AI algorithms – takeaways from Fintech+

The Fintech+ conference with its AI thread was unique. The morning sessions included presentations from Nvidia and Google, and use cases and learnings from Zurich Insurance. Leading into the sustainability & Fintech panel that I moderated just before lunch.

Marc Stampfli, Swiss country manager at Nvidia took us on a journey of AI Fall, Winter and into Spring. He explained neural network concepts borrowed from biology and the initial difficulties of neural network computations outperforming statistical approaches. The 1st tipping point came with increased data availability through the internet, and only then we had evidence that neural networks could outperform statistical models.

After that point, we ran into the next problem which was the lack of computing power to process all this data and multi-layer neural networks. And this is where GPU – a kind of parallel computer –  was created and first used in vector mathematics. This is the technology of Nvidia’s processor.

For me, this historical thread is another example of a solution designed for theoretical mathematics that finds a real-world application that takes us to the next level of the 4th industrial revolution. I associate it to the zero-knowledge proof in cryptography, now used in some blockchain protocols, that allows to verify & validate data without having to trade-off privacy[1].

We are living in a world in which, more or less unconsciously, we increasingly “Trust in Math”. After the GPU adoption in business, we moved to new hardware that is not only faster but also smaller in size. We basically reinventing how data rooms looked.  And this the world from Nvidia’s angle. They have facilitated the growth and new value creation, all powered by #AI tools.  The use cases in Finance are immense. Fintech solutions for:

  • Operations: automating claims processing and underwriting in insurance
  • Customer service & engagement: alerting customer for fraud, chatbots, recommendations
  • Investing/Trading: automating research, trading signals, trading recommendations
  • Risk & Security: fraud detection, credit scoring, authentication, surveillance
  • Regulatory & Compliance: AML, KYC, automating compliance monitoring and auditing.

Evidently, the biggest but fundamental problem that incumbents face in adopting any of these potential use cases, is that they first need to find ways to integrate their data and then to upgrade their data rooms to be able to handle the required computing power.

Having said that, Zurich insurance, one of the large Swiss insurers, shared with us their AI projects and research which started as early as 2015. Gero Gunkel spoke about their very successful AI applications in automating the review of medical records with the aim to arrive at a valuation. A process that entails reviewing reports ranging from 10-40 pages and that may take on average 1hour. They used AI algorithms that reduced this to 5 seconds! That is nearly real time for a business process that is Not low hanging fruit.

Zurich Insurance has also been using AI to automate the time-consuming process of collecting publicly available information towards opening accounts for large corporates. This automated web search can not only offer efficiencies but also become a new service provided to the underwriters of these types of insurances.

“Don’t look for the Swiss army knife”, said Gero Gunkel as AI may seem so promising that one can think it can take care of everything.

Dr. Christian Spindler,  IoT Lead and Data Scientist at PwC Digital Services, raised the important question on how to develop Trust in AI. This is a tricky topic as it beckons for answers around the limits of technology. For now, it is recommended to develop AI algorithms that can also provide explanations for their “Answers”.

I would say that “In Math we Trust” to develop algorithms that Answer “What & Why”.

“Improving lives through AI” is Nvidia’s motto for their Corporate Social Responsibility. See their initiatives here.

[1] Zero-knowledge proof allows a someone to re-assure a validator that they have knowledge of a certain “secret” (data) without having to reveal the secret itself. Zcash is an example of such a blockchain protocol.

Efi Pylarinou is a Fintech thought-leader, consultant and investor. 

 Get fresh daily insights from an amazing team of Fintech thought leaders around the world. Ride the Fintech wave by reading us daily in your email.