OSL launches pilot for ChatGPT-driven trading bot

OSL launches pilot for ChatGPT-driven trading bot

Source Node: 2595378

OSL, operator of a digital-asset trading platform that is licensed in Hong Kong, is going all-in on artificial intelligence.

Hugh Madden, CEO of OSL and its listed parent company, BC Group, says the company is now making wholesale replacements or restructurings of all functions, back end to front, based on language-learning modules (LLMs) such as ChatGPT.

This week, that effort went from internal restructuring to external, client-facing efforts, with a pilot test of OSL’s purpose-built AI trading bot.

“The trading bot is just one aspect,” Madden told DigFin. “The context is much larger.”

ChatGPT

Madden, formerly a chief technology officer at institutions such as HSBC, had been playing around with earlier versions of ChatGPT, the AI chatbot created by US tech company OpenAI. He obtained early access and was impressed.

“We were a decade ahead of where we expected AI to be,” he said, adding that Google’s 2017 white paper on transformers, “Attention is All You Need”, should be held up as a decisive moment, akin to Satoshi Nakamoto’s 2008 white paper outlining Bitcoin.

But while Bitcoin hasn’t evolved from its original construct, AI has, especially in the branch of transformers (machine learning that uses transfer learning, that is, applies knowledge from solving one task to a related problem).

OpenAI released GPT3.5 in 2022, and was then acquired by Microsoft for $10 billion. It quickly followed up this year by releasing GPT4.0. Madden, both amazed and alarmed, moved quickly to overhaul OSL.

“At the start of the year, I said every person at OSL, from customer support to compliance, needs to be trained on AI tools and use them.”

Three phases of AI adoption

Madden has a three-stage framework for understanding where businesses will find themselves with regard to AI.

Phase 1 is getting acquainted with the technology and using AI tools. That is probably where most sophisticated companies, such as banks, are today.

Phase 2 is wholesale replacement of traditional functions with AI. Madden says OSL is now in this phase.

Phase 3 is where corporate AI, powered by access to relevant sets of data, accounts for the majority of transactions, both in volume and value terms.



Madden says “polite discourse” reckons we reach Phase 3 in five years, by 2028. But he thinks for many businesses, the reality is starker.

“Saying five years gives people a sense that they still have time, but I think it’s a two-year horizon.”

The other side of that horizon is a binary situation: companies that adopt and use AI can survive and even thrive, and companies that don’t are in distress.

That’s the impetus behind OSL’s drive to go full tilt into AI adoption.

AI productivity

Madden says it begins with getting everyone in the company to use AI tools. “It’s not just about the trading bot, or relying on a few smart people to make decisions. It’s about leveraging the organization’s DNA to make a material change.”

People who use AI to increase their productivity become assets. Those who don’t are a drag on the P&L. Some departments may be almost fully automated (Madden cites marketing as one area that is now mostly run by machines).

Madden says he began with the legal department so they would release policies to encourage staff to use AI appropriately. “Most companies have policies telling their staff not to use AI,” Madden noted, adding that such companies are hobbling their chances of growth.

Adapting GPT

The company also began to adapt GPT models to its own needs. ChatGPT’s public versions are run on data from the internet up to 2021, so it has no knowledge of subsequent events. This safeguard renders it useless to a trader.

But OpenAI has also enabled plug-ins. For a fee, companies can expose their internal data in real time to ChatGPT. Madden acquired early access from Sydney to incorporate data related to digital assets, wallets, and trading. Now OSL runs its own ChatGPT using graphic processor units provided by software giant Nvidia.

The other aspect to plug-ins is that users can plug in whatever data they like: there is no censorship or political bias that might occur when using services by Microsoft or other Big Tech companies that need to keep an eye on the political temperature in Washington, DC.

The upshot is OSL’s AI Trading Bot is intuitively easy to use: just ask it to send or receive digital tokens or stablecoins, and it processes the trades.

Pleasing traders

One reason OSL decided to make this its primary AI-driven face to the outside world was because, somewhat to its surprise, traders turn out to be difficult to automate.

In the equities world, most trading has long been electronic (or “low touch”). And that’s the case when OSL interacts with big financial institutions, whose desks integrate via APIs. But OSL’s bread and butter is transacting with family offices and hedge funds. And their traders prefer to operate via chat. It’s no more automated than picking up a telephone.

It’s one thing to offer clients a user-friendly trading interface, but OSL’s bot does far more. By correlating clients’ history of trading behavior and portfolio composition with market news and events, as well as historical patterns, the bot can anticipate what clients would want to do in the day-to-day bustle of markets.

It presents them with its options and presents a call to action. It’s designed to make traders want to use a ChatGPT function.

Madden refers to the example of Uber, which was brilliant at removing frictions from the user need to hire a car ride. Ultimately with Uber there was no series of extra clicks: you entered your destination, and that was it. A designated vehicle showed up, took you there, and got paid, without you doing anything other than maybe give the driver a rating.

Regulatory pitfalls

Financial apps have come a long way, but they still require the user to initiate or confirm every step. OSL’s trading bot can essentially operate as a discretionary money manager.

And here’s where things get tricky. OSL is not licensed to manage money or provide financial advice. It’s a broker and an exchange.

“We’re regulated,” Madden said. “We have to make sure we get the right balance between what the user wants without our asking them, and asking them if this is what they want to do.”

He notes that OSL cannot provide discretionary money management or serve up a robo-advisor. It’s not a wealth manager. But technology is now getting to the point where the distinction is blurry. What’s the boundary between financial advice versus a good customer experience?

“We won’t release this until we’re sure of our licensing,” Madden said.

But at some point, OSL will decide it has got the balance right, at which point a small digital-assets player in Hong Kong will be able to provide the same kind of electronic service as global investment banks that spend tens of millions of dollars on similar capabilities.

Personalization

Madden expects most of OSL’s trading (as well as its internal processes) will be handled by machines in a few years. Tech utilization is about scale, but firms that succeed at this will be involved in a very different kind of expansion.

Scale is about making a formula repeatable all the time. Traditionally that means creating an output that serves the masses: go to a McDonald’s anywhere in the world, and the Big Mac tastes the same. In the context of AI, scale becomes about applying that formula to create unique outcomes. Go to an AI McDonald’s to get your Big Mac with any ingredients and packaging you like.

That’s the direction Madden thinks AI is taking financial services and, for that matter, most businesses and sectors. Until now, customization was manual and expensive. Soon it will be the norm because it’s automated.

Madden returns to the point, however, that organizations that only focus on the customer-facing end (like a trading bot) are going to flop if the rest of the enterprise isn’t also built around AI.

Employment is one aspect, both how to train people to use AI tools as well as the fear of mass layoffs. Another aspect is companies that fear releasing their IP or data if they allow their people to embrace AI. Institutions that opt for caution over AI-led transformation are at risk of irrelevance in a short space of time.

Agile dinosaurs?

Large financial institutions are obviously more likely to be imperiled. But that may not always be true. Microsoft has deployed OpenAI licenses to its clients using its cloud services. That gives incumbents an opportunity to adopt AI tools quickly and remain dominant.

Big companies will also have a long-term advantage in their ownership of data, or at least access to large data sets. Once the majority of transactions are effected by machines, beating the competition will require proprietary AIs trained on big sets of data that companies either own or license. That also favors incumbents that are proactive with AI adoption, Madden says.

Finally, Madden runs a digital-assets business. He is optimistic that blockchain finance is the perfect infrastructure for a future in which most transactions are initiated and executed by AIs, with smart contracts smoothing the way.

AI tools already point towards a financial market that is cross-border and 24/7, something no regulator can supervise or enforce, so it will make sense to take advantage of the auditability of blockchain finance for regulatory and reporting purposes. Madden believes this will be the near future for finance: AIs transacting over blockchain rails. Jurisdictions will compete both on their sensible regulation of blockchain – and their ability to support companies’ access and governance of data.

Which in turn is likely to put a spotlight on rules around data sovereignty and a need to access the best AI companies, which today are in the US.

But those are questions for tomorrow. For today, says Madden, “It’s just incredibly cool to see how smart the AI has become.”

Time Stamp:

More from DigFin