Artificial intelligence could help humanity to overcome some of its very human frailties.

Zia Chishti, the chairman and chief executive of Afiniti holds that AI is especially good at “the identification of patterns in complex data.” “Do not believe the hype about artificial intelligence” October 18.

I agree but sometimes we also need help identifying what we humans for some reasons are unable or find hard to see in very simple data. 

That’s why I have often tweeted out asking for IBM’s Watson’s (or Goggle's DeepMind) help, in making bank regulators understand that their risk weighted capital requirements for banks have gotten it all wrong. Had they really looked at simple data on bank crises, they would have immediately seen these were never caused by assets perceived ex ante as risky, but always by assets that when booked by banks were perceived as especially safe.

Perhaps the safe-is-risky and the risky-is-safe might just be a too hard concept for humans to swallow and act on. If that’s the case, in AI’s assistance we trust.

PS. Another useful factor is that AI has the advantage of being able to admit mistakes easier, as it has (I pray) a smaller ego than humans

PS. Here is an aide memoire on the mistakes in the risk weighted capital requirements for banks.

@PerKurowski

Janet Yellen, Mario Draghi, ask IBM’s Watson what algorithms he would feed robobankers, to make these useful and safe

Sir, Sam Fleming reporting from Jackson Hole writes “Janet Yellen, the Federal Reserve chair said regulatory reforms pushed through after the great financial crisis had made the system “substantially safer” and were not weighing on growth or lending. … If the lessons of the last crisis were remembered “we have reason to hope that the financial system and economy will experience fewer crises and recover from any future crisis more quickly”, “Yellen warns opponents of tighter financial rules to remember lessons of crisis” August 26.

As I see it Yellen has not yet learned at all that past and future financial crisis have not, nor will ever, result from excessive exposures to what was or is perceived as risky, these will always result from unexpected events, like when that was perceived, decreed or concocted as very safe, turned out ex post to be very risky.

Since regulators do not to want listen to anything else but their own mutual admiration net-works’ risk biases, I wish they would contract IBM’s neutral Watson to ask it the following:

Watson, while considering the purpose of banks as well as the real dangers to our financial systems, what algorithms would you suggest feeding robobankers with?

THEN Yellen, Draghi and colleagues should compare that algorithm with what they are feeding the human bankers with; the portfolio invariant risk weighted capital requirements that assumes that bankers do not see or clear for risks by means of size of exposure and risk premiums charged.

Then these regulators would understand that with their over-the-board incentives for banks to invest or lend to what is safe, like AAA rated securities and sovereigns, like Greece, they are in fact creating those conditions that dooms banks to suffer huge crises, sooner or later, over and over.

Then these regulators would understand that their regulations induce banks to stay away way too much from lending to what is perceived risky, like SMEs and entrepreneurs, something which clearly must weigh heavy against the prospects of our real economy to growth.

Janet Yellen, Mario Draghi, please ask Watson! Perhaps you could find him on LinkedIn 😆

@PerKurowski

How many qubits would we have needed to install in the Basel Committee in order to avoid its monstrous mistake?

Sir, Jeremy O’Brien writes: “The catch is the need to correct the inevitable errors experienced by qubits. Because quantum computers are prone to errors in ways that conventional computers are not, it is currently understood that to create a quantum computer with 100 logical qubits, we would need a system with about a million actual qubits.” “The future is quantum” March 25.

Our bank regulators created capital requirements based on ex ante perceived risks of assets, instead of on the risks that assets might pose for banks ex post. That is why they came up with risk-weights of only 20% for what is AAA rated, where there could be the buildup of dangerously excessive exposures, and 150% for the so innocuous below BB- rated.

Sir, I wonder how many qubits a scientist like Jeremy O’Brien believes we would have to install in the Basel Committee, or in the quantum computer controlled bank regulations we might have in the future, in order to avoid costly errors like that.

I guess it should not take too many… I guess IBM’s Watson would already not make such a mistake… as it would clearly start by asking: “what is the purpose of banks?” and what has caused major bank crisis in the past.

http://subprimeregulations.blogspot.com/2016/12/must-one-go-on-hunger-strike-to-have.html

@PerKurowski

We would appreciate Google’s DeepMind (or IBM’s Watson) giving the Basel Committee some tips on intelligent thinking

Sir, I refer to Clive Cookson’s “DeepMind overcomes memory block to bring thinking computers a step closer”, October 13.

Here again we read about so much research going on in the world of artificial intelligence. Though clearly still a lot needs to be done, the current advances could perhaps suffice in order to give some good tips to some humans who do not seem to be able to get their thinking quite right.

Yes! You’ve guessed it Sir. I am indeed referring to the Basel Committee of Banking Supervision and their risk weighted capital requirements for banks. Perhaps it would be easier for the regulators to hear out some observations on how to regulate banks, if it came from an impressive “differentiable neural computer” with AI capability, and not from a simple non-expert human like me.

So, if Google’s DeepMind (or IBM’s Watson) were able to only convey the importance of first defining clearly the purpose of banks before regulating these; and second to do some empirical research on why bank systems fail, that could be extremely helpful for the banks, for the real economy, and of course for the future of our grandchildren.

Then regulators, swallowing their pride, could perhaps, with luck, understand both that the main social purpose of banks is to allocate credit efficiently to the real economy; and that no major bank crises have ever resulted from excessive exposures to what was ex ante perceived as very risky, as these have always resulted from unexpected events, or from excessive exposure to what was ex ante considered very safe, but that ex post turned out to be very risky.

That could help to free us all from our banks being guided by dumb risk-weighted capital requirements… more ex ante perceived risk more capital – less risk less capital.

Not only do these cause our banks to misallocate credit to the real economy, like no credit to “risky” SMEs or entrepreneurs; but also to make our bank system more unstable by pushing the build-up of exposures to what is perceived, decreed or concocted as “very safe”, without requiring sufficient capital to cover for the unexpected events.

PS. DeepMind, or you Watson, if you would also care to explain this to those in the Financial Times, that would be doubly appreciated. I have tried to do so with literarily thousands of letters, but still no luck… I guess I am not as impressive as you are.

@PerKurowski ©

IBM, Watson could have a role in regulations that accept the need of the real economy for banks to take credit risks

Sir, I refer to Richard Waters report on the difficulties IBM faces in expanding the application of its Jeopardy champion Watson, “FT Big Read: Artificial Intelligence: Can Watson save IBM” January 6.

In it quotes Lynda Chin mentioning the challenge that “On Jeopardy! there’s a right answer to the question, but, in the medical world, [in the real world] there are often just well-informed opinions… [So how to know] how much trust to put in the answers the system produces. Its probabilistic approach makes it very human-like… [Watson] Having been trained by experts, it tends to make the kind of judgments that a human would, with the biases that implies.”

Indeed how much trust is just another way of stating how much risk is one willing to take.

For instance if one wants driverless cars to provide absolutely security, then traffic will probably become very slow, or even come to a standstill. And one of the difficulties these cars will encounter will be based on defining the acceptable amount of risk taking.

Likewise, if one wants our banks to be absolutely secure, then one would be better off with hiding money under mattresses in bank vaults… but the real economy would be languishing because of the lack of credit.

So there might be a big role for Watson in bank regulations. First of all it could help me convince the Basel Committee of that their credit risk weighted capital requirements are based on a very faulty human bias against risk; something which at the end of the day only endangers banks, since it causes excessive exposures to what is perceived as safe, precisely that which has caused all major bank crisis.

And, if fed with continuous information on bank credit and the state of the real economy, Watson could also be used to automatically send out countercyclical adjustments. Too much growth in credit… increase capital requirements somewhat… too little growth in credit reduce capital requirements somewhat. The most important thing needed for that would be to make Watson immune to lobbying pressures of all sorts.

What I would not allow Watson to do though is to display that kind of human arrogance of thinking itself capable of setting different capital requirements for different assets, so as to distort the allocation of bank credit as it thinks fit to distort.

To do that, I would still want a human to be behind that kind of risk taking… of course a human who understand what he is doing and is willing to be held very much accountable, if taking the next generations down the wrong path.

@PerKurowski ©