Why Swami Sivasubramanian’s Data & Machine Learning Keynote so heavy on data, and so short on ML and AI at re:Invent (Live from Las Vegas with Inawisdom)

Transcript

Hilary:

Rahul, it’s day three. You’ve slept for three and a half hours since you arrived four days ago. We have plans to dunk you in an ice bath just to keep you going. And luckily, there’s one just across from the koi pond up here on the roof of the Nobu Hotel. When they ask you to leave here on Friday, don’t let them. This is our podcast’s home now.

Rahul:

Well, Friday’s a long way away, and of course there’s so much more to come.

Hilary:

Don’t I know it? All right.

Swami Sivasubramanian, VP Database Analytics and ML at AWS is about to take the stage.

We have Alex Kearns with us here to parse the announcements. Alex is an AWS consultant and Senior Solution Engineer at Inawisdom in the UK. If you are talking about AI, ML or data analytics, you want to have Alex in the room and we do. So let’s get to it.

Audio:

Please welcome the Vice President of Data and Machine Learning at AWS, Dr. Swami Sivasubramanian.

Welcome!

Hilary:

Live from re:Invent in Las Vegas. This is AWS Insiders, an original podcast by CloudFix about the services, patterns, and future of cloud computing at AWS. CloudFix is a tool that finds and implements 100% safe. AWS-recommended cost savings. That’s fixes not just analytics.

I’m Hillary Doyle, joined always by Rahul Subramaniam. Rahul, Swami just left the keynote stage where he opened with a Selipsky-style setup that looked at inventors from Isaac Newton to the inventor of the microwave oven. His opening thesis, these moments of innovation, these light bulb ahas, they can only happen when you’ve got the right context, the right information working in the background so that you can make the connections that lead to something revolutionary. In other words, move over Isaac Newton, because AWS can now operate as your enterprise brain making those connections for you, assuming you’ve laid the groundwork to capture that data properly.

So according to Swami, the key is building a future-proof data foundation within AWS and then creating neural networks by weaving a connective tissue across the organization. It’s so easy.

We’ve been watching the action here at your watch party, and it’s been a pretty quiet room here on the rooftop.

Rahul:

Yeah, I think the problem has been with some of the storytelling.

Hilary:

Yeah, yeah.

Rahul:

The connections are crossfired all over my head at this point. But-

Hilary:

Do your best.

Rahul:

Here’s how I unravel this entanglement. I think as everyone knows who’s been dealing with anything ML or AI, it’s all about the data.

Hilary:

Of course.

Rahul:

So if I were to break down Swami’s keynote, it boils down to data connections in terms of where do you source all these desperate pieces of data. Second, how do you process all of this data and get it into shape so that you can actually do something with it? And then the third part of it is what interesting insights can you gain from it, both from a BI and analytics standpoint at the low end of it, and then machine learning models.

So I think if I were to where my technologist hat, the one thing that I think Swami addressed way better than Adam did was he did touch on at least all of those aspects throughout his keynote, though I think the storytelling could do a little bit more work.

Hilary:

Okay. All right. Anything in here that surprised you?

Rahul:

No, I think I was hoping to see some more innovation in the no-code, low-code space or trying to make it easier for customers to A, acquire new data sets. Because the problem that everyone has is data. In fact, when he was just talking about the fact that there are data sets that customers struggle with, I was hoping that they would actually create a marketplace of different data sets and start curating it themselves for the customer base. That would’ve been a really cool idea. I think they did a little bit of that with the GIS data. But again, that’s such a niche that I’m not sure they addressed a broader segment of their customer base.

Hilary:

Okay. That’s interesting. We have been hearing a lot about recent Amazon layoffs. They’ve cut into the Alexa team really heavily. You don’t hear much from Amazon about the generative AI space. And I’m just wondering, the ML, AI innovation at AWS is really deep, but what’s the lane they’re looking to own here?

Rahul:

I think like the rest of AWS, they are not that opinionated about the lane. They’ve just been so focused on creating a bunch of building blocks and hoping that their customer base tells them what lane to be on. So they’re counting and the customer base to basically do all the innovation and the experimentation and come up with use cases that’ll hopefully inform them about which way to go. So I think they’re still discovering it for themselves.

Hilary:

We had a great, fast conversation after the keynote wrapped up.

Audio:

My second slide was… Don’t-

There’s no cool stuff-

Hilary:

You’re being hounded by journalists, so sometimes you have to step out. But it is interesting. It feels like this has been a really subdued year for the keynotes.

And as I’m saying that, I realize in the last two days, we’ve heard about AWS helping companies cure cancer, save lives in disaster zones, improve access to computer science education.

At this point, are we just spoiled? Is the infrastructure so impressive that the work is done and now it’s just about improvements to a really solid base? Or is this a problem of Amazon struggling under its own weight a little?

Rahul:

No. I think over the last five years, Amazon has put in an enormous amount of effort trying to basically have technologists believe in the idea of innovating in the cloud. And I think they nailed that over the last five years or so. This year, given the economic…

Hilary:

It sounds like we’re in the middle of a heist. Welcome to Las Vegas, guys. You’ll hear the choppers in moments. It’s all right if they come and pick one of us up. The other will finish out this episode. Continue.

Rahul:

So like I was saying, the last five years were being spent by AWS nailing the message to the technologists. And I think they convinced everyone that if you want innovation now, it’s got to be in the cloud, it’s got to be on AWS.

This year with all the economic uncertainty, I think they are really trying… I’m not entirely sure they’re succeeding, but they’re really trying to convince the CEOs and the CIOs, more the management of the organization, that they are in the right company, that this is the right bet that they have made.

And I think the message feels a little mixed. Their storytelling hasn’t been to point. It’s, in fact, gotten a little confusing. But at heart, they are a bunch of technologists who have built some amazing services and some amazing building blocks, and I think most organizations realize it. So I think the keynote is just a little bit of a blip.

At the end of the day, the services that they’ve launched, the APIs that are now available for customers to go try out, those are going to be the real winners and we are going to see how everyone’s starts leveraging them.

Hilary:

Fair to say that this year’s re:Invent really isn’t for the technologists, it isn’t for the developers, it’s for the C-suite?

Rahul:

I think the keynotes. I would say that keynotes definitely seem to be geared towards the C-suite. But if you go down to the expo floor, I think there’s no doubt it’s all about the technologists.

Hilary:

We’ll get some audio for you guys from the expo floor. Sounds like a really happening place. I’m just here on the rooftop watching things from above.

Okay. We have our special guest waiting in the wings slash sitting right next to us. Let’s bring Alex Kearns into the fold. We’ve got a lot to talk about.

Alex Kearns is an AWS consultant and a Senior Solution Engineer at Inawisdom. Inawisdom is a UK-based AI machine learning and data analytics consulting company. Alex, we are delighted to have you with us.

Alex:

Thank you so much for having me.

Hilary:

Oh, God. Great to have your accent joining this show. Thank you. I know it’s fake, but keep it up.

Alex:

Absolutely, [inaudible 00:08:11].

Hilary:

We that coming into re:Invent you were looking forward to this keynote. What is your immediate takeaway from this morning?

Alex:

I think very similar to yesterday’s keynote, really, in that it-

Hilary:

Uh oh.

Rahul:

So I wasn’t alone.

Alex:

The feel of being at the keynote was great. But when you get back to the hotel and you look at the announcements that have come out, and there were I think 13 or 14 that came out of yesterday, but there’s not a lot of push on the new shiny stuff, the things that make you really get excited.

Hilary:

What would’ve gotten you excited today?

Alex:

So what I really came into re:Invent looking for was, particularly in the ML space improvements to SageMaker, but for builders. Making SageMaker more flexible and less focused on the SageMaker studio, the all-in-one ID, the no-code, low-code, which is the direction that everything really seems to be pushed into. So the announcements that came out for SageMaker today, we had two machine learning announcements in a keynote that is to be focused on machine learning and-

Hilary:

A bit of a miss. A missed opportunity, shall we say.

Alex:

Yeah, you walk away from it and think, “Is the focus on data rather than machine learning? Is that where Amazon see the immediate future?”

Hilary:

Do you think that data is the focus for Amazon right now? I mean, you’d be forgiven for thinking that.

Alex:

I think data is where the money is. I think machine learning is probably a good follow onto the data. Perhaps Amazon have a strategy to get all data into AWS first, and making it easy to do that is a way to enable machine learning later.

Rahul:

I think it was less about the data itself, but more around the governance of data.

Hilary:

That was interesting.

Rahul:

Because data is so critical to most of these organizations. I mean, AWS is pushing incredibly hard for these organizations to take all the data that they have internally across all of their systems and bring them to the cloud so that you can do something interesting with it.

And I think the biggest fears that these organizations have around the data is governance. What happens if this data leaks? All the compliance and GDPR and all that other stuff that’s going on that they have to be compliant with? That’s the biggest concern of the CIOs, the legal departments at all of these large firms that have any data worth working with.

Because all the other stuff, the machine learning pipelines and the workflows around SageMaker, the Canvas, all of that stuff, that the technologists are already sold on. I mean, you’d speak to any technologists, they’re like, “Yep, that’s the one I want to go with because I’ve got all of my tools. I’ve got all of the stuff. I’ve got all the incident types that I need. It’s all dynamic. It’s on demand. I can do whatever I want.”

But the problem is data. “How do I get the data to work with?”

And I think that’s what they’re trying to address. It’s the same story that I had with EC2 and the basic infrastructure maybe 12, 13 years ago. I’m seeing that play out with data. And data is way more sensitive to these organizations than the barebone machines were 12 years ago.

Hilary:

What are some of the announcements from today that are actually going to practically affect both of you, day to day?

Alex:

I mean, I think for me from today and from yesterday as well, DataZone is one that I see as being really important, particularly at that enterprise level.

So DataZone is a new way to do data cataloging and manage the governance of that within AWS. I really like the idea of being able to have these data produced and data consumers and be able to, as a consumer, subscribe to a data set within an organization.

But it’s not a new idea. To me, it felt, particularly today like it’s a case of Amazon filling in the gaps. You’ve got products like Data Hub, which is a data catalog outside of AWS. The announcements around things like the 20 new app flow connectors. Previously, if it wasn’t an app flow, you’d go and use Fivetran or you’d use Airbytes or something else.

So it feels like AWS is just really trying to put together a proposition where people aren’t forced to go and use something different.

Rahul:

And I think from my standpoint, it just boils down to the governance model. As long as there is one system, one way of doing things, I think most enterprise organizations feel comfortable with that. If you had to go to 10 or 15 different systems to try and coordinate all of that, it just feels like too much of an overhead for a lot of these enterprises. And I think that’s the gap that AWS has realized as they talk to their customers. And that’s what they’ve been trying to fill out.

And I think that could have been a lot more succinct in Swami’s keynote. They could have made sure that they doubled down on this message that this is the most secure place to put your data and the most secure way to deal with your data. Don’t just keep copying stuff around from here to there and have shadow IT within your organization, create copies of your data, which you have no track of. Instead, here’s the way to do it. Double down on the governance, sell the story, get them convinced, and then… Yeah, it’s-

Hilary:

Get out of there.

Rahul:

Get out of there, yeah. I mean, it’s home run from that point on, right?

Hilary:

Yeah.

Rahul:

So I think that’s the opportunity that AWS missed today. But if I put on my technologist hat, I actually feel pretty happy about closing all of these gaps because that has been one of the big pain points: moving data around making sure the right people have the right stuff, making sure their governance is in place. And I think they’re starting to close the gap on that.

Hilary:

Okay. I disagree with the direction that AWS is taking in the spirit of their low-code, no-code focus.

Alex:

I think the boundary should be wider. I think it should be more distinct to say we’ve got tools like SageMaker Canvas, we’ve got tools like SageMaker Autopilot, which are very much designed for, “I’ve got some data. I want to predict a value in this particular column, go and do some magic.”

The problem with that is if you don’t know what magic it’s doing, even at a high level, there’s so many problems that can come from that. If you’re building a model on a data set that has protected characteristics in it like race or gender, there’s so much potential for bad bias to be introduced to machine learning. Which, the moment you start using that into any business process, you are opening yourself up for massive liabilities.

Rahul:

Yeah. I have a slightly different take on this, and if you had spoken to me about 10 years ago, I would probably have the same views. But what I realized is that when you have a platform at the scale of something like AWS, with that many APIs and that many services, you’ve basically given developers a whole lot of ammunition to shoot themselves in the foot. And it becomes imperative at that point to have opinions, have certain ways of doing things that you define upfront. Because too much choice is also a problem. And yes, anecdotally there are some things to watch out for and stuff like that, but when you take a step back and look at things at scale, then you identify patterns of things that people are doing wrong. And AWS has an incredible amount of data that talks about what people are doing wrong. And the fact that they’ve actually taken that, especially in this area, and condensed them down to certain best practices, things like Autopilot…

When I was talking to the SageMaker team about how developers were completely messing up their hyperparameter tuning and struggling with it. The fact that Autopilot was created to just do all the stuff that they were doing, do it efficiently, use ML to figure out what kind of hyperparameter tuning you needed, I think those are all great steps because they really help get the users to the outcomes they want. No one realizes that the process of getting there isn’t as important as getting to the final outcome that you really want. And the fact that they can reduce down the choices, make it easier to get to that end goal, is, I think, the whole objective of this exercise. So I think from AWS standpoint, that’s what is key and I like what they’re doing.

Alex:

I think the making it easier is absolutely the right direction. I wonder if either messaging around it or the documentation or something around it needs to be almost a big warning sign saying, “Do you realize what you are doing?” And that’s-

Rahul:

Let’s put a bumper sticker on that one.

Alex:

Yeah. So with the AWS machine learning suite of tools, I think in the keynote yesterday they showed this three swim lane type diagram type where you’ve got the real custom stuff at the top. Then you’ve got SageMaker Studio and SageMaker Canvas in the middle. Then you’ve got the niche, the managed: Rekognition, Textract, Polly, those single services at the bottom. So they work really well. They are super focused models. You can do a little bit of custom stuff with them, but not huge amounts. But actually, that’s okay. They’re designed to be an API where I can say, “Here’s a photo. Is it a celebrity?” And they’re really good at that.

I think where the problem starts to creep in is the more blurred those lines get, if people start to think, “Okay, if this is possible and I can just throw some data at this and it works, why can’t I throw some data at this and it works.” And it might work, but it’s the why and not necessarily the how, but knowing the pitfalls of what you could be doing here.

Rahul:

My takeaway from that is they need to be more opinionated.

Hilary:

Thanks for answering the whys for us today. Alex, we’ve been so happy to have you here and wishing you the very best for the rest of the conference.

Alex:

Thank you, you too.

Hilary:

Alex Kearns, everybody.

In each of our special episodes from re:Invent, we are taking questions from our audience. For Rahul today, we have a question from Marion about Amazon Go. She says, “Amazon Go was a revolutionary idea, but we haven’t heard anything about eCommerce innovations in these keynotes. Should we be worried about touchless eCommerce going the way of the Alexa Dodo bird?” Why aren’t we hearing-

Rahul:

Interesting question.

Hilary:

Yeah, why aren’t we hearing more about eCommerce innovations?

Oh, the drag racers are out.

Rahul:

Absolutely. Okay. So-

Hilary:

Must be 10:30 AM. Good.

Rahul:

So the one thing that AWS is very particular about is not telling the customers and how they should do. I mean, I’ve said this so many times.? That’s the culture, that’s the way they operate. They love to see what their customers do with things. And I think what is disappointing in the keynotes is that they didn’t highlight enough of what their customers were doing with all of this amazing innovation. So I would not expect them to showcase Go as a big theme and do something like that in any of the keynotes. But yes, Go is pretty damn revolutionary and I think we are going to see a lot more stuff coming out in that space.

Hilary:

All right. Before we go, what’s the look ahead for tomorrow?

Rahul:

Okay, so tomorrow we are going to have the talk that we’ve all been waiting for.

Hilary:

Oh my God.

Rahul:

The one that’s addressed towards the technologist.

Hilary:

I know!

Rahul:

So that’s going to be one of Vogels. I’m super excited. I also keep a particular eye out for the music that he’s chosen for this year. But yeah, so that is going to be a fun, fun talk to attend and I’m really looking forward to it.

Hilary:

I know. I’ve been planning my outfit for days. And while I’m planning my outfit, our wise listeners should send us their questions, their queries, and their reactions to podcast@CloudFix.com.

Rahul:

And please don’t forget to rate us and review us.

Hilary:

Please, five stars. That’s it. We’ll see you tomorrow.

Rahul:

Absolutely. Bye-bye.

Hilary:

À demain.