Withthe public release of large language models like Chat GPT puttingArtificial Intelligence (AI) firmlyon our radar, this episode explores what benefits this technologymight hold for statistics and analysis, as well as policymaking andpublic services.
Joininghost, Miles Fletcher, to discuss the groundbreaking work being donein this area by the Office for National Statistics (ONS) and acrossthe wider UK Government scene are: Osama Rahman, Director of theONS Data Science Campus; Richard Campbell, Head of ReproducibleData Science and Analysis; and Sam Rose, Deputy Director ofAdvanced Analytics and Head of Data Science and AI at theDepartment for Transport.
Transcript
MILES FLETCHER
Welcome again to Statistically Speaking, the official podcast ofthe UK’s Office for National Statistics. I'm Miles Fletcher and, ifyou've been a regular listener to these podcasts, you'll have heardplenty of the natural intelligence displayed by my ONS colleagues.This time though, we're looking into the artificial stuff. We'lldiscuss the work being done by the ONS to take advantage of thisgreat technological leap forward; what's going on with AI acrossthe wider UK Government scene; and also talk about the importanceof making sure every use of AI is carried out safely andresponsibly. Guiding us through that are my ONS colleagues - withsome of the most impressive job titles we've had to date - OsamaRahman is Director of the Data Science Campus. Richard Campbell isHead of Reproducible Data Science and Analysis. And completing ourlineup, Sam Rose, Deputy Director of Advanced Analytics and head ofdata science and AI at the Department for Transport. Welcome to youall. Osama let's kick off then with some clarity on this AI thing.It's become the big phrase of our time now of course but when itcomes to artificial intelligence and public data, what preciselyare we talking about?
OSAMARAHMAN
Soartificial intelligence quite simply is the simulation of humanintelligence processes by computing systems, and the simulation isthe important bit, I think. Actually, people talk about datascience, and they talk about machine learning - there's noclear-cut boundaries between these things, and there's a lot ofoverlap. So, you think about data science. It's the study of datato extract meaningful insights. It's multidisciplinary – maths,stats, computer programming, domain expertise, and you analyselarge amounts of data to ask and answer questions. And then youthink about machine learning. So that focuses on the development ofcomputer algorithms that improve automatically through experienceand by the use of data. So, in other words, machine learningenables computers to learn from data and make decisions orpredictions without explicitly being programmed to do so. So, ifyou think about some of the stuff we do at the ONS, it's veryimportant to be able to take a job and match it to an industrialclassification - so that was a manually intensive process and nowwe use a lot of machine learning to guide that. So, machinelearning is essentially a form of AI.
MILESFLETCHER
Sois it fair to say then that the reason, or one of the main reasons,people are talking so much about AI now is because of the publicrelease of these large language models? The chat bots if you like,to simpletons like me, the ChatGPT’s and so forth. You know, theyseem like glorified search engines or Oracles - you ask them aquestion and they tell you everything you need to know.
OSAMARAHMAN
Sothat's a form of AI and the one everyone's interested in. But it'snot the only form – like I said machine learning, some otherapplications in data science, where we try in government, you know,in trying to detect fraud and error. So, it's all interlinked.
MILESFLETCHER
Whenthe ONS asked people recently for one of its own surveys, about howaware the public are about artificial intelligence, 42% of peoplesaid they used it in their home recently. What sort of things wouldpeople be using it for in the home? What are these everydayapplications of AI and I mean, is this artificial intelligencestrictly speaking?
OSAMARAHMAN
Ifyou use Spotify, or Amazon music or YouTube music, they get data onwhat music you listen to, and they match that with people who'vebeen listening to similar music, and they make recommendations foryou. And that's one of the ways people find out about new music ornew movies if you use Netflix, so that's one pretty basicapplication, that I think a lot of people are using in thehome.
MILESFLETCHER
Andwhen asked about what areas of AI they'd like to know more about,more than four in 10 adults reported that they'd like to knowbetter how to judge the accuracy of information. I guess this iswhere the ONS might come in. Rich then, if I could just ask you toexplain what we've been up to, what the Data Science Campus hasbeen up to, to actually bring the power of artificial intelligenceto our statistics.
RICHARDCAMPBELL
ThanksMiles. Yeah, a few things that ONS has been doing in this verybroad sphere of artificial intelligence, and it's really in thatoverlap area that Osama mentioned with data science, so I'd pickout a few sorts of general areas there. So, one is automation. Youknow, we're always keen to look at how we can automate processesand make them more efficient. It frees up the time of our analyststo conduct more work. It means that we are more cost effective. Itmeans that our statistics have better quality. It's something we'vedone for years but AI offers some new opportunities do that. Theother area which Osama touched on is the use of large languagemodels, you know, we can get into the complexities of data. We canget much more out of data; we can complete tasks that would havebeen too complex or too time consuming for real data scientists.And this is good news, actually, because it frees up the datascientists to add real valuable human insights. Some of the placeswe've been using this. So, my team for example, which is calledreproducible data science and analysis, and we use data science andengineering skills to develop computer systems to producestatistics where the data is a bit big, or what I tend to call abit messy or a bit complex for our traditional computer systems. Weuse AI here through automation, as I mentioned, you know, reallymaking sure that we're making systems as efficient and high qualityas possible. Another thing we're interested in doing here is quiteoften we’re doing something called re-platforming systems. So, thisis where we take a system that's been used to produce ourstatistics for years and years and look to move it on to newtechnology. Now we're exploring with Osama's team the potential forAI to do a lot of the grunt work for us there to sort of go in andsay, right, what is going on in this system? How is it working, howwe can improve it? One other thing I'll mention, if Osama doesn'tmind me treading on the territory of his team, is the Stats Chatfunction that we've used on the ONS website. So, this is using AIto enable a far more intelligent interrogation of the vast range ofstatistics that we've got, so it no longer requires people to bereally knowledgeable about our statistics. It enables them to askquite open questions and to be guided to the most relevantdata.
MILESFLETCHER
Becauseat the moment, if you want to really explore a topic by gettinginto the depths of the data, into the granular data, you’ve reallygot to know what you're looking for haven’t you? This again is anoracle that will come up with the answers for you and just presentthem all ready for your digestion.
RICHARDCAMPBELL
That'sright. And I tend to think of these things as a starting point,rather than the whole answer. So, what it’s enabling you to do isto get to the meat of the issue a lot quicker. And then you canfocus your energy as a user of our statistics in doing the analysisthat you want rather than thinking “how do I find the rightinformation in the first place?”
MILESFLETCHER
Osama,that sounds like an intriguing tool. Tell us precisely how it worksthen, what data does it capture, what's in scope?
OSAMARAHMAN
Sothe scope is publicly available documents on the ONS website. Andthere's a specific reason for that. So, these AI tools, you canhave it look at the whole internet, you can have it look at subsetsof data, you can point it to specific bits of data, right? Andwhat's important for us is actually the work of the ONS, thatstatistics we produce are quality assured and relevant. And byproviding these guardrails where you know, Stats Chat only looks atONS published data, we have a degree of assurance that the datacoming back to the user is likely to be of good quality and notbased on who knows what information.
MILESFLETCHER
Becausewhen you use, to name one example, ChatGPT for example, the littlewarning comes back saying “ChatGPT can make mistakes, considerchecking important information.” And I guess that's fundamental toall this isn't it. These tools, as intelligent as they might be,they're only as good - like any system - as the information that'sgoing in the front end.
OSAMARAHMAN
That'sabsolutely correct, which is why we have these guardrails where,you know, the functionality on Stats Chat is focused on publishedONS information.
MILESFLETCHER
Thatdoes mean that something that's offered by an organisation like theONS does have that sort of inbuilt potential to be trustworthy andwidely used. But of course, you might say, to have a really goodtool it's got to be drawing on masses of information from rightacross the world. And it's interesting how, and you mentioned thatit's open-source data, of course, that's most available for thesetools at the moment, but you're seeing proprietary data coming inas well. And this week, as we're recording this, the FinancialTimes, for example, has announced that it's done a deal with one ofthe big AI firms to put all of its content into their database. Doyou think there's scope for organisations like the ONS around theworld to collaborate on this and to provide you know, reallypowerful tools for the world to exchange knowledge and data thisway?
OSAMARAHMAN
Sothere is collaboration going on. There's collaboration, both withingovernment - we're not the only department looking at these sortsof tools; there's also collaboration internationally. I think thedifference you know... our information on our website is alreadypublicly available. That's why it's on the net, it is apublication. But there's a difference in situation with the FTwhere, you know, a lot of the FT information is behind apaywall.
MILESFLETCHER
Yeah,it has a sort of democratising tendency that this publiclyavailable information is being fed into these kinds of sources andthese kinds of tools. That's big picture stuff. It's all veryexciting work that's going on. But I'll come back to you Rich justfor a second. What examples practically, because I think that theStats Chat project is still a little way off actually beingavailable publicly, isn't it?
RICHARDCAMPBELL
Yeah,I think it is still a little way off. So, I think the key thingthat we're doing at the moment and something we've done for years,but AI is helping is the use of automation principles. Just makingthings quicker. Now in a data science context, this might be goingthrough very, very large data sets, looking for patterns that itwould take an analyst a huge amount of time and probably far toomuch patience than they would have to find.
MILESFLETCHER
Sofor example, in future then we might find that - and this is oneissue that recurs in these podcasts - obviously about thelimitations of official statistics is they tend to lag.This is another wayof making sure that data gets processed faster. And therefore, thestatistics are more timely, and therefore the insights they provideare really much more actionable than perhaps they might be at themoment.
RICHARDCAMPBELL
Yeah,that's spot on. There's potential in there for pace of getting thestatistics from the point that the data exists to getting it intopublished statistics. There's potential there for us to be able tocombine and bring more sources together. There's also some behindthe scenes stuff that helps as well. So, for example, quite oftenwe are coding up the systems to produce new or improved versions ofofficial statistics. And we're looking at the possibility of AIspeeding up and supporting that process, perhaps for example, bygiving us an initial draft of the code. Now, why does that matterfor people in the public, you know, does anybody actually care?Well, what it means is that we can do things quicker and more tothe point we can focus the time of our expert data scientists andother analysts in really helping people understand the data and theanalysis that we're producing.
MILESFLETCHER
Okay,so lots of interesting stuff in the pipeline there. But I’d like tobring in Sam now to talk about how AI is actually being used ingovernment right now. Because in your work Sam at the Departmentfor Transport, you've actually been working on some practicalprojects that have been gaining results in the realworld.
SAMROSE
Wehave - we've been doing loads actually, and my poor team probablyhaven't had any time to sit still for the last 18 months or so. AndI think like most ministerial departments, we're doing lots andlots of work to automate existing processes, so much like Rich hasalluded to in your space, we're looking at the things that take upmost of the time for our policy colleagues and looking at how wecan automate those. So, for example, drafting correspondence, orautomating policy consultation processes, or all of that kind ofcorporate memory type stuff. Can we mine big banks of data be ittext or otherwise and summarise that information or generate newinsights that we wouldn't have been able to do previously? But Ithink slightly more relevant maybe for you guys, is the stuff we'redoing on creating new datasets or improving datasets. So, a fewthings. We're training a machine learning model to identify heavygoods vehicles from Earth observation data. And that's because wedon't have a single nationally representative data set that tellsus where these heavy goods vehicles park or stop outside ofexisting kind of service stations, and what we want to understandis where are those big areas of tarmac or concrete where they'reall parking up as part of their routine journeys, so that we canlook at when we're rolling out the green infrastructure for heavygoods vehicles, we're looking at where the important places that weneed to put that infrastructure are. And that data doesn't exist atthe moment. So we're using machine learning to generate a newdataset that we wouldn't otherwise have.
MILESFLETCHER
Andhow widespread are these kinds of projects across government in theUK now?
SAMROSE
SoI think that there are loads of different things and I wouldn't beable to speak on behalf of everybody but I know lots of differentareas of government are looking at similar kind of automation andproductivity projects like our kind of drafting all of theknowledge management area. I think there's things like Osamaalluded to where DEFRA for example, I think they're using Earthobservation data to assess biodiversity for example. So, there'slots of stuff that's common between lots of government departments,and then there's lots of stuff that's very specific to individualdepartments. But all along the way there's lots of collaborationand working together to make sure we're all learning continuouslyand where we can collaborate on a single solution that weare.
MILESFLETCHER
Iguess one of the central public concerns about the spread of AIonce again that it will cost jobs, that it will do people out ofthe means of making a living that they've become used to. And Iguess from government's point of view, it's all about doing much,much more with the resources that we have and making governmentmuch more effective.
SAMROSE
Yes,absolutely. And it's not necessarily - and I think Rich mentionedthis earlier - it's not necessarily about doing our jobs for us.It's about improving how we can do our jobs and being able to domore with less, I think, so freeing up the human to do the bit thatthe human really needs to do and enabling the technology to dotheir very repeatable very automatable parts of the job. Andindeed, in some instances, this technology can actually do the workbetter than humans. So be it identifying really complex patternsand datasets, for example. Or a good example from us in transportis we've trained machine learning model to be able to look atimages of electric vehicle charge point installations and be ableto identify that similar or the same image that has been submittedmore than once. Now that's estimated to have saved over 130 manyears of time, you know, that's not a task that we would have beenable to do with just humans.
MILESFLETCHER
Andyou would have to be pretty alert as a human and have a very highboredom threshold to process all that material yourself and spotthe fraudsters.
SAMROSE
Yeah,well, quite. And that's, I think, a really nice example of whereagain, it's not taking our jobs, but it's enabling us to dosomething that we wouldn't have been able to do previously andimprove the service that we're providing.
MILESFLETCHER
Now,our ability collectively, whatever sort of organisation we'reinvolved in, our ability to make the most of AI depends on ofcourse having the right skills, and Osama I guess this is where theData Science Campus comes in as the government's Centre ofExcellence for data science, principally, but I guess also in thiscontext, artificial intelligence as well. What work have you beeninvolved in to make sure that the supply of those skills andknowledge is on tap for government?
OSAMARAHMAN
Sofirstly, I would say we are a (one) centre of excellence withingovernment. I think you know, what's been brilliant to see sincethe campus was set up has been that actually more and moregovernment departments have excellent data science, AI teams. Samleads one at DfT. There is, of course, 10DS (or 10 Data Science) atnumber 10 [Downing Street]. There's a Cabinet Office team. So,there's lots of teams that now work in this area. Some of the stuffwe've been doing is we have various training programmes that wehave run. We have senior data masterclasses so that actually,senior leaders within government can understand better the power ofdata. 10DS, Sam's area, have all been running hackathons, whichactually improve skills as well. So, it's no longer just us who arebuilding capability. I think it's great to see that acrossgovernment and across departments there are teams improving skillswithin their departments, bringing in others from outside to workwith them. So, there's a lot going on there.
SAMROSE
Justreally quickly, it's important to think that skills are not justskills of data scientists, but skills of everybody's ability to usethis kind of technology. There's a lot of work going on at themoment looking at what we need to do both internally to government,but also out there in all of our sectors to make sure that ourworkforce has the skills it needs to be able to more rapidly kindof adopt and be able to take advantage of all the benefits thatthis technology brings to us. I mean from a very personal point ofview, and I don't really know all of the answers to this, but youknow, I'm thinking about what actually, if large language modelscan help us to generate efficient code, then actually, what skillsdo I need in my data scientists? If it's not writing code, is itactually the analytical thinking and being able to understand howto apply these kinds of technologies? So, I think it changes whatwe need in the workforce that we have.
MILESFLETCHER
Inevitably,though, if we're talking about this kind of technology being rolledout across government and thereby increasing the power ofgovernment to know more about more people, then concerns obviously,about the ethical use of data come in...
RICHARDCAMPBELL
Maybeif I can just come in on that one Miles. Using data safely andresponsibly - it's built into our very DNA in ONS and acrossgovernment. And our keenness to sort of learn how to do new toolsnew techniques is always going to be tempered by our need to ensurethat we are responsibly using the data that's been entrusted to us.And I think we need to sort of strike a balance here. We need toensure that we don't take this responsibility as an excuse to nottry and adopt new technology such as AI, but it also means we haveto do so with care and responsibility and to do it at anappropriate pace. The key thing, I think, for me is ensuring thatwe can retain control of the data that we've been entrusted with.And so, understanding what AI is doing with that data, consideringwhat data we're giving access to it, what data is being processed,and what data is being generated. And this is really at theforefront of our minds and our collective use of this. I think ourapproach - and Osama touched on this earlier - is to sort of benovel and start with open source and non-sensitive data first, sothat will help us learn how we can effectively use it before we goon to some of the more sensitive data that wehold.
SAMROSE
Wehave to have ethics and data protection at the heart of everythingwe do, which then does have the tendency I think necessarily toreduce the pace of our ability to roll things out a little bit. Butas government we do, I think have more responsibility. We can'thave those kind of oops moments that some of the big tech companieshave had when they're trying to reverse engineer the data to removebias and that you know, things like that that then fundamentallyundermine the output of their models. I think when you're doing ajob that affects individual people, and providing services thataffect citizens then we don't really have the luxury of getting itwrong like that, and we have to try to make sure we get it rightfirst time. So, all of the things that Richard said about startingwith, you know, safer datasets and working our way up before wedeploy these models is kind of fundamental to how we're going tolearn and ensure that we're doing it safely andsecurely
MILESFLETCHER
Osamawhat's your take on the ethics question?
OSAMARAHMAN
Firstof all, I would echo everything just said. You know the StatisticsCode of Practice is an annex to the Civil Service Code, it appliesto all of us not just statisticians - I'll point that out. It is Ithink, not just in the ONS, I think for analysts and datascientists and specialists across government, this is kind of builtinto their DNA. Central Digital and Data Office has put togetherguidance and circulated it across government on the safe use of AIwithin government. So, within government, we do take this quiteseriously. And then actually in terms of the use of some of thesetechniques, I think pointing these tools at data and informationthat we know is accurate is an important starting point - so havingthose guardrails. If it's going to be used for decision making,then having a human in the loop is quite important to make surethat the use is ethical. So, there's a bunch of safety checks thatwe do put in which I think allow for us to have some assurance thatthe use of these tools will be safe and ethical.
RICHARDCAMPBELL
Ithink just as one additional point is you know; this isn't a newchallenge for us. It's a different flavour of a challenge that wefaced in considering new technology in the past. So, we can thinkin fairly recent times the use of cloud technology to securely andsafely store data. If we go further back the use of the Internet,go back further, again, the use of computers to hold data. And whatI think we've demonstrated time and time again, is that we doapproach these things responsibly and maturely. But we do findopportunities to use all of them to improve the quality ofstatistics and analysis and the service that we offer thepublic.
MILESFLETCHER
Lookingto the future then, and this is a very fast-moving future ofcourse, I'd like to get your takes on also where you see us in fiveyears’ time in 10 years’ time with this. I mean starting with theOffice for National Statistics – Osama and Rich particularly onthis. How will we start to see the published statistics and the bigkey topics, but also the granular insights that we provide on allkinds of areas. How will we see that changing and developing do youthink? Where are you going to put your money?
RICHARDCAMPBELL
Ithink predicting the future in this way is quite a dangerous game.I’m thinking back to you know, if we had this podcast in the year2000 and we asked ‘’how would the internet form part of our workinglives?’ We would have predicted something which would have beenquite different from the impact that it had. Saying all that Ithink it will make a fundamental difference to the way that wework. I see that it will be integrated in the day-to-day tasks thatwe do in a similar way that we used computers to speed up andchange the way that we produced statistics. I think it will enableour users to far better interact and engage with data and analysis.So, it will be less of us producing a specific finalised productfor them, and more for them to be able to sort of get in askquestions, probe and really, really interact. And I think lastly,it will give us more potential to work and analyse data because onething, and I think this is really important to say, AI will givemore opportunities for analysts. It won’t take them away. It willgive them more space, more tools to work with to produce better,more complex, more useful datasets and analysis for ONS and for itsusers.
SAMROSE
Iwas just going to add that I think it will fundamentally change thenature of what we do. A little bit like Rich said, the sort of workthat we do will be different, but really critically, I think in afew years’ time we won't really notice that change. I was thinkingthat most people have forgotten that 10 or more years ago beforeyou left the house to go somewhere new, you would have consultedyour map. Whereas actually nobody, or very few people, do thatanymore. So, I think we're going to forget very quickly that lotsof what we will be doing will be AI driven.
MILESFLETCHER
Soit's a big evolutionary step forward, if not quite a revolution. Doyou agree with that Osama?
OSAMARAHMAN
Absolutely,because some of us have actually been using sort oftransformers-based models, which is what these large languagemodels are based on for... My team has been working with those forat least the last eight years. But I wanted to just pick up on whatRich just said. And it is an evolution right. And you can'tseparate the tools from the data. And one of the things we'regetting now is data that is much more granular and of much highervelocity than the data we were used to. So that allows us to lookat things at a more local level, at a more timely level. What I docompletely agree with Rich on is actually a lot of these tools andmethodologies allow the technical production of statistics to getmore efficient, which then allows you to produce more statistics ata disaggregated level - at a regional level or local authority arealevel or looking at different sub populations. It allows us toupdate statistics more frequently. But then also what it allows usto do, because it's not just about the production of thestatistics, it's about what those statistics actually tell you isgoing on. And I think it allows the people we have at the ONS andother government departments to spend more time on the real valueadded which is “what does this mean?”
MILESFLETCHER
It'sinteresting if you're researching a particular topic, it must begood to sort of evolve your methodology quickly and to refine yourprocesses on the run as it were to explore a particular topic. Onething of course we need in statistics is consistency of methodologyand approach. Does that limit do you think, either of you, theability for statistics to get more insightful to get more germaneto issues because we have to stick to accepted methodologies toprovide that consistency over the long run?
RICHARDCAMPBELL
Idon't think it does Miles. I mean, you're right there. There'salways a challenge for us in that, that consistency is reallyimportant, that comparability in a time series. Equally, users dowant us to look for improvements, more detail, whether that'sgranularity or whatever else. And actually, we've got a really goodsuccessful track record of both maintaining the consistency of ourstatistics, while at the same time introducing new and improvedmethods. We do it with GDP. We do it with inflation, we do it withpopulation, it's something that we do time and time again. And,actually, I think automation AI offers up some really excitingopportunities here in terms of methods that can be applied. There'sactually an element of it, which will help us in the understandingand documentation and consistent application of the methods aswell. It’s perhaps one of the less – if you don’t mind me using theword - “sexy” applications of AI but using it to ensure that ourdocumentation is absolutely spot on and done quickly. To ensurethat we are applying methods really quickly and consistently. Ithink AI offers us potential to do that evenbetter.
MILESFLETCHER
“Sexy”in the particular way that we refer to progress in datascience.
RICHARD CAMPBELL
Yes, quite.
OSAMARAHMAN
CanI just come in on this? And it's possibly worth using a specificexample of putting out statistics on prices. In the old days you’dbasically have people going out into the field, and that's whereyou'd find a basket of goods, and using pen and paper would collectprices. Now where a lot of national statistical organisations aregoing to is actually getting scanner data, because most things whenyou pay for them nowadays in many parts of the world, it's scannedfirst, electronically rather than rung up through a cash registerof some sort. So, scanner data provides a lot of information aboutwhat is being purchased, and at what price it's being sold atvarious retail outlets. And so, you have this data which again ismuch more granular and has much higher velocity then price data youcan collect through surveys, and you know, how you integrate thatinto the production of pricing statistics and other economicstatistics is really, you know, a really interesting question andwork that a lot of national statistical organisations are workingon. So, there's still the basic methodology remains the same. It'syou know, kind of defined a basket of goods, but you expand thescale of the basket, we'll get prices on at what each of theelements are those baskets are being sold at, and then produce aprice measure an inflation measure, right. But these tools and theincreasing quantity of data allow us to do that. But you know, thebasic methodology is kind of the same, but actually the increase inthis data allows us to do that in kind of a different way. It's anevolution.
MILESFLETCHER
Itdoes all suggest though, that perhaps the survey might finally bereplaced - the big social surveys that the ONS runs. Do you thinkthat the surveys days are numbered, therefore because ofAI?
RICHARD CAMPBELL
No.
OSAMA RAHMAN
No.
[LAUGHTER]
MILESFLETCHER
Aresounding no.
RICHARD CAMPBELL
Thatwas a resounding no, and it's not a pre-rehearsed one. And maybeI'll just take us back Miles. So, if we went back about the bestpart of 10 years, everyone was talking about big data. You know,the days of a survey was gone. What we needed was these big,complex, sometimes quite messy data sources that were collected fora variety of other reasons, and that we could utilise those to sortof answer all of the statistical questions that we had. Now, whatwe found out actually is that yes, these data sources can give us alot of potential; data science is helping us make the most of them;AI is helping us make even more from them. What we also learnedthough, is that they work best when they're complementing thesurveys, rather than trying to replace. Think of it a bit as horsesfor courses. Actually, though, I want to give an example of whereAI might be able to help us improve the response rates on surveys.So, AI might be able to help respondents navigate through some ofthe surveys, helping them understand what it is that they're beingasked. Helping them answer a bit more efficiently. So that mightactually remove a barrier that some people, some businesses have torespond to surveys. So, you never know we might see a bit of anuptick in response rates with a bit of AI’s help.
OSAMARAHMAN
AndI think the other thing I would add is what surveys areparticularly good at is getting information on the extremes of thedistribution. It's great if you think everything's going to begenerated through digital footprints data and online services, butactually not everyone... some people... apparently dumb phones arecoming back into fashion. Or there are groups that you know, forwhatever reason, are not picked up in other forms of data. Andactually, surveys are really important for accessing, gettinginformation about hard to reach groups at the end of thedistribution.
MILESFLETCHER
Ithink that’s kind of reassuring that for all the promise of AI inthis brave new world, that we hope won't be a dystopian future, butwhether it will deliver all those things that we've been talkingabout in terms of better insights, faster statistics, and all that.It's still good to hear though, isn't it, that there is nosubstitution from speaking to real human beingsdirectly.?
OSAMARAHMAN
Iagree entirely.
MILESFLETCHER
Well,that's it for another episode of Statistically Speaking and, insummary, I suppose the use of AI feels like a natural evolutionwith a number of potential benefits, and potentially huge benefits,but with its adoption we need as always to be thoughtful andethical. So, thanks to all our guests: Sam Rose, Osama Rahman, andRich Campbell, and of course, thanks to you as always forlistening. You can subscribe to future episodes of this podcast onSpotify, Apple podcasts and all the other major podcast platforms.You can also follow us on X - formerly known as Twitter - via the@ONSfocus feed. I’m Miles Fletcher and from myself and producerSteve Milne. Until next time, goodbye.
ENDS
