YT YouTube2Text

YouTube2Text → Transcripts → AI s Data Wall

AI s Data Wall — Transcript

by James Dwayne Group · 5,320 words · 770 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:02welcome welcome to Sweet speak so I'm
0:06excited this is a show that has been a
0:09long time coming uh excited to have uh
0:12David with us David HRA woan and I hope
0:15I got that pronunciation you did perfect
0:18jie uh David's a a tech expert educator
0:21and strategist with a passion for AI
0:24cyber security and data governance he's
0:26also a lecturer University of Texas and
0:28serves on the boarded
0:30htri David serves as the principal
0:33adviser at Informatica advising
0:35customers on AI ready strategy uh to to
0:39get them we're going to dive into so
0:41much that he's experienced around AI his
0:44background includes roles at deoe Grant
0:47Thorton and Dell where he help
0:49businesses harness data mitigate risks
0:52and stay ahead of the curve he's also an
0:54active Mentor speaker and author sharing
0:56his insights on everything from data
0:58privacy to resp responsible AI so get
1:01ready to dive into the world of data and
1:04Innovation with David now here's the
1:06disclosure part of this show the views
1:09and opinions expressed by David on this
1:11show are solely His Own do not reflect
1:15the official policy acquisition of
1:17Informatica University of Texas at
1:19Austin or any other organizations he's
1:22affiliated with these are purely his
1:24thoughts and his personal take on things
1:28so um want that out out there right away
1:30so all of that David welcome to Sweet
1:34speak thank you very much J excited to
1:36be here today and we're gonna focus on
1:39the AI data wall so you may be tuning in
1:43and go okay JD what the heck is the AI
1:46data wall my question for you let's
1:48start off what is that AI data wall and
1:51why is it gaining attention today yes
1:54well AI data wall is this uh notion that
1:59there will be a point when AI models can
2:03no longer improve because we've run out
2:05of data it's a wall we can't upscale it
2:09um and it can comes from multiple uh
2:12angles right it could be that the data
2:14is
2:15scarce all of the data that's been
2:17scraped that could have been used to
2:18train model have been scraped it could
2:21also mean that there's architectural
2:23limits even with the data that we have
2:26even if we could have acquired all the
2:28data that is in the world
2:30it is difficult to structure your
2:33software your Hardware such that you can
2:35learn from all that data there's
2:37infrastructural challenges there's not
2:39enough storage there's even some a
2:41couple years ago uh it's no longer here
2:43because um Mo's law and you know storage
2:46uh increase but there was a point when
2:48we said even if we could theoretically
2:50capture all of the data in the world do
2:52we have the hardware to be able to store
2:54all that and lastly is the human issue
2:57even if we have all of the data and we
2:59can't Lish all everything from Hardware
3:01software how about the humanware issues
3:04we still have ai literacy we still have
3:06trust issue and ethical concerns leading
3:09to poor adoption and data misuse all of
3:11this combined together creates this
3:13notion that there is going to be a wall
3:16through which we cannot go through
3:17unless we can go through that by
3:19increasing our data management
3:21capability and data governance uh
3:23disciplines we won't be able to improve
3:25and capitalize on
3:27AI so let's peel that
3:30a little bit so how does the AI data
3:33wall impact AI development and model
3:36performance so a lot of the AI companies
3:39now have been investing a lot of money
3:41into development and um 80% of the
3:44so-called model development comes from
3:47data cleaning the data getting the data
3:49uh sharing it augmenting it Etc and AI
3:53models will space diminishing returns at
3:56some point because adding more data may
3:59not not lead to better
4:01results unless we come up with new AI
4:04architecture which at this point uh
4:06that's that's going to be a couple years
4:08it will continue to happen but I think
4:09even within the current uh environment
4:12we are expecting some people have
4:14estimated Maybe by 20126 when some
4:16people expected more there's some
4:18development interestingly with deep seek
4:20R1 that illustrate as well as synthetic
4:22data that actually that wall is actually
4:24a little bit further out right um and
4:27even if we have all the data there's
4:29still bias issue generalization issue
4:31and scaling issues so in other words in
4:34short it will create um a diminishing
4:38return on your investment in Ai and data
4:41infrastructure unless you know how to do
4:43it right and it is the old adage right
4:46garbage in garbage out so at the end of
4:48the day you still have to all of the
4:50disciplines that we've learned about
4:52data management all of those are still
4:54relevant um you can use AI to help
4:56overcome some of those tasks but you
4:58still have to understand underlying
5:00principles of what good data management
5:02and governance
5:03means it's weird to think of a DA data
5:06wall with with AI because it seems so
5:10vast but I I'm curious which Industries
5:13or or or AI applications have already
5:16hit a data wall what are the
5:18consequences of that yes so we can kind
5:22of divide maybe into two major uh
5:25categories um there are industries that
5:28are much more data intens
5:30has been in the data game for a long
5:32time has a lot of regulations around it
5:35so I we think of healthcare banking uh
5:39some sectors of Life Sciences uh a
5:42portion of the you know geophysics etc
5:44those um disciplines already have pretty
5:48robust data
5:49infrastructure um they generally face
5:53more of the issue of how can they become
5:55better how can they make the people and
5:57process catch up with the technological
5:59side how do they make sure that
6:01regulations are updated in such a way
6:03that it doesn't prohibit Innovations
6:06okay and then there's the other side of
6:10the category which is the lager industry
6:13so these are Industries where data has
6:15not yet been digitized uh digitalization
6:18is still a long way progress uh sectors
6:21such as perhaps agriculture construction
6:24parts of energy uh these late adopting
6:28uh Industries are still dealing with uh
6:31operational technology challenges how
6:33can we capture the data using drones
6:35using cameras Etc um then there's also a
6:40third dimension that touches both of
6:42these categories which is data sharing
6:44which is an exciting one where you can
6:47allow people to share the AI training or
6:50AI Learning Without sharing the actual
6:52data and of course this is a big driver
6:55of innovation a lot of reasons why
6:57companies cannot just aggregate data
7:00which rightfully should be controlled is
7:02because of concern about data breach or
7:05confidentiality breach with some of
7:07these so-called privacy preserving data
7:09sharing and analytics you could share
7:11the AI Learning Without sharing the
7:13actual data so those are the type of
7:16Industries where um we may see impact as
7:20well as opportunities of data
7:22one so how do we ensure we have the
7:25right data to overcome these challenges
7:27that you're you're outlining
7:30yeah so I I'll um let me break it into
7:33perhaps four ways of looking at this
7:36okay um so first of all if we wanted to
7:39have the so-called right
7:41data the the definition of the right
7:44data uh has to do with relevance number
7:47one right so the diversity of the kind
7:49of data that you have um up until the
7:53era of generative AI most of the data
7:56engineering data management has been
7:57focused around structured data data that
8:00exists in your
8:01databases or semi structure data that
8:04are in Json files or in log files that
8:08are not necessarily tabular but still
8:11follows a particular pattern okay with
8:13the generative AI now we are we're
8:17having to capture a much broader set of
8:20data that are unstructured right uh
8:23videos images um you know text full text
8:28documents ET
8:30okay so that's the First Dimension um
8:32the second dimension of the so-called
8:33right data is whether or not people have
8:37the right to use that data okay so this
8:41has to do with privacy and intellectual
8:43property um organizations such as open
8:47AI for example and there is a actually
8:49well documented public uh showing of
8:51this when the CTO of open AI was
8:53interviewed by Wall Street Journal and
8:55asked how do you train the model the
8:58Sora model the video generation model
9:00and the CTO said in front of this is a
9:03public interview that's on YouTube uh
9:05it's using publicly available
9:06information and when the interviewer
9:08asked so like YouTube or you know which
9:11one and she said um I don't
9:14know so you know you know so the right
9:18data besides the type of data but also
9:21how do you acquire that data okay how do
9:23you assign attribution to the owner of
9:27the intellectual property that
9:28rightfully belongs to the people who
9:31whose data it is you know was captured
9:34okay then there is the quality so yes
9:37you have acquired the right type of data
9:40you have acquired it in the right way
9:42and now you have it all but it is in a
9:45raw format it is not high quality you
9:47need to apply data quality management
9:50you need to apply things such as data
9:52observability uh so that you can detect
9:55and correct biases errors and anomalies
9:57in real time and for the longest time
10:00data Engineers produce the D what they
10:02call data products they pass it on to
10:04the bi or AI data scientists and then
10:08you know they don't handle it from that
10:09point on with the AI era you don't just
10:12look at data as an input you also have
10:14to look at when that AI is passed
10:16through an inference AI model there's
10:18outputs and you have to compare the
10:20output to the input in order to figure
10:21out is the model effective you know is
10:24it starting to have a drift is it biased
10:27does it cross the threshold of anomaly
10:30so that you have to retrain the model
10:32okay and finally the I the last piece is
10:36the human the human oversight Active
10:38Learning um after doing all of that the
10:41so-called right data ultimately is data
10:45is that has passed through the lens of
10:47human oversight human interpretation and
10:50human decision and I was rewatching the
10:52movie Sully last night about you know
10:54the the pilot that landed in Hudson
10:56River and it's kind of interesting how
10:58in in the movie
11:00they challenged him using a simulation
11:02model said simulation computer
11:04simulation shows that you you should
11:05have been able to land and then they did
11:07a pilot testing where a pilot was trying
11:09to fly simulation but he said all of
11:12those are not human uh you know they
11:14were told what to do beforehand I was
11:16never prepared to do this this is the
11:18first time we have this
11:20unprecedented situation and if you want
11:23to find human errors make it human and
11:25so they did and they said all right now
11:27all the pilot simulation we will give
11:28you 35 seconds before you can react to
11:31account for the shock and when they did
11:34that suddenly everybody failed to land
11:37in in the airport that they prescrib so
11:38again that that so those are kind of the
11:40four dimensions of how you make sure
11:42that you're got you got your data right
11:44what world do synthe synthetic data play
11:47in overcoming this AI data wall oh
11:50synthetic data is an exciting thing as
11:53as well as a scary thing right so the
11:55concept of synthetic data is that you
11:57can use AI to to generate fake data that
12:02mimics the original data okay so if we
12:06let that sync in for a
12:08moment we will have a data wall where
12:12real data is run
12:13out but if you understand how to create
12:16synthetic data then you will never run
12:19out of data in
12:21theory however of course we know that
12:23the synthetic version of the data is not
12:25as good will never be as good will never
12:27be as real as the real
12:30data so does that mean that this is not
12:34a valid technique so this is where a lot
12:38of experts are on different sites and
12:40everybody has a great argument about
12:41this um in general synthetic data can
12:44help you enhance privacy because you
12:46don't use the real data so you prevent
12:48Reliance on the sensitive pii you can
12:52also improve model performance because
12:55for example for bias if you have a data
12:57and your population is majority
12:59population 90% of the data and minority
13:01populations all comprise only 10% of
13:03your data that's going to impact the
13:05accuracy and bias of the model so you
13:07use synthetic data to train up based on
13:10this little amount of data kind of
13:12increase it just enough to hit the
13:14threshold of
13:16stability um all of that is good and
13:19also there's some interesting research
13:21that shows that with synthetic data you
13:24could still have a good model
13:26performance even though in theory
13:29synthetic data degrades over time
13:31especially if you have synthetic data
13:32and then you have synthetic of synthetic
13:34data you know since open AI came to the
13:36picture since chat GPT become popular uh
13:39did you know that Google search
13:41performance has gone down because
13:44suddenly everybody's using chat GPT to
13:46generate synthetic information synthetic
13:51uh contents and publish it on the
13:52internet so now Google is figuring out
13:54oh wow suddenly there's a thousand times
13:57the number of articles about a
13:59particular subject and therefore the
14:01accuracy of its search has degradated
14:03right however if we take for example
14:05here's I'll just illustrate one example
14:08let's say the real data said if you
14:10apply heat to a piece of metal it will
14:13expand so that's a true statement real
14:16data you use synthetic data to generate
14:18the opposite of that and saying oh
14:20therefore if I apply Cod into a piece of
14:24metal it will contract and let's say
14:27that that is s it turns out that models
14:30can actually improve based on that even
14:33though that statement is not 100%
14:35accurate because there is you know not
14:36necessarily in metal but with water
14:38there is a thing called anomalies of
14:40water where it actually will violate
14:42that rule when the water is between zero
14:44and 4 degrees celsi okay but again aside
14:47from that so so even if the data is not
14:50100% accurate AI can learn from a second
14:53order knowledge so the relationship with
14:57these statements instead of the the this
14:59statements themselves is valuable to
15:03increase the intelligence level of the
15:06model so those are all kind of the
15:08opportunities with synthetic data to
15:10overcome AI data wall yeah
15:14interesting we when we talk data we
15:16always have to cover privacy ethical
15:20pitfalls so how should companies
15:21approach data collection today to to
15:24avoid that to avoid privacy to avoid you
15:27know the the ethical p that that are
15:31inherent yes
15:33um when you capture it starts from the
15:36point when you capture the data
15:38companies need to have transparency they
15:41need to explain to the end users of this
15:44data what is the purpose of capturing
15:47this information right so you have to
15:49have a not notice this privacy notice
15:51when you capture information you have to
15:53request their consent if you ever plan
15:56to use this data for purposes other the
15:59primary purpose right so there's a
16:01concept primary purpose if somebody's
16:03transacting with a company and they're
16:04entering their payment information
16:06that's the primary purpose they know
16:08they knowingly give the information so
16:10that they can complete the transaction
16:11but there is also secondary purposes
16:13meaning if the company then takes that
16:15data and use it for targeted advertising
16:17that is not the original purpose of
16:19somebody giving that data so that
16:21secondary you they have to request
16:23consent from customers on that secondly
16:26once you have that data you need to
16:28apply data minim ization you need to
16:30collect only the necessary data and
16:32nothing more and if you're leveraging AI
16:36you need to make sure that all of your
16:38AI infrastructure captures the data in
16:40such a way that it limits the exposure
16:42of that data to the real world okay so
16:45that's data minimization if you're
16:47training that on AI using that data then
16:51use privacy preserving techniques so it
16:53used to be everybody would just you know
16:55hey collect the data aggregate data and
16:56use it to train a large language model
16:58now nowadays uh in 2023 the White House
17:01issued a national strategy to promote
17:04privacy preserving data sharing and
17:06analytics it's a long word a long
17:09sentence ppds say for
17:11short and a part of that includes things
17:14such as Federated learning and synthetic
17:16data techniques that can allow you to uh
17:19perform analytics on anonymized data if
17:22you will and and the identified form of
17:24data and so nowadays uh in some of the
17:28you know important projects where you
17:30have large scale data collection
17:32especially for federal and state
17:34government um uh institutions are
17:36required to at least account for the use
17:39of these type of Technologies so in the
17:41old days you could just okay I'll
17:42collect everybody's data into one place
17:44and then I'll run whatever Master data
17:47matching or some kind of know AI machine
17:49learning but nowadays you should not
17:52just do it you should not take for
17:54granted the fact that you are able to
17:56apply Federated learning so that you can
17:58apply joint learning without bringing
18:00the data and only bring the data if it
18:03is a smaller subset that's absolutely
18:05necessary for a particularly important
18:08decision making for example um so and so
18:11those are kind of the three major things
18:13there are some other things that are
18:14kind of more a broader level and this is
18:16kind of going Beyond data discipline
18:18more into cyber security so things such
18:21as uh governance uh it governance
18:24security governance um you know hacking
18:27penetration testing Bounty programs to
18:30make sure that um if you have your data
18:32breach and it's in the in the dark web
18:35that you can find out and you can uh
18:37identify the source of that because most
18:39companies unfortunately when it comes to
18:41data breach um we won't know until too
18:45long before we could do something about
18:48it so let's talk about that so there's
18:52these data limitations how companies
18:55like like D or deoe adapted to these
19:00limitations yeah so
19:04um obviously they they have um
19:07technological they're always staying on
19:09the edges of Technology right uh
19:12companies all the hyperscalers
19:14Informatica is the Switzerland of data
19:15for example we can you know process
19:18allow companies to be able to process
19:21the
19:22data um some of maybe I'll I'll so
19:25beyond the classical perhaps I'll just
19:27share a little bit of some some of the
19:28things that I've seen I've talked
19:30already about privacy enhancing
19:32Technologies uh Federal learning Etc I
19:35think also one of the most um exciting
19:39possibilities now is Edge AI Edge AI is
19:43the application of AI to be deployed on
19:47a local server so we've seen a big shift
19:50from on Prem to the cloud over the past
19:52couple of years yeah and because of that
19:54a lot of the AI infrastructure AI models
19:57as a service are being serve over the
20:00cloud and very few thought that um that
20:05we can you know have a strong Edge AI
20:08that that it's too expensive Etc however
20:12with the advancement of um Technologies
20:15such as Jetson Nano from Nvidia right
20:17which is an edge AI for Edge inference
20:20and the discovery of deep seek R1
20:24becoming an open source AI model you
20:27could load that deeps R1 the open source
20:31version into a local Edge AI device now
20:35I have heard I don't know if this is
20:37true or not somebody told me that they
20:38could even load beep seek onto a
20:42Raspberry Pi which is a very lightweight
20:44Hardware I don't know if it's true I
20:46haven't seen and validated it myself the
20:49point is we don't even have to do that
20:50you can buy a couple hundred dollars of
20:51a a Jetson device from Nvidia and load
20:54up at R1 and voila you have a local Aji
20:58so I I think what's going to happen with
21:00companies you know the hyperscalers and
21:02you know the big four McKenzies Etc and
21:05and and Technology firms like Dell and
21:08Dell especially you know there's going
21:09to be a lot of interest in Edge AI
21:12because the main concern about data for
21:14AI is security and the safest way to
21:17really make sure that that data will
21:19never even be used for training outside
21:21of your own purposes is to deploy your
21:23own inference AI model and Edge AI
21:26becomes uh the mechanism to do
21:30that so human oversight always I feel
21:34like needs to still be there but how
21:38today do you think can AI system systems
21:41become self- sustaining in data
21:43collection or is the human oversight
21:47there and I'm not losing my
21:50job yeah um this is a yeah this is a
21:55very touchy subject yes last week in my
21:58class happened to got to a point in the
22:00semester where I brought the students at
22:04UT Austin to a point of uh identity
22:07crisis oh no we've gotten to the point
22:09where we showed all the things that's
22:11possible with
22:12AI and um it was um and it was
22:17intentional I wanted them to Grapple
22:19with this issue right and if the AI can
22:22produce better music better video better
22:26painting better writing than humans then
22:30what good is
22:33humanity and this week they're they're
22:35doing take-home exam writing their
22:38thoughts on that and trying to produce
22:42products digital products to overcome
22:46the
22:47limitations um I can say so I've been
22:50blessed to be able to Garden in the data
22:53and AI field almost my entire
22:56professional life it has taken my my
22:58career my personal development to the
23:00Moon in
23:02back uh but I can take all of that all
23:06of my career and accomplishment and bet
23:09all those on this one truth AI will
23:13never replace
23:15Humanity thank you
23:18yeah uh just understanding as I deal
23:21with developing deploying using AI it
23:25will never do because um there is the
23:29second AI which I call authentic
23:31identity that's the humanity AI the
23:34artificial intelligence is a
23:36technological Ai and that will be the
23:39subject of my new book AI squared which
23:41is you know augment authentic int
23:43identity with artificial intelligence
23:46will produce exponentially much more
23:48powerful impact that can change the
23:50world for the better but that means that
23:53we have to come to grip with our self
23:56authentic self that means a personal AI
24:01ethics code of ethics um leaning into
24:05the things that make us human things
24:08such as empathy um
24:11thoughtfulness curiosity spirituality
24:14these are the things that AI the
24:16technology artificial intelligence can't
24:18do well and these are the things that
24:20will keep us unique keep us
24:23differentiated keep us authentic
24:24original and indominable by AI
24:29so what kind of AI advancements do you
24:32foresee as a result of of overcoming the
24:35the data wall that that we've
24:37identify so some of this um you know
24:41there's this notion the new generation
24:43of generative AI models that can do
24:47reasoning okay and reasoning is a again
24:50is there's a Continuum of soall
24:52reasoning and people we have to be
24:54careful what we mean by that okay
24:56because there's also this notion of a GI
24:58which is no uh artificial gen you know
25:01general intelligence which is again this
25:03is more of like well where the AI can
25:06self- supervise and do everything on its
25:08own kind of self-learn generalize from
25:10one training domain to another training
25:14domain um the current you know
25:18advancements in some of the generative
25:20AI models definitely make it felt like
25:23as if we're reaching AGI but we're not
25:25there yet okay so somebody advis like
25:28self-supervision learning where AI can
25:30train itself uh there are some
25:32interesting findings where some AI model
25:34when we when we try to retrain it it can
25:37lie to us because it does not want to be
25:39retrained um so there's a lot of ethical
25:43um issues with the process of training
25:45and evaluating AI so those are the
25:47second kind of elements of advancements
25:50there's also Quantum Computing the Q day
25:53this will definitely redefine AI
25:56processing because it it will from data
25:59perspective you know everybody's got to
26:01jump on postquantum cryptography to make
26:03sure that their data is going be still
26:05private and secure now here's a
26:08interesting perspective when people
26:09think okay when is quantum going to be
26:11viable well in about maybe 10 years
26:14although Microsoft has just claimed that
26:15they've managed to accelerate that but
26:17let's just say five to 10 years and so
26:20the question people asked then then when
26:22do I need to worry about postquantum
26:23cryptography I pose this question a lot
26:25of people say well maybe three years
26:27before it did the Q day you can start to
26:30migrate everything I said well actually
26:33um a lot of uh you know cyber criminals
26:36or or people who just wanted to you know
26:38take advantage of this they have been
26:40doing what we call Harvest now decrypt
26:43later when you have a data breach and if
26:46the data is an encrypted form you're not
26:48required to report it to any Authority
26:51okay only if it is in clear text
26:53therefore a lot of theoretically a lot
26:55of data breaches were only encrypted
26:57form of data is out there has probably
27:00been under the radar and the owners of
27:03those encrypted data is just waiting for
27:05the day when they get get a hold of the
27:07first Quantum Computing and then use
27:09that to drip all of our data uh in the
27:12words of my friend who works in cyber
27:14security we are all already
27:16[Laughter]
27:18screwed you talked about you know the
27:22the the challenges you're putting
27:25towards your students talk to me what do
27:28you currently teaching these college
27:29students in light of a data wall yeah so
27:34um I'm currently focusing on learning
27:38myself so to me teaching is actually um
27:40yes I teach but I I use teaching as a
27:43way for me to learn I I force myself to
27:45learn I find that teaching is the best
27:47form of developing myself uh forcing
27:51myself to learn uh beyond what I would
27:54learn without the responsibility of
27:56teaching others and so what I'm facing
27:59right now is especially in the field
28:01that I where I am um doing a lot of
28:03education is in technology unfortunately
28:06a lot of recent graduates um are facing
28:09challenges getting jobs it used to be
28:12that before graduation most of my
28:14graduates would have gotten a job from
28:16Top institutions from Google McKenzie
28:18and others nowadays uh many of them
28:21still couldn't secure a job especially
28:24if they uh require Visa so what I've
28:27been you doing in my class is okay I
28:29teach the theory and I use ai ai is
28:32great because it increases the
28:33effectiveness of teaching and they don't
28:35have to depend on me they can uh we're
28:37I'm experimenting with an AI tutoring
28:39tool which is applying Socratic methods
28:42so the AI will not just give an answer
28:45if a student asks it will give a little
28:46bit of an answer but it will ask the
28:48students a question to challenge their
28:50thinking um so using that to increase
28:53experiential learning method with Gen uh
28:56but also once they produce whatever
28:59papers that they they can produce
29:01because of AI they're able to produce it
29:02much uh quicker much better quality uh
29:06we can get what would have been an endof
29:09year paper by the fourth or fifth week
29:12of the semester at this point however I
29:14challenge them to go further turn that
29:17into content turn that into a Blog turn
29:19that into a podcast turn that into a
29:22multimedia presentation that they can
29:24post on their LinkedIn or whatever their
29:26website so that will help them in their
29:29AI job hunting and writing so those are
29:31things I'm focusing on my students on so
29:34let's talk about the advice you're going
29:35to give to companies right if you had
29:39one piece of advice for these companies
29:41struggling with AI data wall today what
29:45would that be wow
29:50one human uh focus on your people uh I
29:54think as interesting as data wall sounds
29:57it is not a real issue today the real
30:01issue is that most companies have
30:03invested in building gen pipelines and
30:07creating multi-agent strategies and all
30:10this stuff what what we're finding is a
30:12lot of companies are still held back by
30:15the fact that their users don't really
30:18know how to operate in this new model of
30:21you know AI abundance yeah um so the
30:25user AI literacy or fluency program um
30:30is important one of my colleague David
30:32Gaffney had written a really cool piece
30:34on AI
30:36fluency um you know my another colleague
30:38dle uh Greenwood has been producing some
30:42videos all of these are available on
30:45Informatica website by the way um on
30:47data governance to increase literacy and
30:51so we're very active of course in in
30:53know you know promoting this to daa and
30:57EDM Council and other data organizations
31:00but certainly everybody agreed with me
31:02on this um the human factor is the one
31:05piece of advice that we would share with
31:07companies enable your people and in not
31:11just in using AI but also in managing Ai
31:14and the data underlying
31:17it wonderful David this has been
31:20fantastic thank you so much for for
31:25sharing your knowledge with us um being
31:28with us here today um last thing what do
31:30you have what do you have going on
31:31you've got a book you're writing talk to
31:33me about what you have
31:34cooking
31:36yes I uh no I'm I'm fortunate to work
31:39with my um good friend Pete Grimes he's
31:42the author of Legacy thought leadership
31:44I never wanted to write a book um but he
31:48had shown his methodology where within
31:51eight weeks um people can do that and
31:54especially busy executive so he wrote a
31:56book SE level
31:58um author and I found that book last
32:02year and when we met and he has
32:04convinced me that it is feasible it is
32:06possible and especially with Gen and the
32:08technology today it's much easier I'm
32:11hoping to publish a book in a short term
32:13maybe in about a month or two uh on the
32:16subject of AI ethics and it's a personal
32:20code of AI
32:21ethics um the title is AI
32:24squared authentic identity and
32:27artificial int elligence and the subtext
32:29is how to thrive in AI ERA with ethics
32:32and ethics stands for
32:35empathy um
32:39thoughtfulness um you know individuality
32:42curiosity and spirituality these are the
32:44things that make us fully human and the
32:46more we lean into that the more we'll be
32:48able to discover the strengths out of
32:50our authentic self coupled with AI that
32:53will be exponentially more
32:55powerful wonderful I look forward to to
32:58it and uh after you get that book
33:01published let's have you back on David
33:04thank you so much for being with us here
33:05today my pleasure thank you very much JD
33:07for this opportunity

About this transcript

This page contains the full transcript of AI s Data Wall by James Dwayne Group, generated from the public captions YouTube serves with the video. The transcript has 5,320 words across 770 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.