Sameer Gupta knows about electronic markets. The former COO for J.P. Morgan’s global electronic equities trading and Americas high touch and program trading business has been steeped in trade mechanization since graduating from Carnegie Mellon University in 2003. He’s worked as a programmer at Goldman, a business development executive at NYSE Euronext and an electronic trading implementer at J.P. Morgan. Now, he’s C.O.O. of iSentium, a company that uses intelligent algorithms to turn social media sentiment into tradeable data. And he says banks are getting their approach to artificial intelligence (AI) all wrong.
“Banks have got the wrong kind of approach to their artificial intelligence initiatives and are hiring the wrong kinds of people,” says Gupta. “It’s very, very difficult to get anything new and original like this off the ground on Wall Street. There’s a tendency for banks to hire people who stay for a year or so until the banks realize nothing has actually been produced. It’s much easier to produce original products in AI away from large firms. We’re razor sharp and committed to this alone.”
Most banks have already got teams working on artificial intelligence initiatives. At Goldman Sachs, machine learning specialists sit within the strats group. At J.P. Morgan they’re to be found in the automated trading strategies team. But Gupta says banks’ machine learning experts are both bogged down in bureaucracy and prone to over-complicating things.
iSentium has stopped hiring from banks for this reason: “We learned the hard way that banks have the wrong kinds of people on their AI teams. We found ourselves with people who were too complex, who couldn’t take a simple view and who didn’t have a visceral understanding of the data that they could apply to the real world. What we’re doing isn’t easy. You can’t just delegate it.”
iSentium has offices in Miami, New York and Montreal and it’s hiring. Its president and CEO Gautham Sastri says they’re recruiting, “bright, unpolluted talent,” and that, “anyone who calls themselves a data scientist will not be hired.”
The natural language processing (NLP) element of iSentium’s system is coded in Java, while the front end seen by iSentium’s hedge fund and banking clients is coded in Python. The company’s scientific brainpower is located in Montreal. “Montreal is the new hub for artificial intelligence and all our linguists are from there,” says Sastri. The company’s chief technology officer, Anna Maria Di Sciullo, is an MIT PhD and professor in linguistics and cognitive computation at the University of Montreal and iSentium tends to hire AI experts who achieved their PhDs under her guidance. “It’s extraordinarily difficult to find people who can build NLP systems,” says Sastri. “We needed a system that’s both fast and accurate.”
iSentium is already able to process around 20m social media messages a day and to derive sentiment indicators from them. Ultimately, it aspires to process 200m. You can see an example of the product in action at isentium.live as it analyses the global sentiment behind tweets mentioning Donald Trump.
If banks want to develop their own artificial intelligence products, Gupta suggests they should do so by working with agile third parties rather than using clunky teams in-house: “AI initiatives are still in the warm-up stage at most places while consuming significant time and resources.” Last year, iSentium partnered with J.P.Morgan on a co-branded index (BB: JPUSISEN) which Gupta says returned 16% net of all costs and fees.
Banks are still taking small steps at the moment, says Gupta. Ultimately, however, he predicts they will need to commit more strongly to AI as clients demand it. In the process, he says banks will need to rethink their systems and their hiring: more staff will need to be “multi-lingual across the languages of technology and data and capital markets.” In this context it’s unsurprising that Goldman has reemphasized its demand for STEM staff.