Artificial intelligence (AI) has, despite a flurry of media attention in recent times, been around for sometime. But the ability for humans and computer systems to interface in a more conversational way, drawing on large datasets and freeing people from repetitious grunt work, has sparked a wider debate around the use of the technology. ipushpull CEO, Matthew Cheung spoke to Global Trading about the use of AI on trading desks, the products – and solutions – his firm provides, and why we have nothing to worry about from the AI revolution.
Is AI having a measurable, actionable impact on trading, or is it still very much an ‘edge’ case technology?
There are two answers to this. One, AI has been around since World War Two, so in that sense, it’s nothing new. But generative AI has changed people’s perceptions of what you can do with it. With ChatGPT, people have started looking at AI, chat and chat bots in a different way.
Having the ability to interface, to use more natural language, when querying AI means AI can provide a number of different tools on the trading desk. It can take the unstructured chat of typical chatbots and make it into something more usable. This is definitely an area where I have seen a lot of focus. It is an area where traders and brokers on one side, and salespeople on the other, are very keen to see automation because it solves a problem that has been around a long time – double-keying things, manual processes, emails and spreadsheets.
There is however a reluctance, due to the risk element, to start using these tools externally with clients due to hallucinations* within Large Language Models (LLMs). So, we suggest keeping a human in the loop so that while you are removing a lot of the grunt work, you can still have some oversight in place. Ultimately, however, there will probably be full automation in some capacity.
Can you tell us about what solutions you offer, and what challenges they solve?
We are a sharing and workload automation platform. We are used by traders, brokers and exchanges, data companies, and fintechs. We are solving a lot of problems around data sharing where a great deal of this falls into manual, legacy ways of doing things. We do workflows around data distribution, where we can plug into a bank that may have bond axes, or a broker that has prices they are quoting. We can pull that data and distribute it to lots of different applications and services. For example, a trader at a small hedge fund might be happy consuming data in a spreadsheet, whereas a more sophisticated trader, a quant perhaps, might want a FIX API. An OTC trader might want to consume something directly in chat, because that is where they are doing the majority of their trading. We are agnostic, in that sense, to what format the end user wants to consume things in.
We have other workflows where customers use us to grab data from their clients. That could be sending prices out, to a trader, and then that trader wants to initiate an order, or trigger an RFQ, for example.
We can pull all of the unstructured conversation and text from different platforms and turn it into a structured object, and then plug that into a service to get a response. Then we have different workflows where we plug different components together so that we can provide standalone tools for different people in the market.
If I’m typing stuff into lots of different chats or I’m emailing different people, that information is not stored anywhere, but now you can start capturing it and then start creating analytics based on it. For example, if I’m a trader at a buy-side firm, instead of my broker coming to me with quarterly broker reports telling me how well that broker has performed, I now have my own statistics and analysis. In short, our service makes processes a lot more efficient and allows firms to build data around what they do as well.
Do you foresee a situation where firms allow AI to execute trades entirely without human oversight or intervention?
That is already happening in some areas, depending on the size of the trade. One of our clients uses a bot that, under a certain amount, can trade. But over that amount, the trade will be routed to a particular salesperson to pick up.
I think what we are headed towards is something like a co-pilot, where you can start to digitise some of the workflows, which will give you the ability to automate a lot. Looking ahead, I see the trader and the AI working together where all these trades, services and systems are largely automated with a trader acting in an oversight capacity doing exceptions handling. The human would also get involved if there was a sudden rate cut from a major central bank, or if a bomb went off in a major city.
AI is a tool, like fire, electricity, computers – it will never replace humans. The more you can embrace it and understand it, the better. Therefore, it is essential to have a strategy about how to use AI, both in terms of your own products, and on the trading desk.
We have seen the evolution from floor trading to screen trading to more quantitative algo trading. But there is still a need for humans, with their experience and broad range of skills, and their ability to connect the dots. In terms of headcount, AI doesn’t necessarily mean job cuts. If AI frees people from more rote, grunt work, they can be better deployed, servicing clients. It is very much freeing people’s time up, rather than losing jobs. We will see people moving on to more higher value tasks, rather than sacking them because AI has taken their job. I don’t see that happening because there’s higher value tasks people can be doing, rather than typing stuff into chat and opening up emails and opening up spreadsheets. That’s just a rubbish use of anyone’s time. n
*AI models are trained on data, and they learn to make predictions by finding patterns in the data. However, if the training data is incomplete or biased, the AI model may learn incorrect patterns. This can lead to the AI model making incorrect predictions, or hallucinating.
©Markets Media Europe 2024
