AI, ChatGPT-3 & Tableau - here’s how you can use all three in harmony TODAY!

0:00Hey, it's Tim here. Today I want to talk

0:02about AI, ChatGPT and Tableau. It's been in

0:05the

0:05headlines for the last month. I finally

0:07think formed an opinion on this and so here

0:10I am

0:10to share it with you. As ever, let's get

0:12started.

0:13Okay, so by no doubt you will have heard of

0:15a tool called ChatGPT. This is me logged

0:17into

0:17it already. If you haven't, this is

0:19probably going to blow your mind. So pause

0:21this video,

0:21go and check out this tool, ask it some

0:23questions and discover what it's made of.

0:25In this video,

0:25what I want to do is specifically focus in

0:28on how this kind of technology could help

0:30disrupt or even aid Tableau and/or people

0:33using Tableau in the future. And so as a

0:35very

0:35simple task, what you could do is you go to

0:38ChatGPT and ask it, "What is Tableau?"

0:40That's a very simple question. What ChatGPT

0:43has done in the past is it's gone off and

0:45done a bit of machine learning on a corpus

0:47of data. Corpus of data is just basically

0:49like a folder full of information and has

0:52gained that in 2021, I believe, and has

0:54gone

0:55out and sourced information from the

0:56internet. And it's used that information to

0:58build what

0:58is essentially a statistical model of

1:01speech and language that then enables it to

1:03respond

1:04to questions based on the information that

1:06it has. So it's not actually built a

1:08knowledge

1:09graph. It's built a statistical model of

1:10how language comes together alongside

1:12specific

1:13topics and questions, and that's how it's

1:15responding. And so here you get a response

1:17that is, "For all intents and purposes,

1:19good enough. For the everyday person, one

1:21paragraph, straight explainer, perfect."

1:24What you can then start to do is ask it a

1:25little bit more complex things. So for

1:28example, "How do I connect Snowflake in

1:31Tableau?"

1:32And it's even able to sort of work through

1:34typos or things that you get wrong, and it

1:36will give you a step-by-step instruction on

1:39how to use those two technologies to get

1:41them

1:41together. And more often than not, this

1:43comes back 99% of the time correct. And so

1:45this

1:46is a really, really good way of kind of

1:48using this tool as a conversational

1:50assistance.

1:52The other thing I think this tool does

1:53really well is you can ask a follow-on

1:55questions.

1:56For example, "How would I set up that

2:00connection with more security?" So what I

2:04've done

2:04here is I've asked the ChatGPT to

2:06essentially carry on from where it's left

2:09with that

2:10particular question. And I can just say, "

2:12Well, how do I add more security to that

2:14connection?"

2:14And so here there are a few key ways you

2:15can set up more secure connection between

2:17Tableau

2:18and Snowflake. Use a secure connection,

2:21original. Let's wait for the second step

2:23here. Use

2:23a private network. If you're using a

2:24private network to connect to Tableau,

2:26something like

2:27a VPN, then people can't sort of snoop on

2:29your conversation. Use level security. That

2:32's

2:32specifically looking at the actual data

2:34that you're getting access to. Use single

2:36sign-on

2:36to sign into Snowflake and Tableau so

2:38things are a little bit better. And so as a

2:41conversational

2:42I think this is fantastic because you can

2:44get deeper into a topic asking sort of one

2:47sentence at a time to get your better

2:49understanding of a particular topic. And I

2:51think this is

2:52actually the best way to use this. I can

2:54almost imagine this as a conversational AI

2:56in your

2:57ear with an airport or something. You just

2:59talking out loud and something sort of

3:00telling

3:01you how it works. A bit like the film Her.

3:03I don't know if you've watched it. If you

3:04haven't, that's a great film you should go

3:06watch. To me, this kind of technology is

3:07almost in that realm. It's really, really

3:10good. And it starts to fall over when you

3:13get into sort of more complex questions.

3:17For example, how would you configure a Snow

3:21flake

3:21driver for Tableau server? Okay, so there's

3:25definitely going to be some typos in there.

3:28So configure Snowflake driver for Tableau

3:29server, follow these steps, download the

3:31Snowflake

3:31ADBC driver, install the ADBC driver, and

3:33it goes through the steps. And you can

3:35essentially

3:36get to a point where it essentially trips

3:38up on some of these steps and it tries to

3:40give instructions on how to do things now

3:43in this percent. In this particular set of

3:45instructions, it's made an assumption that

3:48I'm using Tableau server for Windows in

3:51order

3:51to do this. And the other interesting thing

3:54is, as it gets more and more complex, the

3:56more complex questions you ask it, there's

3:58a higher risk that one of these points is

4:00not actually factually correct, but

4:02actually still doing a pretty good job of

4:04detailing

4:05out everything I need to know to bring

4:07these two capabilities together. And it

4:09just keeps

4:10on going. And in essence, if you don't know

4:12anything about this topic, what I always

4:15say

4:15is, the more it gives you the more risk

4:17that there's something is wrong, and you're

4:18going

4:19to make a mistake. So you do actually need

4:21to know this topic well, in order to

4:23actually

4:23make use of this, which makes it pointless,

4:25because if you know this topic, well, why

4:26would you come here to understand how it

4:28works? That's where this sort of tool falls

4:30down.

4:31And I think it's a very interesting

4:32capability. But it's got its own sort of

4:34words of

4:35caution. But that doesn't mean I don't

4:37think it's useful. I'll come back to that

4:38point

4:39a little later on. What I want to move on

4:42to next is how is this technology going to

4:45work inside of Tableau? And the answer is

4:47it's already working inside of Tableau. If

4:49I go over to the Tableau's page for AI and

4:51analytics, you can see they have a page

4:54already

4:54that talks about this. So I want I'd love

4:56to know the web analytics behind this

4:58particular

4:59page in the last month. Has it gone up?

5:01Have people search for these two topics in

5:02this

5:03page come up at the top? I wonder. But

5:06Tableau actually themes its AI and

5:08capabilities in

5:09three general areas, augmented analytics,

5:11business science and data science, probably

5:13worth mentioning a distinction between the

5:16term AI and machine learning. And at this

5:18point, it's probably worth sort of just

5:20highlighting why those two things are

5:22different. The key

5:22thing is machine learning is a discipline

5:25inside of AI. AI is the broad umbrella of

5:28everything that happens inside a particular

5:30discipline. So in this case, it could be

5:32machine

5:32learning, it could be predictive modeling,

5:35all of those things sit under the AI banner

5:37.

5:37And so Tableau has three broad areas. And

5:40if you scroll down, it actually spells out

5:42what those three broad areas mean. So the

5:44first one is augmented analytics, which

5:46means

5:47our state and explain data, two very cool

5:49tools. I always think actually our state is

5:52missing a trip, our state should stop

5:53trying to teach you how to use Tableau by

5:55showing

5:55you the pills, the fields and everything

5:57and should just give you what you want the

5:59chart

5:59itself. Now you ask questions, let you

6:01build charts, maybe even suggest charts it

6:03thinks

6:03you should see based on what you're asking.

6:06Explain data is far more useful. This

6:08basically

6:08looks into a data set and tries to spot any

6:11anomalies that are maybe either related to

6:14that particular data point or statistical

6:16variances in the whole spread of the data.

6:18It's actually a pretty good tool at doing

6:19that and helps people just discover those

6:21insights they maybe weren't aware of. If we

6:24go down to business science, this is where

6:25I think this sort of area starts to become

6:28more a little bit hard to understand

6:30because

6:31no, Tableau have all these marketing terms,

6:33simple, fast, trusted, integrated. But when

6:35you go down, it talks about forecasting, it

6:37talks about predictive modeling functions.

6:40But at the same time, business science has

6:42been used in a bunch of capabilities that

6:44relate to Einstein analytics. It used to be

6:46called Einstein analytics is now called

6:48Tableau

6:48business science. And it's gone on and

6:50changed names a few more times. But in

6:52essence, it

6:53talks here about forecasting and predictive

6:55modeling, which are strictly disciplines in

6:58AI, but they're not typically what people

7:01think about when you're using AI. In this

7:03sort of use case, you have to know a lot

7:05about statistical modeling to really go and

7:07use

7:07these or trust these tools entirely. And

7:11the very last area is data science. This is

7:12of

7:12course giving people the ability to connect

7:14their own machine learning models or their

7:16own AI to Tableau. And so here you've got

7:19our Python, MATLAB and analytics extensions

7:22API. These are all ways of bringing that

7:24intelligence into the platform and then

7:26using it to work

7:27better. So it's actually really nice to

7:29have to have this page, I'm going to link

7:30it in

7:30the description so you can go check it out.

7:33But I hinted earlier that in my opinion,

7:36this

7:36is just one way that Tableau could be using

7:38this information, they could be using this

7:40technology to help people and I think it

7:42makes sense from a product perspective,

7:44Tableau

7:45have this challenge where, you know, the

7:47early adopters of Tableau were the people

7:48who built

7:49dashboards or the people who go out and

7:51build tools and things that people use. For

7:54them

7:54to get over that hump, to get the mass

7:56adoption, to really go viral, something as

7:59viral as

7:59Excel to where everyone knows what it is

8:02and knows how to use it, they really have

8:04to go

8:04to a much, much simpler user base. And in

8:06order to do that, you can't expect them to

8:08be opening up Tableau and doing data models

8:10. That's almost a step too far. And

8:13businesses

8:13just haven't opened up their data stages

8:16and their data warehouses to people in that

8:18way.

8:19And so, Tableau have to use AI machine

8:21learning to surface insights on behalf of

8:23people to

8:24get them using the tool better and that's

8:26sort of one approach they've taken. But I

8:28still fundamentally believe that actually

8:30you can use AI machine learning

8:32capabilities

8:32to help people discover the tool better and

8:35this is sort of a nuanced thing. So bear

8:37with

8:37me. You see, when we were sitting here

8:40asking it questions, this data was trained

8:42on a corpus

8:43that was just scanned from the internet a

8:45long time ago. And the way that ChatGPT,

8:48specifically

8:48the company behind ChatGPT, OpenAI have

8:51done this is they took a technology that

8:53was actually

8:53released last year called GPT-3 and they

8:56combined it with human input. So

8:58essentially they had

9:00people who were editing the input and the

9:02style and the responses as they came across

9:04it. And they did this on mass scale. There

9:06's actually an academic paper which I'll

9:08link

9:08to that you can go check out that talks

9:10about this. And that's what's made it a

9:11really,

9:12really good tool. Now, essentially they've

9:14trained it on what is a broad corpus of

9:16data

9:16about the world and it's coming back with

9:19very good responses. The flip side is you

9:21could train this on a very specific set of

9:23data and you could end up getting really,

9:26really valuable outputs out of the system,

9:29even though it maybe doesn't necessarily

9:31know about things that might be new or

9:33might be coming up in the future. And so

9:35that's

9:35where I've got the Tableau help and the

9:37Tableau community, because if I just

9:40restrict the

9:40scope to just what Tableau has access to,

9:42they've got all the help documentation

9:44going

9:44back. I don't know how many versions. And

9:47the great thing about those is they're

9:49version

9:49specific so you could actually very easily

9:51have Tableau explain something in different

9:54versions if you go out and you train your

9:56own model on this specific corpus of

9:59information.

10:00Essentially all the help documentation that

10:02you put together to help people better use

10:03the tool. This is a highly underused

10:05resource. People never really find it and

10:08what you tend

10:08to find is when you go to the Tableau

10:10knowledge base, they're linking back to the

10:12help documentation.

10:13When you look in the Tableau forums, they

10:14're also linking back to this. And that's

10:16sort

10:16of bringing me to the second dataset, which

10:18is this. The Tableau community pages also

10:21have tons of questions and tons of

10:23different things that people have asked

10:25alongside the

10:26answers and the things they're linking to.

10:27So you could also use that information as

10:29a way of building a model for the way

10:31people are finding problems and the answers

10:34that

10:34are solving those problems. And pairing

10:36those two things, you could maybe build

10:37something

10:38a little bit more intelligent than just ask

10:40data and explain data, which is instead of

10:43helping people see better insights, instead

10:45help people use the tool better so they can

10:48get to those insights faster. It's a

10:50completely different paradigm, but it's

10:52more of an educative

10:54sort of approach. And it's something I

10:56think we've lost touch of in more recent

10:58times.

10:58Back when I was young, Microsoft Word had a

11:00tool called Clippy and Clippy would help

11:02you understand how to use Microsoft Word.

11:04You go to Clippy, you ask it how to do

11:06something

11:06and it would tell you, here's how to do it.

11:08And it would point you to the right section

11:10of the software. We've lost that sort of

11:11finesse in more recent times with

11:13technology because

11:14I think smartphones and tablets have sort

11:17of forced people to make UI and technology

11:19much, much simpler. And that's good because

11:21it makes it more accessible. But then it

11:23also

11:23means the capabilities, really powerful

11:25capabilities are pushed off to the fringes.

11:27And I think

11:28that happens with Tableau a lot. I can't

11:29tell you how many people just aren't aware

11:31of certain

11:32features that are deployed and live right

11:35now inside of Tableau. Just the other week,

11:37I actually met someone who asked, why is it

11:39so hard to do year on year comparisons with

11:41Tableau? And in about three clicks, I went

11:44ahead and I showed them how to do it. Now,

11:46they didn't know that the approach I took

11:48was actually possible because they didn't

11:49know you could create an ad hoc calculation

11:51very easily to just change the number from

11:53being a minus one to being a minus 12.

11:57Because we had months in the view and year,

11:59we had

11:59months in the view and then profit on the

12:01axis. And basically, we're just looking at

12:03year on year sales broken down by month.

12:05And I just changed the lookup to basically

12:07look

12:07back 12 months and create a duplicate of

12:09the view. And there we had it, we had a

12:11year on

12:12year comparison, and you could then do

12:13difference and a bunch of other things from

12:15there on

12:16using percentages as well. And when I

12:19showed it to them, they found that approach

12:21easy.

12:21But what they were comparing it to is Power

12:23BI, which had something baked in already.

12:25And they were actually comfortable doing it

12:27the more complicated way, if they knew it

12:29was as easy as I made it look. And so this

12:31is where I think Tableau really could use

12:33this tool, use it to help surface better

12:35ways of using the product. Hey, you've just

12:38created

12:38a join. But we think the exact same thing

12:41as a data model would actually perform

12:43better.

12:44And here's why. Then an introduction to the

12:46model and just have this AI paying

12:48attention

12:49to what's going on in the software, how you

12:51're using the software, how you're browsing

12:52the

12:52software, they already have this capability

12:54in Tableau server, which looks at what

12:56other

12:56people are using that are similar to you.

12:59And they're surfacing that to you as

13:00recommendations.

13:01It's exactly the same thing. But this time,

13:03you're actually helping people better use

13:05the software. And so I think that's an area

13:07that could work really, really well. The

13:08copyright

13:09issues are endless. If you're not Tableau,

13:11and you're thinking of doing this, I'd

13:13recommend

13:13you to just do this with your own data set,

13:15which in my case is going to be the

13:17translations

13:18of the video into text essentially. So

13:21going and getting the audio transcribing

13:23the whole

13:24plot, but it's not going to be very useful

13:26data set because it's all conversation,

13:27there's

13:28no sort of instructional element to it. The

13:30logs on the other hand are a very good day.

13:32So if you've been writing blogs for the

13:33last decades, now's the time to see if you

13:35can

13:35train an artificial intelligence store to

13:38go and maybe surface this in a way that

13:41works

13:42that surfaces your blogs to people in a

13:43much better way. The other thing is if you

13:45've created

13:46a lot of blogs over the last few years, it

13:47's probably worth putting a copyright notice

13:49on the bottom of your blog just so that

13:52when an AI or tool comes to take a corpus

13:54of your

13:55blog without your permission, at least you

13:56have some sort of protection for when these

13:58tools then do fall under sort of legal

14:00mandate of some sort because at the moment,

14:02it's just

14:03a free for all. And the tools just seem to

14:05be going out and scraping whatever they

14:07want,

14:07because it's the internet, it's an open

14:09internet. So yeah, those are my thoughts on

14:11this issue.

14:12It's a very nuanced issue, I'd be really

14:14intrigued to see where else it goes. But

14:16for me personally,

14:17I think the most useful area is just this

14:19stuff. There's so many simple things in

14:21Tableau

14:22that this tool could already do today, you

14:24could already ask it how to do them, and it

14:26would already go off and explain them. I

14:28can't tell you how many how many simple

14:29things for

14:30example, how do I build a bar chart in

14:33Tableau? I've made a two minute video on

14:36this. So hopefully

14:37you'd find my video and go and watch it.

14:39But here to build a bar chart in Tableau,

14:41follow

14:41these steps, open Tableau and connect your

14:43data, great, drag and measure fields to

14:46column

14:46shelf and the dimension filter the row

14:48shelf. It's even specifically telling you

14:50which shelf

14:50that is. For example, if you want to create

14:52the bar chart showing the cells by region,

14:54you would drag the cells measures to the

14:56column soft and the region dimensions of

14:57the row

14:58shelf. In the marks card, select bar as a

15:00mark type telling you to go to the marks

15:03card.

15:03If you want to customize the appearance of

15:05the bar chart, you can then use the options

15:06in the marks card to change the color, size

15:09and other visual properties for the bars.

15:12If you want to add labels, go ahead, you

15:13can also use the options in the columns and

15:15row

15:16shelf to customize the axis for your bar

15:17chart. For example, you can change the

15:19number format,

15:21axis title and other options. Once you've

15:22finished building your bar chart, you can

15:24then use the options in the worksheet tab

15:26to customize the appearance of the chart

15:28and

15:28titles and captions and save before you

15:31publish your work. If you knew nothing

15:33about Tableau,

15:34tell me that isn't a much better

15:36description than just going off to Google

15:38and trying to

15:38find the right help and then ending up

15:40stuck in a bunch of resources. If you don't

15:43understand

15:43any of the terms, you can always go ahead

15:45and follow that up. So you can say, hey,

15:47what

15:48is the marks shelf? And it understands you

15:53're talking about Tableau and it will just

15:55go

15:55ahead and tell you in Tableau the marks

15:57shelf is the visual element that you can

15:59use to

15:59customize the appearance of behavior. The

16:02marks shelf is located in the marks card,

16:04which is found in the lower right corner of

16:06the Tableau interface, the lower right

16:09corner

16:09of the Tableau interface. Oh, interesting.

16:12That's technically wrong, but I know what

16:14it's referring to in the Tableau interface.

16:17There is nothing on the lower right corner.

16:20It all stops like on the left hand side,

16:22but it means the most lower right thing in

16:25the

16:25Tableau interface. So in that respect,

16:27maybe the language isn't great. It needs to

16:30be able

16:31to just show you images and you can go off

16:32to Google and set it up, but it gives you

16:34an explanation. You can also ask it where,

16:38let's see, let's ask where is the columns

16:41shelf? And again, it understands it. It

16:46worked through my typo. In Tableau, the

16:49column shelf

16:50is a visual element that you can use to

16:52specify what to include as columns in your

16:54visualization.

16:56The column shelf is located in the top left

16:58corner of the top Tableau interface next to

17:00the row shelf. That is actually correct.

17:02And then again, it gives you a little bit

17:04more

17:04detail and you can go off and use it. So if

17:06you're new to Tableau and you're stuck, you

17:10don't understand how to do something, I

17:12think more often than not, this tool is

17:14going to

17:14give you the right answers. And if you

17:16combine that with your research skills and

17:18your ability

17:18to deduce information from the help

17:20documentation alongside the internet, just

17:23doing basic research,

17:24Googling, I can't see how you would go

17:26wrong in understanding how to do something.

17:29Yes,

17:29you might get stuck. You might not know

17:31exactly what you're trying to do. You might

17:32not know

17:32exactly what data set you're working with,

17:34but at least when it comes to the realm of

17:36just understanding how to do something or

17:38knowing where there's knowledge gaps that

17:41you need to go off and investigate, I think

17:42this is a fantastic tool. And I can't see

17:45why people won't be using this more to help

17:47them get their work done faster so they can

17:49achieve what they're trying to achieve and

17:51do less of messing around with tools and

17:53software

17:54that frankly don't matter because when the

17:55insight is what you're trying to get to,

17:57that's

17:57the only thing that matters. And therefore

17:59that's pretty much the only thing you

18:00remember

18:01in the entire workflow. Three, four years

18:03down the line when you've forgotten all

18:04this

18:04stuff, the insight is still what you'll

18:06have in your mind. So maybe Tableau should

18:09just

18:09spend more time making it so easy to

18:11discover rather than trying to add more and

18:13more and

18:14more insights on top of it. It's just an

18:16opinion. What do you think? Let me know in

18:18the comments

18:18below. I'll catch you in the next one.

18:20[silence]

18:26[ Silence ]