In the sixth and final episode of the Sorcero series on building AI-powered life sciences products, Dipanwita Das speaks with Rackhouse Venture Capital Founding and Managing Partner Kevin Novak. They discuss why funding in life sciences suffers from lagging expertise on artificial intelligence (AI) technology. They also talk about the reality that AI is still very early in its development, and why futurecasting — the ability to have increasingly accurate forecasts for complex scenarios — may be a generalizable principle with wide application in machine learning technologies. Finally, they consider the issue of ethics, including the question: even if AI is better than a human being in a certain area, when and how do we consider it good enough?
On the impact of lagging expertise in life sciences funding
When an artificial intelligence founder comes to a venture capital (VC) firm for funding, there are many considerations to keep in mind. Top of mind is demonstrating that this is a good product with good product-market fit. But there is a major obstacle: many VCs might not have AI experts on staff to accurately assess the technology at hand. Not only that, but since a VC is involved in a systematic exercise of building out a portfolio, there is only so much time to assess a company.
Kevin says founders would be well-served to understand that diligence process a bit more. And they should also find compelling ways to present their technology to non-experts. He says, with AI there is the fundamental question of “what is this technology? Is it differentiated? Are their forecasts for how this product is going to evolve realistic?
“When you don't have a lot of domain expertise, that last bucket of diligence, on a hard AI product, takes a ton of time. If I've never seen natural language processing before, and there's this thing called embeddings, if I have to go from first principles all the way to is this company investment grade? in 40 hours... Most firms just throw up their hands…”
Due to a lack of technical expertise, many VCs simply invest in startups similar to other companies that have been invested in by competing VCs. “So,” Kevin says, “you end up with like a bunch of over-competed segments [with] companies that are similar to each other.”
“Rackhouse … has a very technology specific thesis to make my diligence process easier. But also, it's just something I personally love. And we're very industry and sector agnostic. And the reason we're able to do that is because 80% of my diligence time is spent learning about the industry, in the sector and all the other stuff. So I think the lack of creativity in investing … and the lack of domain expertise are pretty intrinsically linked, just from the nature of how a firm diligences a company.”
On why AI futurecasting should get investors excited now
Access to data is not a problem in 2022. There are many data points about any given topic. But is that data usable? This is where AI comes in. It can take a huge data set, distill it, and enable accurate models of trends that help suggest improved behaviors. That could be related to everything from an individual’s financial decisions to better health practices.
Kevin says a fundamental value proposition of AI technology now is futurecasting. In other words, “the ability to have more accurate forecasts further out in the future for increasingly complex scenarios. For example, consider what Sorcero is working on. That is the ability to take increasingly diffuse, complex concepts — in Sorcero’s case embedded in the language — and distill them. That creates a framework that computers can understand, that human beings can understand. That distillation is very fascinating and very valuable to businesses, when you think about what that means for the future."
Better forecasting with AI
“Especially in the industry and sector space, I'm very intrigued by what better forecasting with more diffuse data can do for risk, for lending, for climate modeling…. That same model, which can help health insurance companies price insurance better, or price their risk better, can also be used to suggest [to consumers], do [these] three things to improve your health. Or do three things to … meaningfully move your premiums in your favor.… And here's the prescriptive causal relationships between all of these things.
“When you think about the reams of information out there across all of these different sectors, the sum total of published documents, of strings of characters since humanity has started writing stuff down, and how much of that is functionally walled off from individuals, from computers, and what we can glean from that, I think [it] is really fascinating. The stuff around drug discovery; you see a lot of headlines [about that]. I think that is the leading indicator of something much more generalized.”
On AI and ethics
Two major areas of AI application are health and finance. And since these arenas have a big impact on people’s lives, builders of these AI applications must contend with regulators. They also run into issues with ethics.
Kevin says that the mere existence of this debate is a healthy indicator. He says, “Even the fact that people are anxious about this is itself an indicator of health in society.… In Fascist states, there's very little debate. So first of all, the fact that we are having a debate makes me feel good.
“One of the things is when most people, even product managers, are looking at different technology choices with different approaches. It's a subjective comparison of options on the table. Does AI do this better than people? Does Deep Learning do this better than a rules engine or something? And it is a [subjective] assessment. When you shift the question to 'is this ethical or not?' It is an objective assessment against a third party framework.
“So it is possible to live in a world where a company can say with confidence… AI issuance of mortgages is generally more ethical than human beings. And if you think about problems with redlining, or communities being marginalized, that's [an objective] comparative statement. The other question — is AI mortgage lending ethical enough? — is a different question…. I think people talk past each other. It can both be the least bad option on the table, and still not good enough, or vice versa.”