Let us know about free updates
Simply sign up for Central Banks Myft Digest and it will be delivered directly to your inbox.
The author is a former professor and senior advisor of economics at the Bank of England.
There are many dark visions in the distant future of artificial intelligence. Skynet vs. Humans, as depicted in the Terminator franchise. Super Intelligence running Nick Bostrom’s Amok has transformed the world into paper clips. Or a work-free world with either poverty or abundance – “fully automated, luxurious communism” (depending on who owns the robot).
However, a recent paper – monetary policy sentiment – raises another future outlook that may not be too far away, using modern machine learning technology to crunch a speech by Mario Draghi and Christine Lagarde at the European Central Bank Press Conference. I don’t know if there’s a sci-fi finance film on watches (if you’re stealing an idea, throw some $ as an executive consultant), or even in the industry with a new series, read and take your own judgment.
The ML methods used in this paper are very rough. First, the algorithm is executed, where facial expression video, audio tones, and textual content are data on different types of emotions. Second, these measures of sentiment are compared to financial market data released during and after central bank speeches.
The reasoning in this paper is that the non-verbal information (smiles and frowning) covered by the algorithm is moving through the financial markets. Whether this is causal or correlated, we don’t know for certain. Perhaps the ECB watcher will decode nonverbal information unofficially. Alternatively, there may be correlations as nonverbal information ultimately becomes verbal.
I don’t know how robust this task will be. The merits of this particular paper are not the main issue. The literature for such works is growing and exceeds that. As such research becomes more widely known, replicable and improve, we can imagine central bank watchers dedicating resources to decipher visual and tone information in real time. Perhaps the algorithms used by ECB watchers may decode nonverbal information better than the watchers previously did informally. Your computer can work perfectly and you don’t need to turn off steam-writing jokes in Bloomberg Chat.
The next step in the escalation of the algorithmic war is for central bankers to know that their press conference speeches will be fed into the ML algorithm and try to mitigate their nonverbal communication to ensure that they will affect the financial markets they want. Perhaps they will perform speech rehearsals through their own algorithms built to replicate the watcher’s programs as closely as possible, and try to pre-measure how the watcher’s algorithm judges what they say. Imagine a financial policymaker talking about a small screen bank, guessing what the audience’s ML program is saying through 21st century graphics and allowing him or her to adjust in real time. Or just the equivalent of the producer in the ears of the speakers, “Ms. Lagarde, I need to smile a little” and “raising my eyebrows.”
The arms race then proceeds as follows: The watcher goes back and rinses speeches based on the new algorithm, and deceives a new algorithm that he designed by tricking the old one. Central bankers are forced to do this and raise the game. and so on.
Perhaps algorithms will produce mistakes: inferences about the views on the economy and future policies that policy makers do not hold (for example) will cause yields. These unintended effects need to be responded by more communication, and those communications end again by algorithms and the cycle continues.
Does anyone know where this ends? Perhaps the central bankers will retract to read the pre-written statements and pull back further, and return to release (aivetted) text on the website. In fact, it means a computer that takes over both sides of press conferences, speaking and listening.
If central banking was a clearer science – if everyone on both sides of the table knew what is generating inflation and how to control it accurately, then thousands of processors don’t need to digest the central bank’s speech. But in that world, there is little need for human policymakers.
If central bankers are completely transparent, deceived, ambitious, fearful of being kicked out, or are prone to oppose each other, or forget, the algorithms rarely unearthed. Everything we need to know will be present in policy documents, charts, and forecasts. But then humans are not like that. Because of these shortcomings, we form their committees, and we unite them to average out their frailty and tie them together in the institutions that constrain them.
It’s not difficult to see how these types of techniques can complicate speaking in all kinds of delicate public. The Finance Minister has been delivered or interviewed about the new fiscal plan, concerning the impact that it will have on the funding costs of the plan. The CEO of a public company hopes to give a speech on forecasting new revenue for shareholders, achieving the most advantageous reception possible. Political leaders negotiate with each other to lie and lie. Congressmen listen to their leaders and vice versa. Corporate lawyers over terms of merger or acquisition.
I don’t think humans are still doing substantial policy making and business at some point in the future, and have not been able to get out of work through artificial intelligence programs. Perhaps the real AI singularity we need to worry about is not the solar system of paperclips, but the arms race of algorithms deployed by speakers and listeners in all public sectors, wasting a huge amount of computing power that allows for more productive use.
