Posted on

The environmental cost of Generative AI: A conundrum for FIs

esg
Horizontal bar chart showing that ChatGPT, Gopher and OPT- all LLM models emit alot more CO2 than those emitted by an american in an average year. Chat-GPT emitted 502 tonnes of CO2 while an 18.8 tonnes is emitted within the span of an average human life.
Source: The AI Index 2023 Annual Report,  Institute for Human-Centered AI, Stanford University, Stanford

Generative AI is making a lasting impression in nearly every industry – consumers, CEOs and whistleblowers alike are busy postulating the possible advantages or dangers of the breakthrough technology found in the likes of ChatGPT by OpenAI and Bard by Google.

While new use cases seem to emerge everyday, Generative AI which are powered by sophisticated large-scale, machine learning models have a definitive cost on the environment. New data by the Stanford Institute for Human-Centered Artificial Intelligence suggests that models like ChatGPT-3 and Bard utilize an amount of energy that could power an American’s home for hundreds of years.

The research builds on recent data measuring the carbon footprint of training 4 models: BigScience’s BLOOM, Meta’s OPT, OpenAI’s GPT-3, and DeepMind’s GOPHER. OpenAI’s ChatGPT has released around 502 metric tons of carbon during its training, according to the research. 

Those in the financial industry are not new to these sorts of concerns. The advancements in crypto until very recently motivated the same questions about environmental impact and carbon footprints. Infact, outcry around the impact of crypto technologies on the environment motivated a major crypto player Ethereum to switch its model to a proof of stake mechanism, last year – a switch that was supposed to reduce environmental impacts by 99%.

As the financial industry considers the usefulness of this technology, it may also have to consider how the use of Large Language Models like ChatGPT will impact their ESG-related values and objectives. If emissions by Large Language Models are not tackled, and companies behind these models do not proactively release information on their carbon footprints, we may soon see climate-conscious customers asking some hard questions from their FIs. 

Posted on

AI becoming just another tool in the trader toolbox

In the collective mind, the trading floor is often depicted as a chaotic and noisy place where traders shout out orders at the top of their lungs. The reality is much much quieter. Where humans used to shout, bits and bytes now move silently.

The process started when trading shifted from manual, voice-based actions to computers. Then it developed further when automated order routing was added, saving time and resources. Then, automated trading systems rose in popularity, responsible for about 75 percent of today’s market volume.

In recent years, with the advent of cheap computational power, the next step of automation is trading automation driven by AI or machine learning. It is here that popular media propagates fears about the ultimate takeover of machines.

“It feels like some of the buzz and the hype has died down, which is an interesting stage in the lifecycle,” said Tom Doris, CEO of OTAS Technologies, a market analytics and trader intelligence company. Now, he adds, we can get back to looking at the problems that technologies like these are supposed to solve.

A typical trader, explains Doris, has too many orders in his queue. Faced with this, he will start processing them from the top and work his way down to the bottom. He might linger a little bit more on those that are more volatile or require more attention, but it might be hard to spot those among the noise.

“The market is generally pretty boring,” Doris said. “If you have 100 orders, 95 of them are perfectly ordinary. It is all very predictable. Your task as a trader is to find those 5 percent where something unusual is happening.” OTAS’ technology is able to create a predictive model of how a stock should behave, and alerts the trader when unexpected information changes a stock’s behavior. The technology can be plugged directly into an Execution Management System, so that the trader can act on those alerts instantly.

The trend over the last couple of years was towards increased automation, attempting to take humans out of the loop. Now, it’s understood that there is a limit to what a trader can do systematically.

“You want humans looking at situations where there is a human story going on,” Doris explained.  For example, when a stock starts to rally because of aggressive buying, only a human with a good understanding of the risk landscape and the company’s story can discern if this is informed flow or a trend that may revert.

There have been several hedge funds priding themselves in the use of AI software to guide their decision making, including Bridgewater Associates, Renaissance Technologies, D.E. Shaw, and Two Sigma. Many more firms describe themselves as “systematic”, meaning they base their decision making on computer models, which might not be driven by AI.

Perhaps the most common approach to AI in investing is the use of natural language processing to be able to make sense of unstructured market data and the use of neural networks to identify patterns, relationships and hidden trends.

AI is seen now more as another tool in the toolbox of traders, rather than a magic bullet, Doris concluded.

Posted on

OTAS Technologies’ Tom Doris is creating machines to do (part of) a trader’s job

interview with fintech investor, Dan Ciporin

Tom Doris is CEO of OTAS Technologies

What’s OTAS all about?

Tom Doris, OTAS Technologies
Tom Doris, OTAS Technologies

At OTAS we use big data analytics, machine learning and artificial intelligence to extract meaning from market data and provide traders and portfolio managers with insights that would otherwise lay hidden. Our decision support tools help traders to focus on what’s important and interesting, you could say that we use machines to identify the areas that humans should be paying attention to.

I did my Ph.D. in artificial intelligence, and around 2009, it was clear to me that several of the more sophisticated hedge funds were converging on a set of approaches to market data analysis that could be unified and made more efficient and general by applying algorithms from machine learning and artificial intelligence.

Better yet, it quickly became clear that the resulting analysis could be delivered to human traders and portfolio managers using natural language and infographics that made it easy to absorb and action. At the same time, the role of the trader was becoming increasingly important to the investment process, while the problem of executing orders was becoming more difficult due to venue fragmentation, dark pools, and HFT, so it was clear to me that there would be demand for a system that helped the trader overcome these problems.

How does leveraging artificial intelligence for trading help traders and portfolio managers make better decisions and manage risk?

Experienced traders and PMs really do have skill and insight. With all human skills, it is not easy to apply the skill systematically. We can leverage AI to help humans scale their investment process to a larger universe of securities, and also to ensure they apply their best practices on every single trade.

In many professions, everyday tasks are too complex for a human to execute reliably, for instance, pilots and surgeons both rely on extensive checklists. Checklists aren’t sufficient in financial markets because hundreds of factors can potentially influence a trader’s decision, so the problem is to first find the factors that are unusual and interesting to the current situation. This is the task that AI is exceedingly good at, and it’s what OTAS does. Once we’ve identified the important factors for a given situation, machine learning and statistics help to quantify their potential impact to the human, and we use AI to generate a natural language description in plain English.

What is compelling the increased use of artificial intelligence and big data analysis in financial services?

A basic driver is that the volume of data that the markets generate is simply too much for a human to analyze, but the more compelling reason is that AI and machine learning are effective and get the results that people want. Intelligent use of these techniques gives you a real edge in the market, and that goes to the firm’s bottom line.

How do you see artificial intelligence and big data analysis playing a role in trade execution in the future? Any predictions for 2016?

AI is going to provide increased automation on the trading desk. Execution algorithms have already automated the task of executing an order once the strategy has been selected by a trader. Now we’re seeing a big push to automate the strategy selection and routing decision process. The next milestone will be to see these systems in wide deployment, and with it will come a shift in the trader’s role; traders will have more time to focus on the exceptional orders that really benefit from human input. Also, the trader will be able to drive the order book in aggregate according to changes in risk and volatility. Instead of manually modifying each order, you will simply tell the system to be more aggressive, or risk averse, and it will automatically adapt the strategies of the individual orders.

What’s the biggest challenge in acquiring new customers in your space?

Traders have largely been neglected in recent years as regards technology that helps them to make better decisions. Even when the benefits of a new tool are clearly established, it can be difficult for the trading desks to get it through their firm’s budget. Despite the recent hype around HFT and scrutiny of trading, there’s still a lag when it comes to empowering traders with the best information and tools to support them.

Photo credit: k0a1a.net via VisualHunt.com / CC BY-SA