Sun 17-Mar-2019
Natural Language Processing (NLP) is a domain of artificial intelligence (AI) focused on, well, processing normal, everyday language (written or spoken). It is used by digital assistants such as Siri or Google, smart speakers such as Google Home or Alexa, and countless chatbots and helplines all around the world (“in your own words, please state the reason for your call…”). The idea is to simplify and humanize human-computer interaction, making it more natural and free-flowing. It is also meant to generate substantial operational efficiencies for service providers, allowing their AI’s provide services that were either previously unavailable (human-powered equivalent of Siri – not an option), or costly (human-powered chats and helplines).
Natural Language Generation (NLG) is an up-and-coming twin of NLP. Again, the name is rather self-explanatory – NLG is all about AI generating text indistinguishable from what could be written by a human author. It has been slowly (and somewhat discreetly) taking off in journalism for a couple of years now.1 2 3
NLG is far less known and less deployed in financial services (and otherwise), but given potential for operational efficiencies (AI can instantly, and close to zero cost produce text which would otherwise take humans much more time to prepare, and at a non-negligible cost) it makes an instant and strong business case. There are areas within asset management whose primary (if not sole) purpose is the preparation of standardised reports and summaries: attribution reports, performance reports, risk reports, or periodical fund/market updates. Some of these are so rote and rules-based that they make natural candidates for automation (attribution, performance, perhaps risk). Fund updates and alike are much more open and free-flowing, but still, they are rules- and template-driven.
AI replacing humans is an obvious recipe for controversy, but perhaps it is not the right framing of the situation: rather than consider AI as a *replacement*, perhaps it would be much better for everyone to consider it a *complement* or even more simply: a tool. You will still need an analyst to review those attribution reports and check the figures, and you will still need an analyst to review those fund updates. And with the time saved on compiling the report, the analyst can move on to doing something more analytical, productive, and value-adding. At least that’s the idea (QuantumBlack, an analytics consultancy and part of McKinsey, calls this “augmented intelligence” and did some research in this field which they shared during a Royal Institution event in 2018. You can watch the recording of the entire event here – the key slide is at 16:44. There is some additional reading on Medium here and here).
Some early adoption stories begin to pop up in the media: SocGen and Schroders (who, with their start-up hub, are quite proactive in terms of being close to the cutting edge of tech in investment management) are implementing systems for writing automated portfolio commentaries4. No doubt there will be more.
Disclaimer: this post was written by a human.
1 https://www.fastcompany.com/40554112/this-news-site-claims-its-ai-writes-unbiased-articles
2 https://www.wired.co.uk/article/reuters-artificial-intelligence-journalism-newsroom-ai-lynx-insight
3 https://www.wired.com/2017/02/robots-wrote-this-story/
4 https://www.finextra.com/pressarticle/75910/socgen-to-use-addventa-ai-for-portfolio-commentary
