Ignoring artificial intelligence (AI) is no longer an option. IT leaders have to understand the opportunities and limitations of AI and decide where and how AI makes sense to create business value. To deepen and share ideas on this topic, CIO WaterCooler, in partnership with Globant, joined a Digital Boardroom with a high-level panel of specialists. The event was held in late March and included the participation of Juan Jose Lopez Murphy, Head of Artificial Intelligence at Globant UK.
Artificial intelligence is rapidly becoming mainstream, but poorly implemented AI can lead to a poor customer experience and reputational damage. The organization has to be AI-ready.
The first steps to achieve this are to clean up the data infrastructure and hire the required data science talent. Machine Learning Operations (MLOps) handles, automates, and simplifies the complexity of embedding AI in the organization.
“It seems that we have finally passed the hype wave and are facing an exciting phase of building real, useful solutions. “Exploring” is not enough anymore.”Juan Jose Lopez Murphy, Global Head of AI, Globant
Most organizations use AI in some way
Among the participants in the session, 47% already had AI in their organization, either in production in specific business units or subsidiaries, centrally managed across the business, or in an AI proof-of-concept project. 18% used applications that supposedly have AI under the hood, while 35% had not used AI yet. The Head of IT at a hospitality brand said, “We’re looking at using AI in the booking engine, but before that, we have some catching up to do and get the fundamentals right about our IT systems.”
What we are seeing in AI today is about reaching a competitive advantage, but we are not that far away from the time when AI will be embedded in every aspect of the digital world. AI is going to be a necessity to be competitive and relevant.
What MLOps is and why we need it
Messy data (59%) and lack of data skills (41%) were reasons not to deploy more AI. MLOps extends DevOps with data science and puts processes around the machine learning lifecycle, according to the main speaker. These should be continuous processes, repeatable, and with good governance practices. MLOps should manage versioning, data storage and retrieval, computing resources, access, and process flows. The platform should be extensible because there are always new models, technologies, algorithms, and ways to put things into production.
MLOps is still nascent. We see many offerings with varying degrees of maturity; as that evolves, MLOps will enable more cross-functional teams of specialists – easing up on the current requirements of a cloud + data engineer + ml engineer + data scientist type unicorn we see many organizations trying to find.
AI is part of the data platform evolution
Vendors are busy releasing MLOps platform products. ‘The technology is exploding at the moment,’ according to the main speaker. There is no clear market leader yet. Many will eventually get acquired or go out of business, so make sure you have an exit strategy. The latest MLOps platforms include a ‘feature store’ and build on the concepts of a data warehouse or data lake house, which another speaker used in production. MLOps platforms are the next iteration in the evolution from SQL to batch MapReduce processes to streaming data platforms and near-real-time processing.
As platforms with the evolution in AI turn into the “data-centric AI,” we find a synergy that will enable us to combine large-scale “foundational” models and low-data “expert” models that will permeate previously hard-to-get industries.
AI is a team sport coached by a business.
The CIO of an international investment management company asked whether we should prefer technical or business skills when hiring a data analyst. The speakers generally agreed that business curiosity is more critical to avoid ‘the trap of people falling in love with algorithms.’ What question are you trying to solve? What decision are you trying to make? What is the unknown that you’re trying to make known? In the sessions, the chair had worked with citizen data scientists in the public sector said they could be key players.
We always focus on improving decision-making, specifically when the complexity and frequency of the decisions spiral up. If a simple rule solves most of it, start with that and iterate when improvements warrant it. You are halfway there if you can adequately frame how you would use data to choose the better outcome!
Regulation coming for ‘Explainable AI’
The first speaker addressed the ethics of AI but thought that term was ‘an ivory tower thing,’ preferring words like ‘explainable,’ ‘responsible,’ ‘transparency,’ and ‘trust.’ Regulation is in the works from governments, standards bodies, and the World Economic Forum. The speaker had ‘done some work for the Indian Government around this.’ The legislation would appear in a year or two and look similar to GDPR. Organizations that start implementing ‘the rules of the road for AI’ now voluntarily would have an advantage ahead of legislation.
“Ethics” is more valuable as a grouping factor than something solvable; hence a focus on the actionable components mentioned would enable decision-makers to be accountable and the impact of the algorithms better governed. “Explainable” is surrounded by controversy about viability and usage, but the other factors benefit from the start.
To explore further, visit https://more.globant.com/reinvent-with-data-and-ai or write to us at email@example.com for a free consultation.