• How AI Just Defeated World Class Poker Players Over 20 Days
  • The Coming Flood of New Investment Intelligences

Yet another sudden and surprising landmark for Artificial Intelligence this week…

News from the Rivers Casino in Pittsburgh has filtered through that a poker playing AI has trounced human players after a gruelling run of 20 straight days and 120k hands.

Libratus, a program developed at Carnegie Mellon University, was trained to play a variant of the game known as no-limit heads-up Texas hold ’em – an especially complex game where strategies play out over dozens of hands.

While the tiring humans slept it used the time to hone its performance, learning from its growing experience and the accumulating data from hands played.

“The first couple of days, we had high hopes,” said Jimmy Chou, one of the professional gamblers.

“But every time we find a weakness, it learns from us and the weakness disappears the next day.

It’s like a tougher version of us,” he said.

An AI with a Trillion Regrets

Is this the moment when AI learns to trump human intuition?

Not quite.

Libratus won through a rigorous course of reinforcement learning (basically extreme trial-and-error).

What interests me is how Libratus independently developed its strategy – one that the humans couldn’t compete with.

It did this by using three layers of AI.

The first algorithm relied on counterfactual regret minimisation.

In short, it started playing at random and tracking regrets over trillions of hands of poker in training, it discovered ways to play the game in ways that humans couldn’t…

It then used an end game solver – looking at the state of play at the table and running through possible scenarios.

The last algorithm ran at night….identifying the weaknesses that the humans had picked up on.

A Flood of New Intelligences

Libratus has already received inquiries from companies across the spectrum: from those developing military and cybersecurity strategies, to medical treatment planning.

I expect this type of AI to appear in apps for individuals and organisations in e-com, loan negotiations and a great number of other scenarios where a little human deception will pay off.

But it is also indicative of the new breed of algorithms that are being adopted across the investment community…

After all, Texas Hold ‘em is a complex game….the opponents’ hands remain hidden with imperfect information, indeed withheld information. Bluff and deception are the very essence of the game.

It is no coincidence that many successful hedge fund managers are top notch poker players.

So just imagine the impact on stock market trading and asset allocation among those poker playing hedge fund managers and others besides.

We’ve seen how AI-driven hedge funds are outperforming the rest of the market, making eerie moves, spotting patterns, making wild connections.

It’s not just a case of outperforming humans – most AI is deeply disappointing and primitive when compared to human faculties. It lacks common sense. It doesn’t in any way know what a good investment actually is.

And it is difficult for investment managers to understand and assess and is often ‘too brittle’ where there’s a very low signal to noise ratio and/or high uncertainty.

What’s more there are plenty of badly coded algorithms out there making specious connections and reading false signals. Tyler Vigen, the author of “Spurious Correlations“, notes that margarine consumption is closely linked to divorce rates in Maine, and Nick Cage films to swimming pool drownings.

But the point is that we are beginning to release algorithms, like Libratus, that can train themselves. And they are very capable of developing successful strategies to outsmart humans.

In the coming years, we are going to create many, many new types of intelligence to solve specific investment problems…
— An algorithm that predicts social unrest.

— An algorithm that looks for black swans.

— A master algorithm that organises other algorithms.

— An algorithm capable of cloning itself many, many times.

— A algorithm using quantum computing whose logic is not understandable to us.

In prospect: a stock market that is colonised by deceptive, self learning algorithms that are not just focused on high frequency trading and scalping, but also strategising, asset allocating, perceiving risk, arbitraging and looking to completely frustrate and compete with professional investors.

In other words: A restless market of competing non-human Librati.

I’ll have more on this subject in a report I am preparing on the coming ructions in the investment community from increasingly autonomous and deceptive AI.

In the report, I look at the opportunities, the best investment strategies for outplaying AI and why I think the most impressive advances — in terms of AI driven wealth and asset management — could come from China.

There are also some strange and rather disturbing implications for the stability of the stock market.

More on that once I have completed the research in the coming weeks.

Drop me a message here if you have any queries or questions.

Follow me on twitter here.