Episode 5: Generating the Future of AI with Peter Prettenhofer of DataRobot

In our fifth episode, Peter Prettenhofer, VP of Engineering for DataRobot, joins Pavan Agarwal for a discussion on generative artificial intelligence and its disruptive potential. What exactly is generative AI? Where does it fit within the broader field of AI research? How will owners of large data sets be able to leverage the recent advances in capabilities?

Go Deeper:

Transcript

The below episode transcript has been edited for clarity.

Pavan Agarwal

Hi everyone and welcome. My name is Pavan Agarwal and I’m a partner in Foley’s Washington D.C. office as well as chair of the firm’s Innovative Technology sector. Our featured guest for this episode of Innovative Technology Insights is Peter Prettenhofer. Peter is Vice President of Engineering at DataRobot and we’ll describe the company in a moment. He’s been there for nearly a decade during their growth from just over a dozen folks to over 1,000 employees.

He led and managed the engineering team that was focused on machine learning for quite some time and over the past couple of years has been working on strategic initiatives in the office of the CTO with a focus on product innovation. Prior to DataRobot, Peter was active in the open source area. He’s a core developer of one of the most widely used open source toolkits for machine learning, with more than 8 million users and 30 million monthly downloads. He’s also authored research papers that have more than 70,000 citations.

Peter, welcome. What do we just start with? Who is DataRobot? Where do they fit in the market?

Peter Prettenhofer

DataRobot was founded in 2012 in Boston, Massachusetts and is a pioneer in automation for AI and machine learning. The key value proposition of DataRobot is to help customers optimize their business using AI and our strategy is to collaborate with customers to make AI a core competency by providing basically two things. One is a ‘data with AI’ platform that is a complete lifecycle platform integrating [the customer’s] existing infrastructure investments, and the second is applied AI expertise; basically professional services derived from the knowledge gained by our data scientists from working with about 1,000 customers and tens of thousands of AI use cases for nearly a decade.

So where do we fit into the market? Unlike other companies, only we have the full AI lifecycle platform, broad technical integrations, and professional services to help customers envision what’s possible with AI and how to achieve it. Moreover, contrary to many of our competitors, we were founded as an AI company and stay true to our roots.

Forty percent of the Fortune 50, eight of the top ten U.S. banks, seven of the top ten pharmaceutical companies, seven of the top ten telecommunications companies, as well as four of the top ten global manufacturers are current DataRobot customers. So the kinds of use cases that DataRobot helps customers with are varied: in the banking space it could be detecting money laundering for example. With pharmaceutical companies and retail companies, we help with planning and demand forecasting. For the telecommunication companies it’s all about churn modeling, and for global manufacturers it’s a lot about predicting.

Pavan Agarwal

If I used the words ChatGPT or OpenAI six months ago, I don’t think anybody would know who they are or what that is. What is ChatGPT? What is generative AI?

Peter Prettenhofer

Generative AI is a new field that describes systems and methods that can be used to create new content including text, images, audio, video, and much more potentially in future. ChatGPT is a concrete generative AI system that is based on text so we call it a language model. It’s a very simple thing – give it some content and it tries to complete it. But it can be instructed by sending feedback to answers in a certain way, reflecting that it was initially designed as a chat bot for question/answering. You should really try it if you haven’t, I think you’d be very surprised by the quality of the question/answering.

Pavan Agarwal

Generative AI is obviously getting a tremendous amount of attention, is that what AI is? Is it just generative or are there other areas of AI?

Peter Prettenhofer

No, generative AI describes models that analyze data and generate new data. It belongs to broad class of AI that’s characterized by what we call weak supervision by a human. It seeks to reconstruct your data and use that to solve a specific task. For example, grouping similar objects to generate new patterns. Other areas of AI are characterized by much stronger supervision, such as discrimination models like classifying whether or not an email is spam or whether a patient will be readmitted in the hospital in the next 30 days. Time series and forecasting is another very important area of the investment. Until now most enterprise use cases have been supervised and thus it’s the main focus for AI companies out there.

Pavan Agarwal

I’ve heard mention of foundation models. Why is that important? Do you see this evolving at a more accelerated rate? Or are we going to continue to move along in AI the way we have been?

Peter Prettenhofer

So I think it’s a drastic change to how we approach AI problems in the future. Foundation models are the generative models trained on very broad amounts of data that can be adapted to a wide range of downstream tasks. If you train a machine AI by using foundation models, you can adapt them for your given task at hand with very little additional data. So this significantly lowers the barrier of entry into AI, allowing new players to build AI systems like e-mail authentication that they couldn’t previously because of the computational burden. This potential for customization and personalization will fuel new innovation in human-computer interaction and around conversational interfaces.

Right now these foundation models are fairly limited to text and images but I do expect that moving forward we’ll probably see expansion into other domains. That’s where we really imagine most of the disruptive power; some of you might have seen a system that’s called Stable Diffusion that really changes how the designers work. They might prototype not by starting from a blank canvas, but by instructing the system to generate the first version and then refining it. Similar innovation is happening in software development where you can instruct the machine to write code that implements your intention. Not perfectly, but often good enough.

Pavan Agarwal

Excellent. I’ve heard that this is the iPhone moment for artificial intelligence. Give our team a little bit about what you think that means.

Peter Prettenhofer

That is basically what we think of ChatGPT even though it has certain flaws. And if you start using it, ChatGPT can come up with non-truthful answers, etc. But I think it really massively demonstrates the art of the possible. And I do think that once you show people what is possible, innovation will rapidly catch on. It’s going to open up so many more applications and use cases that I think we will see that sprawl of innovation.

Pavan Agarwal

Do you see the public perception of AI changing from being mostly weird to interesting? If so, how and what will the impact be on businesses?

Peter Prettenhofer

One thing that I don’t see changing is the view where there’s a mass algorithm somewhere in the back that implements intelligence. I do think that data in particular will become even more important. Given the nature of the foundation models, training is extremely costly to do so. Most people do not have access to the sheer amounts of data you need to train them. And so they will be centralized. Most of the big players such as the Big 5 [technology companies] will position themselves for this future. But I think they will be fairly commoditized and because everybody will have access to these there will be a little differentiation.

The value then comes not from the [foundational] models but from adapting these models to your use case and you can do that by using your proprietary data – customer interactions, etc. which can be summarized.

And so I think that companies that own their own customer and interaction data, they will be able to innovate and create new products and services using generative AI. I do not expect there’ll be fierce competition for building foundation models because it’s extremely expensive and ultimately chip manufacturers probably benefit the most.

Pavan Agarwal

And so what about inside companies? You and I know historically, and this isn’t true for all companies, there’s been a technical team looking at AI. Do you see it migrating up to the executive level? Is it already there? What are you finding in as you look across the marketplace?

Peter Prettenhofer

Companies have been talking about AI and its future potential for the last decade. But many have been slow to act. Most have been experimenting with AI but the economic downturn put pressure on those efforts. The focus is on cutting costs and one thing we see is that many companies are now being asked about their AI strategy and trying to rapidly develop it. It’s a bit too early to say how far up this goes.

Pavan Agarwal

I have one last question for you. Do you see any transitions in the marketplace? Are we talking about different kinds of M&A or new players?

Peter Prettenhofer

So here’s my perspective. On one hand we already see the hyperscalers competing for market share right now. And they need to work together with the chip manufacturers who are providing the picks and shovels for this gold rush moment.

Many startups are also opportunistically dropping into position in this new growth market for generative AI development tools to help others take advantage of it. Existing AI players like DataRobot, for example, will not want to miss out on this opportunity as it provides growth potential. Investors are hungry to see growth after a slow 2022. And from this perspective I do see the AI space becoming a bit more contested and also noisier.

But on the other hand I also anticipate greater M&A activity because of two forces. Some companies might see acquisitions as a short path to building an AI strategy. If you look at what the Google and Microsoft have done it makes others say “what are we doing in this space?” Separately, for non-profitable companies this might be the quicker way for an exit or they might even be forced to sell given the tough funding environment. So I do expect more M&A activity.

Pavan Agarwal

Thank you so much. I can’t tell you how insightful it is to hear from somebody that’s really studying the field and living through it. So we really appreciate the insight you’ve given us here.

Peter Prettenhofer

Thanks for having me.

Foley & Lardner’s Innovative Technology Insights podcast focuses on the wide-ranging innovations shaping today’s business, regulatory, and scientific landscape. With guest speakers who work in a diverse set of fields, from artificial intelligence to genomics, our discussions examine not only on the legal implications of these changes but also on the impact they will have on our daily lives.