o started up you could do it too models are open no one's that no one can be that part ahead of you the stuff just came out a year ago that's what's pretty exciting about it call y'all got some endurance here ples I don't quite know how to phrase this but um I know that if a bunch of doctors or so on uh working on separate things or somewhat similar things get together in a conference and there's an awful lot of uh Comm uh communication between them and learning from one another uh do AI machine

s ever get together and I mean is it possible oh sure yeah I mean actually one of the the the trivia examples of that is is a for adversarial learning so one AI doing one thing and one AI trying to do the opposite so people back here mentioned misinformation think about one AI trying to generate information misinformation and another one trying to detect it one trying to generate fake images the other one trying to detect it U we've had tons of success with that with gaming like here learn how t

o play chess we're not going to teach you how to play chess but you play against this other machine that also doesn't know how to play chess and play B billion games and you learn how to play chess and what's particularly interesting about the chess example is that these machines discovered the same strategies that people have come up with over you know the history of human existence here now your your question was a little broader than you know I got it very narrow there with one against anothe

r this abial uh learning context but uh yeah you can think about maybe to to your example of like lots of different people with lots of different opinions coming together um yeah what would that what would that look like I don't know where you I don't know but certainly in the microcosm of one and twos yes somebody probably knows a lot more about that than I do thank you very much everyone for being here tonight thanks for coming out the rain thank you thank you for having [Applause] [Music] me

[Music]