“We need to start spreading awareness of what AI can do for businesses, not only to executive decision makers but also to young people, younger than college. We need to start developing and embedding the quantitative and analytics skill sets that will be needed for jobs that will be available over the next decade or so,” Danzig said. “We could really inspire a larger generation of skilled workers.”
One of the council’s first projects is developing an infographic identifying the barriers and accelerators to AI Adoption, similar to what other councils have created. That should be published in the first quarter and it’s only the beginning, Tatiraju said.
“I’d like to see us detail and standardize the AI lifecycle, specifically the machine-learning lifecycle, starting from data gathering and cleansing all the way to production––which entails serving models, monitoring the models and a continuous feedback loop,” he said. “We should be able to identify various roles within a lifecycle and then perhaps work on corresponding certifications for those roles.”
Standardization of machine-learning lifecycles would also be critical in getting buy-in from big technology companies looking to implement AI, Tatiraju said.
“Overall, we’re looking to accelerate AI adoption and demonstrate value, and also maybe tackle issues such as removing barriers to adoption, including ethical considerations, and addressing the lack of guardrails,” he said. “For example, there was a recent AI competition where someone found a way to cheat to get to the top of the leaderboard. Machine learning models may allow us to analyze things like that and be more deliberate or proactive in our approach to put in some guard rails and have meaningful discussions that discourage that behavior.”
Explaining AI and Machine Learning
AI is unique compared to other emerging technologies such as internet of things, drones or robotics because progress and innovation isn’t as visible, happening more “behind the scenes,” said Danzig. Two examples are chatbots and the automation of customer service and guiding a customer through a customer journey, both of which leverage AI to provide great value to businesses and customers.
The council can also help vendors, solution providers and customers recognize the absurdist view that AI can do more damage than create value, the so-called “killer robot theory” that theorizes AI will take over our lives.
“Look, we haven’t achieved artificial general intelligence (AGI), not even close. That’s not a big concern. That said, there are definitely some issues to address in terms of privacy, AI ethics and the social impact,” Tatiraju said. “For example, new generation machine learning models are being used to generate deep fake videos and images now. How do we combat this and identify what’s real and fake? Generative adversarial networks help with that to an extent. There are issues around reproducibility and explainability as well. For instance, in healthcare where questions of accountability and transparency are particularly important, it is paramount that the machine learning models’ decisions are explainable and reproducible. Chasing the next benchmark, GLUE score, or SOTA (state of the art) vs. focusing on AGI, and some of these issues needs to be balanced too.”
There are other misconceptions too. For example, in the finance industry many people assume that AI solutions pile through numbers to identify profitable trading patterns. And, while that is a valid application and one that is of interest, it is not even close to topping the list of ways in which AI adds value to financial services companies, Danzig said.
“AI is used far more frequently for operations, for the actual execution of trades. Suppose you needed to offload a $100 million block of shares in a single company. You certainly would not simply offer up the entire block all at once, as this would result in the market moving away from you and cannibalizing your own profits,” he said. “Instead, AI is leveraged not only to send out small test orders, but also to incorporate information that arises from the market’s reaction to future orders – how quickly it was executed, at what price, and so on. The ability to automate, streamline, and remove human emotion from this process is one of the most compelling ways in which fund managers use AI to deliver excess returns to their clients.”