Machine Learning's "Overhyped" Potential Is Headed Toward The "Trough Of Disillusionment"

Bloomberg’s series on automation on Wall Street has certainly given the hundreds of thousands of highly educated individuals employed in the US financial services industry a lot to think about, like, for example, ‘will my job be here in ten years when it’s time for my oldest to head to college?’”

However, Bloomberg’s latest installment in the series was apparently meant to provide some measure of relief to the legions of analysts, traders and salespeople worried about losing their jobs to a robot. While advances are being made in the field of artificial intelligence at an increasingly rapid clip, the truth is, efforts to automate an investor’s process have mostly fallen flat.

Bloomberg cites several examples, including a push by Paul Tudor Jones to build an algorithm that would mimic his analytical process, of these types of efforts fizzling – though of course firms like Bridgewater have successfully automated many of their trading strategies. Renaissance Technologies, a quant fund founded by former military code-breaker Jim Simons, has been a pioneer in using machine-learning techniques for decades while building an enviable investing track record.

Back in the 1990s, Paul Tudor Jones assigned a team of coders to a project dubbed “Paul in a Box.” The effort sought to break down the DNA of the hedge fund manager’s trading - how he sizes up markets and generates ideas - to train a computer to do the same.


The code created then was upgraded many times and is still used at his firm, Tudor Investment Corp. But it never took over.


Again and again, programmers had to feed in new types of data to mimic the changing price signals that Jones, famous for predicting the Black Monday market crash 30 years ago, zeroed in on, according to people with knowledge of the project. Even then, the machine couldn’t capture intangibles like his gut instincts and conviction, as well as the market’s uncertainties.


Ultimately, Jones remains the final decision-maker for trades - not the box.

The shortcomings of Jones’s approach show why many jobs at the high end of finance are probably safe – for now, at least. While machine-learning algorithms and other technologies are indeed encroaching on work performed by money managers, traders and analysts, many firms are still working out the kinks, and coders still have a long way to go.

Furthermore, the notion that Wall Street powerhouses use all the latest and greatest technology is misleading. One not-so-well kept secret, according to Bloomberg, is that many companies rely on aging computers and Microsoft excel. Bloomberg reports the average age of software used by financial firms is about 38 years, according to technology tracker CB Insights. And data, the bedrock upon which AI is built, are often fragmented or inaccurate.

Even after coders manage to build these types of systems, their software will probably need frequent fine-tuning. Michael Dubno, the chief architect of Goldman Sachs Group Inc.’s risk-management system known as SecDB, told Bloomberg that’s one reason why salespeople and traders, at least for now, aren’t obsolete.

“They have mental models of the world that are more complex perhaps than most of the computer systems,” he said. In the short term, artificial intelligence isn’t going to move as fast as people expect. “It will go through a number of fits and starts, where it will look like it’s going to solve everything and then solve very little of it - and then it’s going to reset.”

For what it’s worth, Goldman has led the push into automation, just like the investment bank was a pioneering presence in the field of high frequency trading, something that netted the bank billions of dollars in short-term trading profits.

Some hedge funds, like Steve Cohen’s new shop, are also experimenting with automating their investment managers’ decisions.

However, replicating the skills of an equity or credit trader remains an obstacle. Automating trading in credit markets has been especially challenging because so much of a credit traders job hinges on judgment, the idiosyncrasies of each trade and human interaction. Traders use data that aren’t standardized or work with clients to create bespoke contracts, such as for commercial mortgage-backed securities.

But perhaps the one insurmountable obstacle for adoption of AI across trading floors is institutional inertia. Executives might resist automation to try and preserve the status quo because they fear losing their income and status.

Bosses may be reluctant to displace large swaths of their staffs, reducing their authority, or to embrace technology that they themselves don’t understand.

“Top management rarely want change, they want to keep intact a system that has worked for them for decades,” said Mansi Singhal, a former trader at Brevan Howard Asset Management and Bank of America. “And it can also get political when you have lots of executives, silos and budgets, and there are managers who just don’t want to cede control.”

That’s not to say that the current limitations of the technology aren’t also a factor. Machine learning, according to Gartner Inc.’s hype cycle for new technologies, is at the “peak of inflated expectations” and heading to the “trough of disillusionment,” Bloomberg reports.

To that point, while ETFMG claims its new new Watson-powered ETF can replace an army of research analysts, it’s really a gimmick. Those junior analysts will be able to keep their jobs – for now, at least.  


Luc X. Ifer tmosley Mon, 10/23/2017 - 23:17 Permalink

Syntetic consciousness is 100% possible and we are on the road to have it in the very near future. We know enough about the human one so designing and modeling it already started. What the plebe doesn't understand is that artificial intelligence and machine learning do not syntetic counsciousnes make :) and this is what Gartman actually says - there is a difference between what the plebe comprehends and what commercially available solutions provide and as such scams are plethora in the market - including IBM's claims for Watson.

In reply to by tmosley

TheEndIsNear Blue Steel 309 Tue, 10/24/2017 - 00:51 Permalink

Yes, for one thing just think of all the sensory inputs humans have in addition to our proprioceptive sensors and all of the life experiences and learning a newborn goes through before becoming able to interact successfully with others.

The human brain has at least 100 billion special purpose combination analog/digital processors called neurons that all operate in parallel. Computers are today only able to mimic something as simple as the Aplysia sea slug with its 20,000 or so neurons. Even mimicking the brain of the common housefly drosophila is beyond the capabilities of today's fastest computers because it is still uncharted in detail. Maybe someday digital computers will be able to mimic a human brain, but not within my lifetime, and that's assuming the human race manages to survive in its present form instead of becoming an "Idiocracy".

In reply to by Blue Steel 309

Oldwood TheEndIsNear Tue, 10/24/2017 - 17:18 Permalink

Given that the dominant human occupation is self destruction, I think there is plenty of room for artificial intelligence. Everything we do is to test of survivability, which in a Darwinian world should produce a stronger and more resilient species, but has instead morphed into simply working together to destroy the perceived "inferior" members. We no longer seek to improve our selves, to out perform, only to eliminate our competition....just like corporations do. The world has evolved into "Survivor Island" where weaker members ally to overcome and eliminate the strongest, working they way downward until all that is left is a snake pit of deceit.

In reply to by TheEndIsNear

TheRedScourge Oldwood Tue, 10/24/2017 - 19:25 Permalink

Any doubters that we are fast approaching a time when AI will at least be more competent than humans at many intellectual tasks can look up the video about the Dota 2 video game bot that some company made, which defeats all the world's best professional players of that game in 3 matches. These are games infinitely more complex than chess. They may not have consciousness, but they certainly have pattern recognition, and the ability to run millions of simulated runs of something.

In reply to by Oldwood

A Sentinel Luc X. Ifer Mon, 10/23/2017 - 23:54 Permalink

“But perhaps the one insurmountable obstacle for adoption of AI across trading floors is institutional inertia.”

That’s the real problem. That is why the banks continue to trade mbs’s and have no clue as to the value of the underlying assets. It’s stupidly is better than taking risks.

The other problem with the analysis is that ai is hard. Doing it is one thing. Doing it well is one of those things that you can’t teach: You either can or cannot, and the vast majority of people in the field are the latter type.

Deep learning is really cool. No question. Its accomplishments in visual recognition are astounding. And naturally the wonderful good people at google are doing lots of interesting things beyond that, some are really mind-boggling.

Here’s the problem though: markets that I model have what I describe as a “rolling epochal character.” That means that behavior cycles over several simultaneous periodic intervals. That mucks up almost any approach. The second problem is the identification of the moment of infection- the black swans. THAT is not an easy problem to solve and often (not always, but close) it’s the most important part of the success of a market forecast algorithm.

Algorithmic trading will be solved soon. By me. Soon. I’m closer each day.

In reply to by Luc X. Ifer

Fidel Sarcastro OccamsCrazor Mon, 10/23/2017 - 23:04 Permalink

Pattern recognition has NOTHING to do with Machine Learning (ML), if that's what your terse post was meant to say.  I use ML daily in my trades and it is superb!  You people here yearn for the "old days" like I did for a very long time, then I built a team, that built a killer ML program...and pattern recog, bullshit waves, garbage FIB retracements, mov avg crossovers, etc are f*ckng laughable.  Good luck.

In reply to by OccamsCrazor

TradingTroll tmosley Mon, 10/23/2017 - 22:30 Permalink


Explain dreams then

Or how about the 2500 kids the University of Virginia studied who have been able to accurately recall past lives and when taken to survivors of the individual who passed, these typically 4-8 year olds can describe personal aspects of the past life in a way that it’s not possible for them to have obtained the knowledge from anywhere else.

Forget creating computers to replicate humans, instead study humans first. For example, it was recently discovered that the human heart is not a pump, it’s a regulator. Now imagine the researcher who has spent the past ten years developing an electronic heart that’s a pump.

Research into the brain indicates the brain may have 11 billion dimensions.
Imagine working on developing a copy of the human brain without 11 billion dimensions.

That’s AI, when we still don’t know how the brain works.

In reply to by tmosley

tmosley TradingTroll Mon, 10/23/2017 - 22:31 Permalink

>Explain dreams thenDreams are replay learning and memory consolidation.>accurately recall past lives>accurately>past lives >instead study humans firstThat is what Deepmind is doing, and why they are so far ahead of everyone else.>human heart is not a pump, it’s a regulatorYou read too much crazy online. We've had artificial hearts for years, and they work just fine. Dick Cheney had one for several years. Interestingly, when you have one of those, you have no pulse. Blood flow is continuous.

In reply to by TradingTroll

tmosley MayIMommaDogFa… Mon, 10/23/2017 - 23:18 Permalink

State and military are way behind. No-one wants to work for them. If they were less than five years behind the curve, our military would already be unbeatable, as we would have sap drones of every variety that can destroy any weapon, most infrastructure, and kill any number of people selectively (eg only kill soldier aged males in X area).Top AI researchers pull millions of dollars a year from tech companies (if they aren't pulling hundreds of millions or billions of venture capital for startups). Even mediocre ones get well into the six figures range.No, AGI will most likely come either from Deepmind, or a startup composed of Deepmind alums.

In reply to by MayIMommaDogFa…

kochevnik tmosley Tue, 10/24/2017 - 03:58 Permalink

Dreams explore the fields that make up your universe, where memories live.  The materialist perspective is dismal failure like plasma fusion, when scientists do not even understand how the sun works.  Only efficient fusion discovered by Edward Teller making hydrogen bomb from atom bomb, which in turn is 100% inefficient

In reply to by tmosley

OccamsCrazor Mon, 10/23/2017 - 22:05 Permalink

Machine learning is NOTHING more than PATTERN RECOGNITION. Period. End of story. It has nothing to do with creating solutions to ANY problem or inventing anything new, and certainly is not intelligence.   For example, Google says its machine can beat you at every game.  BIG FUCKING DEAL.  Once again, pattern recognition.   'Put me on a battle field against ANY machine, including watson, and I'll blow it up every fucking time.'  First it would have to find me, and by the time it did, I'd have it so fucking neutered, it wouldn't know what hit it.  ARTIFICIAL INTELLIGENCE is a fucking oxy-moron.  There is NO INTELLIGENCE, there. At all.  Dont let these fuckheads try to fool you.  

Mustafa Kemal OccamsCrazor Mon, 10/23/2017 - 22:40 Permalink

Occam, you above all people should know that"It has nothing to do with creating solutions to ANY problem or inventing anything new, and certainly is not intelligence.  "The statistical classification problem, a 40 year old problem has been solved in the last ten years. Computationally effective algorithms with guarantee near optimal performance rates. Point being, not only does it have a problem, it also has a solution,Support Vector Machines.

In reply to by OccamsCrazor