Living with AI: An Intelligent Today and Tomorrow

How far are we from AI becoming mainstream? What fields will benefit the most from AI and ML? What’s the most important component when building an AI product? Underscore sat down with six AI experts to answer these questions and more!


Emerging technology has – to some extent – impacted nearly every industry on the planet, with artificial intelligence and machine learning being the two most prevalent technologies that business leaders should be prepared for.

With AI and ML making their way into an increasing number of aspects in our world, it’s important to look to the experts to see just how these two increasingly intelligent technologies will impact the way we communicate, do business, and live our lives.

At Underscore’s 2019 Core Summit, we had the opportunity to speak to six such experts in our panel: Living with AI – Imagining an Intelligent Today and Tomorrow. The speakers on the panel were:

  • Heather Ames, Co-Founder and COO of Neurala
  • Nick Brachet, Director of Engineering at Forge.AI
  • Doug Merritt, CEO of Splunk
  • Andy Palmer, Co-Founder and CEO of Tamr
  • Rama Ramakrishnan, Professor of the Practice at MIT Sloan School of Management
  • Nir Shavit, Co-Founder & CEO of Neural Magic

Living with AI Panel

They offered up their opinions on how far we are from machine learning and artificial intelligence becoming officially mainstream, what industries would benefit most from ML and AI, and why companies need unique data generated from ML now more than ever.

First Things First: Differentiating Between Artificial Intelligence and Machine Learning

While often used interchangeably, there’s a clear distinction between artificial intelligence and machine learning.

Artificial Intelligence (or AI) is defined as “machines that respond to stimulation consistent with traditional responses from humans, given the human capacity for contemplation, judgment, and intention.”

Machine Learning (or ML), on the other hand, is defined as “an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed.”

In other words, AI is the umbrella term that ML is part of – AI is the ability of a machine to know, while ML is the ability of a machine to learn. As Doug Merritt puts it, “AI gets misinterpreted as effective machine learning routines, and not true AI.”

When Will Machine Learning and Artificial Intelligence Officially Become Mainstream?

When posed with presenting a timeline for machine learning’s shift to the mainstream, the panel was sharply divided, but agreed the date is indeed inevitable.

“I think mainstream AI is 100 years away, [maybe] 125. ML being baked into almost everything we touch? I think that is happening now, and we’re five to ten years away from where I think a product would be difficult to survive without a meaningful ML backbone,” said Doug Merritt.

Nick Brachet dug into the question deeper, saying, “Well, I think you have to define mainstream. Mainstream is really everybody’s using it every day, and it’s part of our lives – whether we know it or not know it. I think that’s probably…10 years? But otherwise, we’re not that far from having AI being part of many processes that we use on a daily basis.”

After asking how many panel attendees used Google, Rama Ramakrishnan declared, “ Frankly, I think it’s mainstream.”

Calming the nerves of anyone expecting a new ruling class of supercomputers, Andy Palmer agreed with Rama. “This is math, compute, and data – and it’s the same freaking math that I’ve been studying since I started working with Marvin Minsky back in the early 80s.”

“It’s math,” he continued, “and we’re going to continue to use it faster and faster. And maybe it’ll get so fast that it starts to approach something that mimics general intelligence, but I just think we’re a century away from that.”

What Fields Would Benefit Most from AI and ML?

When asked which industries desperately need more artificial intelligence, the panel was largely in agreement in two fields: health care and, well, every industry.

On the topic of health care, Heather explained the group’s consensus on the topic largely comes from the health care industry’s lack of exposure to ML and AI to date and the immeasurable impact the movement would have on it, saying “This decision not to approach healthcare [was] because of the long lead times in sales cycles and just a lot of bureaucracy to get such an early technology off the ground. But we recognized very early on that the data that’s amassed in healthcare is just so perfect for AI that it’s a shame that it can’t be used more broadly. And also it has real implications on human life, which is kind of a beautiful marriage with the technology.”

A few other areas members of the panel thought would benefit?

  • Nir said storefronts.
  • Rama said climate change.
  • Doug said supply chain optimization.

On the topic of supply chain, “there’s so much inefficiency and waste,” Doug noted. “And it’s a practical, boring application that will revolutionize different aspects in the industry.”

What’s the Most Important Component When Building an AI Product?

The panel was asked to choose between four different options as the most important aspect of the building of a product with AI:

  • Great product management.
  • Great data modeling.
  • Great data itself to train the models.
  • Something else entirely.

Heather kicked off the discussion by advocating for the last option: something else entirely. “It’s not just one thing.” Heather said. “You can’t do anything with AI until you define what you’re using it for. There’s too many people that just want AI for AI sake without solving an actual problem.”

“You can’t build a model with terrible data – garbage in garbage out,” she said. “The data needs to answer your problem that you’ve defined with product management.” Finally, she suggested that, “we need to take [data] out of the AI lab and into the realm of productization, and really treat it as such rather than as a technology but as something we can productize.”

Nick also advocated for the “something else entirely” option, saying what’s most important is “a clear definition of the objective and the goal, which I guess falls under the product management, but it’s more than that.”

“AI is difficult to determine,” Nick continued. “What is it you’re trying to achieve? Have you achieved it? Are you there yet or not? That definition is tricky, and so maybe the journey to get there is product management, but this is to me the most difficult.”

Living With AI Panel

Rama advocated for the importance of great data. “If you have disproportionate access to some unusual data that nobody else has access to, you can couple it with mediocre product management, a fuzzy problem description, and whatever else is on your list, and you could still make it reasonably successful – while if you have all the other things, but your data is a commodity and everyone else has it, it’s going to be a really ugly fight,” he said, adding that when companies focus mostly on product management, they “completely miss it on the data, and then they have thirty competitors with the same data, and [they] can’t differentiate [themselves].”

“Your focus should be on getting your hands on data that nobody else has access to, and try to preserve the competitive window for as long as you can, because that gives you the breathing room to get all your other stuff done and in order so you can actually come to market with a compelling offering,” Rama concluded.

Nir and Andy advocated for great data modeling as the most important component to a new AI product.

“In machine learning right now, the models that we have and the way we analyze data [are] a reflection of the people that are doing the analysis, Nir said, and so I think it actually is about the model right now. It may change, but right now, that’s where it is.”

Andy agreed that great data modeling is most important, but did not discount the importance of data. “I think it is about the model, but with one caveat, which is: many of these models have a core assumption at the beginning,” being “What data do you have?”

“There’s so much dark data out there,” he said. “There’s so much data that is available that’s laying around on the ground, and dirty in all kinds of ways, that your models can change significantly when you change what data you have available in order to drive those models.”

The key takeaway? An AI product needs a combination of great product management, great data modeling, and great data, but at different stages, each of these criteria can take priority – so long as the data is unique enough to stand out on its own.

An AI product needs a combination of great product management, great data modeling, and great data, but at different stages, each of these criteria can take priority – so long as the data is unique enough to stand out on its own. Click To Tweet

What’s Next for AI and ML in Business?

While the panel acknowledged that data and ML considerations will be required for most new products to succeed, most panelists agreed that ML is not completely here yet. This seems to give companies time to catch up and establish a plan to manage AI and ML in their businesses – but there’s no time to waste.


Have another opinion on any of the above questions? We want to hear it! Please share it in the comments below or tweet at us here. And, as always, if you’re an early-stage startup using ML or AI in transformational ways, get in touch!