It has been said that people do not become wiser with age, they just become “more”
What we’ve done well – and more importantly, what we’ve done badly – has been magnified. The same is true when we add computers and data to human decision making.
Algorithmic / machine learning / artificial intelligence (AI) tools are increasingly ubiquitous in the investment world. They determine the risk tolerance of investors in portfolio management and are applied to the selection of alternative information among other tasks as well as the selection of actual securities.
The debate over whether to “use AI” is a sensitive one: these tools will appear in the process of investor research and hold research, even on the most fundamental-based considerations. The right focus then is on “model awareness”: how can we harness the fact that machine learning, alternative data and AI are not just widespread, the impact is growing?
Model awareness is our term for how to think about machine learning, AI, big data sets, etc., or the spectrum of capital market rules, machines, or data-driven processes. In order for the model to be aware, each trustee, assignee and manager should start with an overall focus on the process question: where are the most opportunities and risks?
It lies with people.
Remove human drivers and pedestrians from the road and self-propelled cars will work perfectly. Collaboration between humans and machines is a “minimum bandwidth” connection. Think about how easily we can turn a door tube and walk outside or a computer can present a complex image. Compare how difficult it is to represent our problem or get feedback on its consequences. Human-machine collaboration is both the key to success and the opportunity to exploit vectors.
The problems and opportunities are how we see computer- and model-based approaches in the market. They are either in our team or in another team.
Humans and machines can examine each other’s perspectives: can we replicate existing human outcomes with a machine-learn model? And if so, what do our standard tools tell us about the flaws in the model?
We can “counter” the predictions of the models that computers make and will reliably like or dislike them.
The concept of “alpha decay” is real. Something is coming to take our alpha generation away. We can use the flaws of human-machine collaboration to exploit that problem by seeing each other as adversaries.
Adversarial machine learning is a tool and strategy that seeks to overcome intelligent opposition. For example, a team of researchers using sophisticated deep learning networks used image-boring spectacle frames to identify Reese Witherspoon as Russell Crowe.
Even the most advanced, well-defined problem space can be resisted. What can we learn from this? It is important to monitor and coordinate models to address “intelligent anti” behavior. A simple working method is to create a “red team” for an existing prudent approach or to form a human red team to deal with a model- or rule-based strategy.
The concept of “red team” has been borrowed from espionage and military agencies. This means reading the same facts, playing the devil’s advocate and creating an internal opposition to support the opposite decision. Each of us has its own unofficial version of the red team: we think about manipulation of GAAP / IFRS earnings vs. slips from cash or large block trades and revise our analysis and plans accordingly.
To formalize such a red team model, we can incorporate these methods into our data set with additional “counterfactual” data points and act as if an intelligent adversary is trying to resist us. It echoes Nasim Taleb’s call for clarification on how our methods will work. “All possible worlds,“We didn’t just have a world in mind. In this way we can create strategies that benefit from decay and disorder.
Hybrid Human-Machine Behaviors
After we separate ourselves from the machines and “inspect” each other, we must remember that humans and machines are not really different. Machines often mimic human social biases. Human-machine collaboration can improve some bias, but it can make it worse, create or transform another:
- Improve: Decision-making from human hands can alleviate or even resolve some behavioral biases. For example, a hedonic treadmill ক্ষতি feeling the loss more intensely than gain-is not a problem for a well-configured algorithm.
- Bad How models are designed – often their assumptions, parameters, hyper parameters and interactions with people – can exacerbate some of the problems. The relative volatility spikes between markets and asset classes are strongly tied to this increased impact. Computers quickly recover from the asymptotes of their parameters and are almost like a mathematical “reflective boundary”.
- Creation: The continued growth and reliance on model-, rule-based and new data sources has led to new behavioral biases. Problems with “hybrid” human-machines include the effects of the black box. This indescribable result – the oscillation of interrelated instability, for example – develops out of nowhere and mysteriously disappears. Hidden machine-machine interactions can also pop up, such as “machine learning reunion” where machines conspire with each other without human instruction.
- Conversion: Dimensions of human behavior take on new forms when they are bound to computing or data sets. Peak-end rules, in which the best and worst aspects and the end of an event are felt more intensely than the remaining experience, are presented in fancy ways when people and machines cooperate.
What can we do today? We can start by thinking about how this set of collaboration gaps affects our strategies. Can we “red team” or “counter” our models and human processes? Which hybrid behavior dimensions will change our basic ideas about how people view the world?
If you liked this post, be sure to subscribe Entrepreneurial investors.
All posts are the author’s opinion. As such, they should not be construed as investment advice, or the opinions expressed must not reflect the views of the CFA Institute or the author’s employer.
Photo Credit: © Getty Images / Dong Wenji